Professional Documents
Culture Documents
For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance
Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy is not
required from the publisher.
Printed in Singapore
“…and have always in view not only the present but also the coming
generations1…”
Leo and Aaron
Maggie and Kate
Aleksandra, Michael and Teodora
1
Seven generation sustainability (Wikipedia, 2015).
May 2, 2013 14:6 BC: 8831 - Probability and Statistical Theory PST˙ws
vii
viii Geometry, Language and Strategy—Vol. 2
Louis H. Kauffman
December 16, 2015
Preface
ix
x Geometry, Language and Strategy—Vol. 2
their experiences with me. This has occurred over many years and I am
not able to provide a detailed accounting. But I am able to point to a few
people who have been especially helpful for this phase of the project.
I am grateful to Stephen Wolfram and members of his staff at
Wolfram Research, Inc. (WRI) for ongoing support and especially from
Maryka Baraka and others from the Partnership Program Support at
Wolfram Research. Lou Kauffman strongly supported this project and
provided in this new phase, access and feedback from his students.
Gordon Kane provided feedback on the physics and pedagogy based on a
draft of the manuscript. Jack Behrend read and provided edits of
countless early drafts for which I am grateful. I received useful and
supportive feedback on my philosophical stance from Steve Rosen. Lisa
Maroski and Helen Kessler have held me accountable for my project and
goals as well as providing feedback on specific sections. They have
honed my understanding of Systems Dynamics as a deterministic
approach applied to a variety of real world applications. Keelan Kane
collaborated with me on some classic and paradoxical problems in the
literature on the prisoner’s dilemma, which led to chapter 7. Col. Edward
Sobiesk introduced me to the new field of Network Science, and the
interest at West Point in applications. I have benefited from the support
of the Milwaukee School of Engineering (MSOE) during the course of
this project. My Colleagues at MSOE have patiently listened to my
attempts to clarify my ideas. Most recently, I am particularly grateful to
Bharathwaj Muthuswamy and Jovan Jevtic and our weekly meetings to
discuss a variety of research topics including this one. Finally I am
grateful to my daughter Liisa M. Thomas for her legal advice and to my
wife Kathleen for her patience throughout the course of this project and
her commitment to seven generation sustainability.
G. H. Thomas
Contents
xi
xii Geometry, Language and Strategy—Vol. 2
Table 5.1. Player Fixed Frame Model—Persistent and Dynamic Components .... 170
Table 5.2. Time Equations .................................................................................... 175
Table 5.3. Gradient Equations on Surface Normal to Flow .................................. 178
Table 5.4. Player Fixed Frame Model—Electromagnetic Fields .......................... 195
Table 5.5. Player Fixed Frame Model—Tidal Fields ........................................... 197
Table 5.6. Player Fixed Frame Model—Inertial Components .............................. 201
Table 9.1. Player Stakes and Decision Choices for Sensitivity Analysis ...... 348
Table 16.1. Mesh Analysis in WWW Model .......................................................... 639
xxi
May 2, 2013 14:6 BC: 8831 - Probability and Statistical Theory PST˙ws
Figure 8.7. Nash solution for : non-zero game value ............................. 307
1
Figure 8.13. Player interest field in the co-moving frame (PDM). ........ 316
Figure 8.14. Electric field f j in the co-moving frame (PDM). ................... 316
Figure 8.15. Player engagement e in the co-moving frame (PDM) ........ 316
Figure 8.16. Diagonal components ln jj in the normal form frame(PDM) .... 316
Figure 8.17. ln jj in the normal form frame (PDM) .................................. 319
Figure 8.18. Flow components of the Killing vector E j (PDM) ..................... 319
Figure 8.19. Prisoner 1 inactive Killing vector components E (PDM) ........ 319
1
Figure 8.20. Prisoner 2 inactive Killing vector components E (PDM) ........ 319
2
xxiii
xxiv Geometry, Language and Strategy—Vol. 2
Figure 16.33. f0AD model pressure contours along the streamlines ...................... 632
Figure 16.34. f0AD model acceleration potential contours along the streamlines.. 633
Figure 16.35. f0AD model streamline contours for free fall................................... 634
Figure 16.36. f0AD model defense (blue) charge ................................................... 634
Figure 16.37. f0AD model attack (red) charge ....................................................... 635
Figure 16.38. f0AD model code of conduct charge ................................................ 636
Figure 16.39. Gauge condition surface with mesh 4 .......................................... 638
Figure 16.40. Gauge condition surface with mesh 10 ........................................ 638
Figure 16.41. Pressure in the f1WWW model ........................................................ 639
Figure 16.42. Acceleration potential in the f1WWW model .................................. 640
Figure 16.43. Streamline “tube” in f1WWW model ............................................... 641
xxviii Geometry, Language and Strategy—Vol. 2
Figure 17.25. Decision pressure for the f2AD model ............................................. 680
Figure 17.26. Decision process acceleration potential ........................................... 680
Figure 17.27. Player bonding scalar field .............................................................. 683
Figure 17.28. Player shear vector field 11 ....................................................... 684
Figure 17.29. Player shear vector field 22 ...................................................... 684
Figure 17.30. Player shear vector field 12 ...................................................... 685
Figure 17.31. Player shear vector field 13 ...................................................... 685
Figure 17.32. Player shear vector field 23 ...................................................... 686
Figure 17.33. Composite payoff ............................................................................ 687
Figure 17.34. Seasonal payoff vector field ............................................................ 689
Figure 17.35. Allowed region for βn, in the f2AD model ....................................... 694
Figure 17.36. βn for the f2AD model...................................................................... 697
Figure 17.37. The dual-*βn for the f2AD model ..................................................... 697
Figure 17.38. Coordinate streamlines for the f2AD model ..................................... 698
Figure 17.39. Code of conduct passion density plot for the f2AD model ............. 699
Figure 17.40. “Blue” passion vector field f2AD model .......................................... 700
Figure 17.41. “Red” passion vector field f2AD model ........................................... 700
Figure 17.42. Code of conduct passion vector field f2AD model........................... 701
Figure 17.43. “Blue” player interest vector field f2AD model ............................... 702
Figure 17.44. “Red” player interest vector field f2AD model ................................ 702
Figure 17.45. Code of conduct player interest vector field f2AD model .............. 703
Figure 17.46. Density plot of the source I B for composite
payoff f2AD model .......................................................................... 704
Figure 18.1. Time sequence for a typical stock Jan-Dec 2015............................. 709
Figure 18.2. Recurrence plot for typical stock time sequence ............................. 710
Figure 18.3. f2AD model flow vectors ................................................................. 716
Figure 18.4. Coordinate plot u a for f2AD model ................................................. 717
Figure 18.5. Initial flow for f1-hf AD model ........................................................ 719
Figure 18.6. Travelling wave phase gradient along z for f1-hf AD model .......... 720
Figure 18.7. Persistence of Memory by Salvador Dali [2015]............................. 726
Figure 18.8. Player charge for “blue” when “blue” is aggressor for f0AD model 731
Figure 18.9. Phase space plot of player engagements for “blue” as
aggressor in f0AD model.................................................................. 732
Figure 18.10. Phase space plot of player engagements for “red” as
aggressor in f0AD model.................................................................. 733
Figure 18.11. Active acceleration Q x , y , 0 for f0AD model .............................. 736
xxx Geometry, Language and Strategy—Vol. 2
Figure 18.12. Active acceleration Q x , y , 0.1 for the f0AD model .................... 737
Figure 18.13. f0AD model of the time metric factor .............................................. 741
Figure 18.14. Initial co-moving frame acceleration q x , y ,0 for f0AD model .... 742
Figure 18.15. Player engagements for the f2AʹD model ......................................... 747
Figure 18.16. Coordinate phase space for the f2AʹD model ................................... 748
Figure 18.17. Frequency dependence of the f2AʹD model transfer function ........ 748
Figure 18.18. Phase behaviors for the f2AʹD model ............................................... 749
Figure 18.19. Recurrence plot for the f2AʹD model relative strategy component .. 749
Figure 18.20. f2AʹD model initial strategy flows................................................... 750
Figure 18.21. Time dependence of the f2AʹD model strategy flows ...................... 751
Figure 18.22. Player aggression flows for the f2AʹD model .................................. 751
Figure 18.23. Gradient aggression strategy flow for the f2AʹD model .................. 752
Figure 18.24. Network connections recurrence plot for aggression
preference strategy for the f2AʹD model .......................................... 752
Figure 18.25. Time metric factor for the f2AʹD model ........................................... 753
Figure 18.26. Initial acceleration gradient for the f2AʹD model ............................. 754
Figure 18.27. Acceleration potential for the f2AʹD model ..................................... 754
Figure 18.28. Initial active acceleration vector for the f2AʹD model...................... 755
Figure 18.29. Active acceleration slices for the f2AʹD model ................................ 756
Figure 18.30. Pressure for the f2AʹD model ........................................................... 756
Figure 18.31. Recurrence plot for stock price ratio (dB) between two similar
companies ........................................................................................ 757
Figure 18.32. Mean price of two similar companies .............................................. 758
Figure 18.33. Price ratio of two similar companies ............................................... 758
Figure 19.1. Code of Conduct charge behavior for the f2WWW model ............ 766
Figure 19.2. “Work” entitlement payoff in the f2WWW model .......................... 767
Figure 19.3. “Wealth” entitlement payoff in the f2WWW model ........................ 768
Figure 19.4. Code of conduct entitlement payoff in the f2WWW model ............. 771
Figure 19.5. Composite payoff in the f2WWW model ......................................... 772
Figure 19.6. Phase space plot of player engagements in the f2AʹD
model on initial boundary ................................................................ 773
Figure 19.7. Phase space plot of player engagements in the f2AʹD model on
z 0.02 boundary ........................................................................... 774
Figure 19.8. Code of conduct charge for f2AʹD model......................................... 784
Figure 19.9. Code of conduct charge for f2WWW model with extended
time interval ..................................................................................... 785
List of Figures xxxi
Figure 19.10. Entitled payoff for “blue” in the f2A´D model ................................. 785
Figure 19.11. Entitled payoff for “red” in the f2A´D model................................... 786
Figure 19.12. Entitled payoff for code of conduct in the f2A´D model ................ 786
Figure 19.13. Composite payoff in the f2A´D model ............................................. 787
Figure 19.14. Charge contours z 0.2 for the f2WWW model .............................. 787
Figure 19.15. Charge contours z 0.2 for the f2WWW model ........................... 788
Figure 19.16. Engagement payoff for “work” in the f2WWW model .................... 788
Figure 19.17. Engagement payoff for “wealth” in the f2WWW model ................ 789
Figure 19.18. Engagement payoff for code of conduct in the f2WWW model ...... 789
Figure 19.19. Player shear 11 in f2WWW model ................................................. 790
Figure 19.20. Player shear 22 in f2WWW model ................................................ 791
Figure 19.21. Player shear 12 in f2WWW model ................................................ 791
Figure 19.22. Player shear 13 in f2WWW model ................................................. 792
Figure 19.23. Player shear 23 in f2WWW model ................................................ 792
Chapter 1
Introduction
3
4 Geometry, Language and Strategy—Vol. 2
200 or 2000 such blocks. Towers quickly thrown together may last a few
weeks but not a few centuries. We have developed engineering practices,
i.e. rules of behaviors, which provide guidelines for building large or
complex structures that are built to last. These guidelines help us build
structures that would have been impossible to imagine before
engineering. I grew up in Los Angeles that limited buildings to 12 stories
because of fear of earthquakes but today, the city allows significantly
taller skyscrapers that have a better chance of surviving than structures
built a hundred years ago.
I believe there is a need for an approach that treats decision making as
an engineering discipline to address those problems that involve large
distances or time frames. We will have to define distance more precisely,
but what I have in mind is a strategic distance as opposed to a purely
physical distance. I will argue that seemingly unrelated past choices may
in fact influence the same decision.
This is Vol. 2 of a project to treat decisions as deterministic
processes. Though others have started this conversation, in Vol. 1
[Thomas G. H., 2006], my contribution was the proposal that the
dynamic properties of decision processes are best described by the
mathematics of differential geometry. Illustrative examples using
streamlines showed that decision processes would exhibit not just
persistent behaviors but oscillatory behaviors analogous to weather
patterns with hurricanes, tornadoes and other vortical behaviors.
After completion of that project, it became obvious that there in fact
might be additional solutions of interest. Volume 1 extended the
essentially DC behavior of game theory to an AC version of discrete
steady state solutions. To understand issues with larger time frames and
strategic distances, there are further extensions. I view the discrete
solutions obtained in Vol. 1 as analogous to resonant phenomena in
physical systems. Thus one might expect that in addition, there would
also be transient and steady state AC solutions of any frequency that are
driven by the initial and boundary conditions of the system. It is the
intent of this volume to focus on such solutions and show that they do
exist. These solutions no longer correspond to purely local behaviors, but
may exhibit global characteristics that no longer mimic the purely local.
Introduction 5
1.1 Pedagogy
1
We indicate by bold-italics distinctions that we believe are new ways to speak about
decision processes.
Introduction 7
also assume that the reader is not familiar with the vast literature on
game theory, which is a rather specialized aspect of economic theory.
This presents challenges that are not unfamiliar to those in
engineering and science. On the one hand we must provide motivation to
the reader to make the effort to learn the new techniques and on the other
hand we must provide adequate tools and training for the reader to be
able to apply and use those techniques on interesting problems. The
payoff in the end is the possibility of being able to vastly improve the
readers understanding of real-world decision processes.
Thus the goal for this volume has been to select and bring relevant
components of the vast literature on differential geometry, including the
physics of gauge theories, to the reader interested in decision processes.
The approach is similar to bringing the ideas of physics to engineering
students who want to apply the ideas to realistic problems. We require
only that the student have sufficient sophistication in mathematics and
science. One way to gain that sophistication is to follow the course of
study that has proved successful in engineering schools.
For this reason we have suggested the requirements for a sophomore
level engineering student. The goal in this volume is to make the ideas
accessible to a student who has followed such a course or its equivalent.
We do this at three levels. We provide an exposition of the ideas, we
provide a set of problems to enhance the student’s understanding of the
material and we provide numerical examples with the help of
Mathematica. The latter provides an important visual tool for making the
ideas come alive.
We show how numerical solutions to the decision process equations
can be obtained for numerous illustrative examples. The equations are
Einstein-like partial differential equations and so the techniques and
solutions obtained also may be of interest to those in other fields such as
relativity, gauge theory and differential geometry. As the purpose of this
book is to provide the necessary mathematical and engineering training
to both understand and apply the approach, out of necessity it is cross
disciplinary and hopefully sheds light on the areas from which it
borrows.
We turn now to an introduction of our goals for decision process
theory.
Introduction 9
1.3 Introduction
choices. Such problems are completely solved in game theory and are
also completely solved for the dynamic theory, which we demonstrate in
later chapters. In this chapter, we provide the context for economic
situations with any number of players and any number of strategic
choices.
In order to find solutions to such problems, in the next sections we
look at classical physics and how the approach might relate to decision
process theory. The equations of classical physics are derived from
differential geometry and we anticipate from Vol. 1 that the decision
process theory equations are as well. We start by looking at the physics
equations in the same way as electrical engineers. We provide a brief
survey of the relevant ideas starting from Newtonian mechanics (Sec.
1.5) to the theory of Maxwell (Sec. 1.8). The latter theory is particularly
relevant to decision process theory.
We find in later chapters that decision process theory equations are
similar to equations in general relativity, so our approach of harmonics
taken from electrical engineering (phasors) may be unorthodox even to
physicists and we think a novel approach to solving equations from
general relativity. Thus an introduction to physics may be helpful to the
more general reader.
We present decision process theory following the techniques of
Lagrange (Sec. 1.6). For the expert, this indicates what we intend; for the
non-expert we will say in sufficient detail what we mean so that you can
understand and use the concepts. We argue that it is useful to break
forces into those that are applied and those that are constraints, allowing
the number of variables to be drastically reduced. These techniques relate
to the key physical principle, the principle of least action, which we use
to create a causal theory. Without losing any generality, we can describe
processes using variables that are applied and matched to the problem.
For decisions we suggest that the key variables are the strategies
available to each participant in the decision.
Given the principle of least action, we bring together the key
concepts of decision making (such as strategies and utility, Secs. 1.7-1.8)
and show that they can be described by calculus-based rules that allow
the extrapolation from the local to the global, both in time and space. The
14 Geometry, Language and Strategy—Vol. 2
To make the conceptual basis for cause and effect relevant to decisions,
we frame decisions as events that occur in some space at a moment in
time. Here we review attempts to do this and find that foundational
aspects of game theory provide the perfect vehicle in which to express
our decision process theory.
We start with a foundational view of decision process theory and its
relationship to engineering and to mathematical and physical theories.
The foundational concept from game theory that we find appealing is
that every decision process can be represented in a standard way and put
into a normal form, an insight of Von Neumann & Morgenstern [1944].
They prove this under very general considerations, creating a theory of
games and economic behavior, where games in their context go beyond
recreational games to include real world economic behaviors. They
envisioned that this way of looking at decision processes provides the
basis for a quantitative way of dealing with social phenomena. See for
example, Myerson [1991], for an undergraduate course on game theory
Introduction 15
F mr . (1.1)
Indeed, one might think of inertia as a part of the system that involves
many degrees of freedom about which we know little.
We rarely deal with a single (“rigid”) body in physics. The simplest
interesting case is two bodies attracted to each other by a gravitational
force. Newton observed that the force is proportional to the masses of
each body and inversely proportional to the square of their distances. In
this way he could show that the theory would agree with the observations
of Kepler that equal areas would be swept out in equal times.
A success of Newtonian mechanics is the accuracy of the predictions
that result from these two statements. A success of the scientific method
on which the theory rests is that as challenges to the theory have
occurred, the framework for discussion has remained or evolved in a
systematic way. For example it was discovered by Einstein that the
above examples contain a fatal flaw for Newtonian mechanics, since they
implicitly assume that the influence of one body is felt instantaneously
by the other body. More careful observations show that action at a
distance is not instantaneous, but is an effect that travels with a finite
speed. A related flaw is the notion of a rigid body, which again would
require the influence to travel instantaneously.
Despite the successes for the motion of two bodies under
gravitational influence, such as the motion of the earth around the Sun
and the belief that the laws hold generally, the Newtonian formulation as
it stands does not deal directly with many real world classical situations.
A car moving along the highway might be idealized as a single object,
but the road on which the car moves is an important contributor to the
forces acting on the car. Using Newton’s formulation, you must account
for all the forces, including those keeping the car from sinking into the
pavement. Calculations that include these effects are long and once
complete, have information that is not normally of great interest, such as
knowing the value of the pavement force.
It is at this point that we see a connection to decision theory. Here too
we have many forces to deal with, but have a suspicion that their
inclusion will yield information that may not be of great interest. One
might be tempted to argue that the forces are not knowable and use
probability arguments. However, we shall stick to the deterministic
approach using the ideas of Newton.
Introduction 19
Fk m k rk . (1.2)
This represents many (second order differential) equations and takes into
account all the forces.
The forces of interest are applied and external (the motive force
moving the car forward) and the forces that are not of interest are
constraints or internal (the gravitational forces holding the car to the
road, the atomic forces preventing the car from sinking into the road as
well as those holding the bits of the car together, and possibly symmetry
forces such as the car is on a planet that is spinning). These definitions
are not absolute, but relative to the interest of the person solving the
problem.
A rigid body is held together by internal forces and such forces
normally do no work (see below). A formulation that even partially
excludes such internal constraining forces is advantageous. To appreciate
this approach, it is helpful to introduce the concept of work, which is
defined as an integral made up of the sum of the product of the force
times incremental distances along a specific path. Work has the
dimensions of energy. Newtonian mechanics distinguishes two types of
energy.
First there is energy associated with motion as kinetic 1 2 m r r
defined as one-half the product of the mass and the sum of the squares of
the velocity components.
Second there is energy associated with forces, as potential energy:
typically the variation of the potential energy with distance along a given
axis determines the force along that axis. In both cases energy has
dimensions of a force times a distance.
20 Geometry, Language and Strategy—Vol. 2
Fk p k . (1.3)
F k
k p k rk 0 . (1.4)
F
k
a
k
p k rk 0 , (1.5)
where F a k are the components of the applied force. The positions of the
objects are not independent but constrained.
The Lagrangian point of view is that any set of independent
coordinates q (not necessarily equal to the above rk coordinates) can
be used to rewrite the above equation. Lagrange showed that the
equations can be written in terms of the kinetic energy T now expressed
in terms of the independent forces and the applied forces Q as they
relate to these independent coordinates:
d T T
Q . (1.6)
dt q q
2
The constraint force does no work because the force moves through no distance.
22 Geometry, Language and Strategy—Vol. 2
T 1 2 mk rk 2
rk . (1.7)
Q F a k
q
V d V V d V
F
a
k Q . (1.8)
rk dt rk q dt q
d L L
0. (1.9)
dt q q
We see that all of the information about the problem is now reflected in
the Lagrangian that has dimensions of energy.
Further these equations of motion solve the problem posed, namely
the equations involve only the applied coordinates, though all aspects of
the kinetic and potential energy are taken into account. In Sec. 3.1, we
Introduction 23
for mixed strategy choices are determined from these pure strategy
payoffs. We must find a way to incorporate payoffs into our decision
process theory.
We believe therefore that we must incorporate two related concepts
from game theory: the concept of a game value and the concept of an
equilibrium set of strategies. Decision process theory is not derived
from game theory so we don’t have a precise method for adding these
concepts. Our method can be illustrated best by using an example and
showing how it is treated in game theory and how we might then
incorporate these game theory ideas into our framework. The example
we have chosen is the Prisoner’s Dilemma.
The prisoner’s dilemma has been extensively investigated by game
theorists. As an example, see Rapoport [1989]. It has been scrutinized in
both theoretical and empirical contexts. The scenario is that two
prisoners are being held for a crime where it is suspected they have acted
in concert. Each is given a choice to confess or not confess with penalties
that are supposed to induce confession of their guilt. However if neither
confesses they will get off lightly. If both confess they will be penalized
but not as severely as the case in which one confesses and implicates the
other. The dilemma is that the prisoners would be collectively best off by
not confessing. However the game theory analysis below recommends
they both confess.
We start with the formulation of the prisoner’s dilemma as a game in
normal form between two players/prisoners specified by the following
payoffs to player 1.
G121 N2 C2
N1 0.1 1 . (1.10)
C1 0 0.9
The payoff values are nominal and don’t influence the conclusions
drawn. There are identical payoffs to player 2:
G212 N1 C1
N2 0.1 1 . (1.11)
C2 0 0.9
Introduction 25
Game theory supposes that there are quantitative payoffs for the matrix
G k of possibilities as illustrated above for player k 1,2 .
We describe the meaning of the elements of the matrix in detail for
player 1, noting a similar description holds for player 2. For player 1, the
rows are labeled (C1) if player 1 confesses and (N1) if player 1 does not
confess. The columns are labeled in a similar manner (N2) and (C2).
The payoffs reflect quantitatively the posed problem: if both confess
player 1 loses 0.9 units, if both don’t confess player 1 loses a much
smaller value: 0.1. If player 1 confesses and implicates player 2 who
does not confess, then player 1 loses 0.0 units. On the contrary, if player
1 does not confess and is implicated by player 2, player 1 loses 1 unit.
Though rather simple, the example illustrates several general
properties of game theory. The payoffs for the two players need not add
up to zero; when they do, Von Neumann & Morgenstern [1944] called
them zero-sum games. Zero-sum games have been analyzed in detail and
have an equilibrium value that is argued as follows.
Suppose the payoffs Eq. (1.10) for player 1 were a zero-sum game.
The most conservative strategy for player 1 would be to determine the
minimum outcome for each of its pure strategy choices. If player 1
chooses (N1), the minimum case is 1 . If player 1 chooses (C1) the
minimum is 0.9 . The maximum of these two is 0.9 . The most
conservative strategy is to choose this max-min. There may not always
be a max-min solution however.
What Von Neumann & Morgenstern [1944] showed however, is that
for a zero-sum game with any number of players, each with any number
of strategies, there is always a max-min using mixed strategies. This is
called an equilibrium strategy. It is the best one can do acting
defensively. When there is equilibrium, there is also a game value equal
to the payoff associated with the equilibrium strategy (or weighted
payoff if the strategies are mixed).
When the game is not zero-sum and there are separate payoffs for
every player, the above argument has been replaced by Nash [1951]. He
showed that again there will be a best choice now called the Nash
equilibrium. Nash says that the best strategy has the property that no
player can do better against this strategy as long as all the other players
adhere to their best strategy. There will now be multiple game values,
26 Geometry, Language and Strategy—Vol. 2
one for each player at the Nash equilibrium. By inspection, the Nash
equilibrium for the prisoner’s dilemma is that each player confesses. The
game value for each player is 0.9 .
Notwithstanding the great insight provided by the Nash equilibrium
and the importance of considering the payoffs separately for each player,
we believe insight into the dynamics comes first from looking at each
zero-sum game and studying how one computes that equilibrium
strategy. For the example chosen here, it is possible to determine that
strategy by inspection. In the general case the max-min problem is solved
numerically. A general method is to first convert the game to a fair game
(one whose game value is zero) which is equivalent in that by
construction it has the same equilibrium value. For player 1, the general
method assumes that the fair game has one additional strategy,
commonly called the hedge, whose payoff is adjusted to get the required
equilibrium:
Fab1 N2 C2 N1 C1 H
N2 0 0 0.1 0 0
C2 0 0 1.0 0.9 m 0.9
1
. (1.12)
N1 0.1 1.0 0 0 1
m 1.0
C1 0 0.9 0 0 1
m 0.9
H 0 1 0.9
m m1.0 m 0.9
1 1 0
This fair game produces exactly the same strategic results as the original
game.
However, computing the equilibrium value for this fair game can be
done as a dynamic process. It is of great interest to us that one of the
methods of finding the equilibrium is to write a differential equation
involving these payoffs:
dV
g eF V . (1.13)
d
The equation says that the rate of change of the mixed strategy is
proportional to the product of the payoff matrix and the mixed strategy.
At equilibrium, the rate of change of the mixed strategy is zero so the
product is also zero. It can be shown that the product is zero is equivalent
to the max-min condition normally ascribed to zero sum games. The
advantage of the reformulation of the game to a fair game is that the
equilibrium is defined by the product vanishing. This technique can be
done for any zero-sum game and any number of players with any number
of strategies.
We believe that we now have a specific attribute to look for in a
dynamic theory. The theory needs to provide in a natural way a payoff
matrix of the type above that is associated with each player. The notions
of equilibrium, Nash equilibrium and game value then have
generalizations in the theory. The dynamics represented by Eq. (1.13) for
a payoff matrix that is antisymmetric is analogous to electrodynamics.
We present the arguments for why this is true and relevant in the next
section.
x Ex y E y z Ez
x Bx y B y z Bz 0
1 Bz
x E y y Ex
c t
1 Bx
y Ez z E y
c t
1 B y . (1.14)
z Ex x Ez
c t
1 E z 1
x B y y Bx Vz
c t c
1 E x 1
y Bz z B y Vx
c t c
1 E y 1
z Bx x Bz V y
c t c
The first equation generates the law of attraction for charges (Coulomb’s
Law); the second equation expresses the fact that there have never been
any magnetic charges observed; the simplest magnet always comes as a
dipole.
The next three equations reflect Faraday’s Law that changing
magnetic fields induce electric currents and hence an electric field. The
last three equations reproduce Ampère’s Law that moving charge
generates a magnetic field. The last three equations however are not
identical to Ampère’s Law because of the presence of the “conduction
current,” the extra gradient of the electric field with time. There is no
contradiction since the early experiments were done using steady state
electric fields and the extra term would have been zero.
Maxwell realized that without the extra term, charge would not be
conserved for time varying fields. With the extra term he got a continuity
equation similar to the equation for conservation of mass:
d
jVj 0 . (1.15)
dt
30 Geometry, Language and Strategy—Vol. 2
We introduce the convention that for repeated indices j we sum over the
possible values x, y, z .
The conservation law agrees with the observation that charge does
not spontaneously appear or disappear: its flow in or out of a box or cell
in space results in an increase or decrease of the charge inside that cell.
So, like mass, charge is conserved within cells.
Maxwell’s equations are consistent for electric and magnetic fields
that vary in time. Moreover, the electric and magnetic fields separately
satisfy wave equations in free space whenever the charge density is zero
0:
1 2E
j jE 0
c2 t 2
. (1.16)
1 2B
j jB 0
c2 t 2
q er e 2e
E 2 r r2 2r . (1.17)
4 r t r x t x
The first term is what one would expect if the force were instantaneous.
In Maxwell’s theory, however, the force is not instantaneous. The
effect of charge propagates with the velocity c and so is not felt at a
given point until it has propagated there. Thus the force one gets is based
Introduction 31
on the “retarded” distance of where the charged body was. The additional
correction terms to Coulomb’s law are based on doing that arithmetic
correctly. The correction terms are not measurable until the speeds are
comparable to the constant c .
Moreover, the equations make connection with a seemingly unrelated
phenomenon: light. We have learned a good deal about light since
Newton. We know light travels as a wave and we know further that
under certain circumstances it behaves like a particle moving in straight
lines. The above analysis construes light as a special case of
electromagnetic radiation. The constant that appears in the equations is
the speed of light (in a vacuum) and is approximately 186,000 miles/sec.
For such speeds, it is not surprising that correction terms were not
initially noticed. A speed of 50 miles/sec, which seems quite fast being
180,000 miles/hour is only 0.00027 the velocity of light. The correction
terms above are actually of the order of this amount squared, since the
coefficient of the corrections to first order is zero.
Maxwell’s equations have been written in a greatly simplified
notation since Maxwell. The electric and magnetic fields are now
understood to be part of a single tensor structure in space-time. We note
the simplifications that occur when the electric and magnetic fields are
written as a single antisymmetric electromagnetic field Fab where the
indices a and b can take on the three spatial coordinates and one time
coordinate (represented as x0 ct ):
Fx 0 Ex Fy 0 E y Fz 0 Ez
Fxy Bz Fyz Bx Fzx By . (1.18)
Fba Fab
x F0 x y F0 y z F0 z
x Fyz y Fzx z Fxy 0
x Fy 0 y F0 x 0 Fxy 0
y Fz 0 z F0 y 0 Fyz 0
z Fx 0 x F0 z 0 Fzx 0
. (1.19)
1
0 Fz 0 x Fzx y Fzy Vz
c
1
0 Fx 0 y Fxy z Fxz Vx
c
1
0 Fy 0 z Fyz x Fyx V y
c
Fab a Ab b Aa . (1.21)
dx 0
cdt , dx1 dx , dx 2 dy , dx 3 dz
g 00 g11 g 22 g33 1
. (1.22)
a b g ab 0
1 1
c Fab g bc gabV b b F ab V a . (1.24)
c c
We introduced for the time component of the flow, V 0 1 .
To complete the comparison of the electromagnetic fields to payoffs
Eq. (1.13) we need the behavior of charged matter in an electromagnetic
field, the Lorentz force:
34 Geometry, Language and Strategy—Vol. 2
dVx
m eEx e Vy Bz Vz By
d
dVy
m eE y e Vz Bx Vx Bz . (1.25)
d
dV
m z eEz e Vx By Vy Bx
d
We put this into the covariant form, which also adds a time component:
dV a
m eF abV b . (1.26)
d
Based on the comparison we see that the form of the payoff (Eq. (1.13))
is the same as this (Lorentz force) form for the behavior of charged
matter in an electromagnetic field.
The comparison suggests we identify the hedge strategy with time,
the flow with the mixed strategy flow and the payoff matrix with the
electromagnetic field. There are differences however that we must still
deal with. The dimensionality above is three space-dimensions and one
time-dimension: the payoff matrix has a dimension n 1 based on the
number n of pure strategies. Distances and energy are appropriate to
decisions not physical distances. Both are a consequence of a differential
geometry. This identification is helpful and we deal with these issues in
the next chapter.
1.9 Outcomes
1.10 Exercises
37
38 Geometry, Language and Strategy—Vol. 2
We restate the basic equations from Sec. 1.8 for the electromagnetic
fields in the language of relativity and geometry by going from the
differential formulation to an integral formulation, [Warner, 1971]. We
use the language of forms from differential geometry to represent the
integral quantities of interest.
We start with a 0-form, which is a scalar quantity . It has the
property that its differential, integrated along any closed path is zero:
Dynamics of Decision Processes 41
d 0 . (2.1)
d a dx a . (2.2)
A dx 0.
a
a (2.3)
A Aa dx a . (2.4)
A B Aa dx a Bbdx b 1 2 Aa Bb Ba Ab dx a dx b . (2.5)
1
We use the summation convention: Cf. Eq. (2.22).
42 Geometry, Language and Strategy—Vol. 2
F dF 0 .
K K
(2.7)
We make use of two general properties of n-forms. First, Stokes’ Law for
any n-form G expresses the fact that the integral of the differential of
G over any subspace K of an n-dimensional space is equal to the
integral of G around the closed boundary K of that subspace:
K
G dG
K
. (2.8)
Dynamics of Decision Processes 43
F
K
mn dx m dx n 0 . (2.10)
BdS 0 .
K
The net flux through the closed boundary is zero. The implication is that
the magnetic flux, defined as the spatial components Fmn of the field
intensity, cannot stop or start inside a volume, but flows continuously.
The differential form of this is
1
F
K
m0 dx m
t K 2!
Fmn dx m dx n . (2.13)
The left-hand side is the circulation of the electric flux around a one-
dimensional boundary, and is defined as the electric potential.
This equation is a special case of Eq. (2.7) where one of the
dimensions is time and the initial and final times are very close. In this
case the two-dimensional surface on the left originates from t times a
one-dimensional integral of the boundary, and the three-dimensional
volume on the right originates from the difference of two-dimensional
volumes over a short interval of time. The above result is obtained by
taking the time difference and dividing by the differential time. We think
of the result as the time rate of change of the spatial components of the
payoff flux flowing through a two-dimensional surface determining the
electric potential.
In the general case of a payoff field, the magnetic flux (calling the
space components of the field intensity Fmn the magnetic field) flows
like a continuous media without sources or sinks (lines are not created or
destroyed). Changes in the magnetic flux with time generates the
electromotive force (unit-less) described by the left-hand side of Eq.
(2.13). The magnetic flux lines cause charged particles to move in
patterns identical to the paths predicted by elementary game theory,
identifying the magnetic field components with the payoff matrices of
the players.
In the steady state case, no electromotive forces are generated,
analogous to the situation of steady state electric circuits: The above
equation reduces to Kirchhoff’s Law of electrostatics. The result of
Dynamics of Decision Processes 45
changing flux with time creates electrical forces that induce currents
(motion of charges). Inertial energy is converted to electromagnetic
energy and vice versa.
1
*j ab f j a dx b dx f . (2.14)
n!
1
*F adef F ef dx a dx d . (2.15)
2 n 1!
Utility can be created and destroyed since the current acts as the source
or sink.
This leads to generalizations of Coulomb and Ampère’s law that
relate the differential of the scaled payoff to the current and are
expressed here as a single relation:
*j d * F . (2.16)
*F d * F *j .
K K K
(2.17)
The right-hand side describes the utility. The left-hand side describes the
effort to produce that payoff in terms of the dual payoff *F . The inverse
of the productivity factor is a measure of the effort required to
produce one unit of utility.
Utility is created or destroyed through the flow of currents. The units
of the current are U Ln . As before, these equations represent different
physical situations depending on whether the components of the
integrand are purely spatial or not. If we suppose that on the right, the
volume is purely spatial, then we find that the integral is the charge over
the surrounding volume,
1
n!
K
0mm j 0dx m dx m .
In this case, the left-hand side represents surfaces which are purely
space like, so one and only one of the components can be timelike. The
left-hand side is therefore the net electric field density that is normal
to the (n – 1)-dimensional strategic surface bounding the n dimensional
strategic volume,
F Sm Qencl .
0m
K
m F 0 m ch ,
1
Sm 0 pqm dx p dx q
n 1!
is the dual of a 1-form in the n-dimensional strategic subspace.
The decomposition of a payoff field into its spatial and time
components is frame dependent. In each frame, the space coordinates
determine a volume; the spatial payoff component (flux) that flows
through any two-dimensional strategic surface of that volume is
magnetic. The space-time flux that flows through an (n – 1)-dimensional
strategic surface is electric. The net number of flux lines that emanate
from this spatial volume equals the enclosed utility. This nomenclature
however is unnecessarily tied to the physics analogy, though at times it is
comforting and provides a useful mnemonic for looking up standard
results. We introduce new terminology to emphasize we are not tied to
that analogy.
We have said that the total utility (“charge”) is determined by the net
flow of the electric components of the payoff field through the strategic
surface. This supports the idea that utility is the source of value as used
in game theory: utility determines the electric field and value as used in
game theory becomes the electric field in decision process theory. We
therefore suggest the electric decomposition of the payoff fields be
valuation fields. Further, we suggest negotiation fields be the magnetic
fields. We suggest this since there are always two distinct negotiation
strategies for which the payoff represents the payout. Each pair of
strategies defines a negotiation plane. Each payout represents a flow of
negotiation flux lines through this negotiation plane. In general, since
the negotiation field is an antisymmetric matrix, at each point we can
always find a frame of reference in which the matrix is block diagonal
reflecting a disjoint union of negotiation planes. If we think of the
48 Geometry, Language and Strategy—Vol. 2
k F jk 0 F j 0 j j
. (2.18)
k F jk j j 0 F 0 j
1 F mm F 0m
2!
t K
m
S mm j S m Sm . (2.19)
K K
The right-hand side we call the enclosed utility current, including the
contribution of the displacement current that is induced from valuation
fields that change in time, flowing through the strategic surface.
The boundary can be thought of as the dual of the 2-form created
from the wedge product of the surface 1-form and a second intersecting
surface 1-form that defines the boundary. The dual of this 2-form is
1
S mm 0 pqmmdx p dx q ,
n 2 !
Dynamics of Decision Processes 49
a ja 0 ,
Q
*j d * j 0 j Sm
m
. (2.20)
K K K
t
This states that the rate at which the enclosed charge (utility) changes in
time is equal to the net current leaving the n-dimensional strategic
(space) volume K through the enclosing n 1 strategic surface K .
Charge is not created or destroyed, but flows like a fluid through space.
We have anticipated our adoption of this generalization of the
electromagnetic field as the payoff field. We have provided the units that
make sense for this generalization. Implicit in our approach is the notion
that decision strategies change continuously over time. In the next
section we introduce the basis for this anticipation.
50 Geometry, Language and Strategy—Vol. 2
In many approaches to decisions (Cf. Sec. 12.1), the key aspect is the
uncertainty of the decision and its independence from past events. The
approach is to use Bayesian or inverse probability to estimate future
behaviors. It is a stochastic approach. Yet in modern businesses, the
opposite holds: businesses practice continuous process improvement as
the basis for improving profits and quality of their products.
Incrementally improving past behaviors is the key to future successes.
We believe as well that continuous change is a necessary attribute of
decisions and can be applied to mixtures of strategies viewed as
frequencies (Sec. 1.4). Decision processes can be put into a normal form,
albeit locally. For sufficiently short periods of time and sufficiently small
strategic variations, the foundational aspects of game theory provide a
framework for decision process theory. This is a statement of “flatness”.
To be locally flat however is not to say that that is true globally.
To discuss strategic change, we distinguish between changes in
choices that occur at a constant rate and those that don’t. The latter
changes are characterized by acceleration, which is the rate at which
strategic rate changes occur. In many cases we are unaware of such
accelerations. We will often refer to the physical analogy of our
perception that as we sit comfortably at home at rest, we are unaware that
we are spinning around the North-South axis once every 24 hours and
rocketing around the sun every 365 ¼ days. Yet those of us that pay
attention to weather are aware of the induced acceleration effects
attributed to Coriolis, because they contribute to the formation of
cyclonic weather conditions [Crawford, 1978]. A discussion of dynamic
changes involves not only the ability to order our preferences, but the
ability to measure the magnitude of change of those preferences and
whether the changes occur at a constant or accelerating rate. As with
weather, we must learn to identify when such acceleration effects are
present. No acceleration is yet another type of “flatness”.
To address these issues we apply what we feel are common sense
notions of space (applied to strategic choices) and time to the domain of
decision making. We assume that choices that are near to each other
strategically or near to each other in time, lead to actions that are
Dynamics of Decision Processes 51
ds 2 dt 2 dx a dx a 0 . (2.21)
a
ds 2 g ab dx a dx b . (2.22)
Dynamics of Decision Processes 53
There are three things to keep in mind: first, if there are repeated indices
with one upper and the other lower, add a summation “sigma” for that
index. Second, allow the indices a, b to represent not only the
allowed strategies but also time. Third, specify all the tensor quantities
so that the desired expression is replicated.
In this case we replicate Eq. (2.21) if the non-zero components are
gtt 1 and the strategy components are diagonal and gaa 1 . The
geometry of this space is not Euclidean but the curved Minkowski space.
For our purposes in this book we call spaces with a Minkowski metric
flat.
x a x 0 x1 x n ,
a b b a .
54 Geometry, Language and Strategy—Vol. 2
y a x 0 , x1 ,, x n
that depend on the old coordinates. The new frame of reference also
provides a holonomic coordinate basis.
These frame considerations allow us to implement the common sense
notion that the specification of pure strategies may not be unique: many
equivalent choices are possible. We thus avoid falling into the trap that
every player adopts the same set of values, since they may be measuring
values using different coordinate bases. We create a decision process
theory in which any frame of reference is an acceptable choice. We
assume covariance, that the fundamental dynamics will be expressed by
a mathematical form that does not depend on which choice we make.
A natural consequence of our dynamic theory is the idea that the
strategies change continuously over time (due to a principle of least
Dynamics of Decision Processes 55
y a x 0 , x1 ,, x n
dy a dx b b y a dx b . (2.23)
x b
Note our use of the summation convention to simplify the writing and
our introduction of a more compact notation to indicate the partial
derivatives. Though the expression can be a laborious computation, the
essential conceptual point is that the new differentials are linearly related
to the old differentials.
We expand the concept of a frame transformation by allowing any
linear transformation of the differentials, including those that are non-
holonomic. The resultant expression E we call a 1-form using the name
from the theory of differential geometry (Sec. 2.1), which, when these
56 Geometry, Language and Strategy—Vol. 2
infinitesimal quantities are summed over a path, provides the basis for
defining a line integral:
Ea E a b dx b . (2.24)
c E ab b E a c ,
which follows from the equality of the mixed derivatives in Eq. (2.23),
c b y a b c y a .
c E ab b E a c
for at least one pair of coordinate values. The usual rules of calculus
don’t hold. Changes to fields now depend on the path taken.
We assumed that at each point there is a frame in which the cause and
effect signal propagates outwardly as a spherical wave. In a general
frame E a we assert that this implies that the propagation is determined
by the vanishing of the quadratic form ds 2 g ab Ea Eb constructed from
the frames. The coefficients are determined by the transformation.
Cause and effect must apply to all physical phenomena. We expect
that regularities of physical phenomena will involve the rates of change
of variables, such as a scalar field x a , a vector field X , or an order 2
tensor field gab . There will be higher order tensor fields to consider as
well. We need rules that apply to any tensor field and its rates of change.
In the frame in which space is locally flat, rates of change are determined
by ordinary calculus, by the derivatives. Transformation to rotating
frames requires that the effects of the acceleration be included in the
definition of the rates of change. We again recall that the Coriolis force
due to living on a spinning sphere (earth) is an example of a rotating
Dynamics of Decision Processes 57
The basic rules for including rotating frame effects for any tensor field
follow from how we treat the rate of change of a vector field. The rate of
change of a vector field along a path parametrized by a scalar is the
difference of the field at two different points divided by the differential
change d along the path. In dealing with rotating frames we must be
concerned with Coriolis-type forces: we must be clear about how we
compare vectors at two different points. In physical applications, we
make the comparison by parallel translating the initial vector along the
curve to the comparison point. This removes the effect of the rotating
frame, exposing only the actual change in the vector field.
As a result, the rate of change of the vector field will be a differential
change of its values plus a contribution due to the frame rotation, which
linearly mixes the vector components:
DX a dX a ωa b X b . (2.25)
dX a b X a dx b .
DX a X a ;bdx b .
X a ;b b X a a cb X c . (2.26)
ωa b a bc Ec .
a Ea b b ,
dX a b X a Eb . (2.27)
In a space that is locally flat, all frames will be torsion free. The 1-form
has properties of a vector field so we expect the two terms for the
differential of a vector 1-form. The 1-form represents a differentially
short path segment of a curve in space. The line integrals are constructed
Dynamics of Decision Processes 59
A B Aa dx a Bbdx b 1 2 Aa Bb Ba Ab dx a dx b . (2.29)
C a bc C a cb
dΕ a 1
2 E
b
a
c c E a b dx b dx c
. (2.30)
dΕ a 1
2 C a bc Eb Ec
c c
Dgab dgab ω a gcb ω b gac . (2.31)
We require
Dg ab 0 .
acb abc .
bac abc ,
We have stated that in decision process theory, the rules for decisions
should be covariant. The basis for this comes from utility theory in game
theory, in which preferences are determined only up to linear
transformations (Sec. 12.2). Thus for any given event, we express the
theory in such a way that the rules are independent of linear
transformations. Theories that are invariant under transformations have
characteristic geometric structures, such as scalars, vectors and tensors,
which we will define in more detail below. We start the discussion by
laying out in detail what this covariance implies.
We identify a geometric structure as a set of related functions in the
theory that maintain their form under transformation. The simplest
example is a single function that transforms into itself. We define a field
to be a scalar field if , which indicates no change under the
transformation other than expressing the field in terms of new variables.
The next more complicated structure is a collection of functions in 1-
1 correspondence to the number of space time dimensions. The simplest
example is that of functions that define a frame. For each direction b , the
frame transformation components E a b form the components of a
contravariant vector2 in the new basis: the collection of such vectors is
E a . (One can equally well look at each direction a , and the frame
transformation is a set of covariant vectors in the old basis; similar
statements can be made about the inverse frame transformation Ea b .)
We consider a frame transformation λE , which is a product between
the matrix λ , which is a frame transformation in its own right, and the
frame components E that transform a set of frames into a new set
(denoted by the prime):
E a a b E b . (2.35)
As with the scalar field, the new frame components are also expressed in
terms of the transformed coordinate values.
2
It is defined in the next paragraph.
62 Geometry, Language and Strategy—Vol. 2
ω C λ x λ y ω C . (2.36)
ω y λ y ω y λ 1 y λ y dλ 1 y . (2.37)
The first term is the transformation of a mixed tensor. The second term
indicates that for rotating frames, the orientation potential is not a mixed
tensor.
It is also true that the ordinary derivative of a contravariant vector
field is not a tensor since the transformation of the differential of a
contravariant vector field is:
Putting this together with Eq. (2.37), we see that the covariant derivative
Eq. (2.25), formed from the parallel translation of the contravariant
vector field, does transform like a vector field.
The transformation components of Eq. (2.37) that don’t transform as
a vector field cancel the corresponding components of Eq. (2.25) of the
differential of the vector field:
In particular this holds for the torsion free coordinate basis vectors:
Frames are geometric structures defined over the whole space; they
transform as vector fields.
We thus have a proof that local gauge invariance for a space that
is locally flat (Minkowski) implies that every frame is torsion free. This
provides a justification for Eq. (2.28). Because the derivative D that
results from parallel translation preserves the tensor field
transformation, the name covariant derivative is appropriate. It is a
differential operator that preserves the geometric structures of the theory.
Although the orientation potentials are not tensors or geometric
structures that are preserved, they are associated with such structures and
64 Geometry, Language and Strategy—Vol. 2
R dω ω ω . (2.41)
This orientation flux field carries energy and momentum through the
2-surface, is a tensor field and provides the physical field that propagates
the cause to the effect. In physical theories it is called the curvature
tensor.
If we can relate this physical orientation flux field to the forces, we
will have specified decision process theory as a dynamical theory. We
make a start at this in the next section where we explain how hidden
dimensions lead to the creation of decision process theory and the payoff
fields suggested in Eq. (1.26) are identified as frame rotation effects.
The challenge is bridging the view we have been developing. On the one
hand, that view approaches decisions in terms of the strategies or choices
available to a player, considered as applied dimensions of a many-body
system. On the other hand, that view sees the dynamic content suggested
from game theory as resulting from payoffs for each player.
We find similarities with a challenge posed in physics to unify the
applied forces viewed in space and time with internal symmetries such as
electromagnetic fields. This is by no means obvious. Suffice to say at this
introductory point that a solution by Kaluza [1921] and Klein [1956] was
proposed almost a century ago to consider the electromagnetic fields as
an internal symmetry or hidden symmetry represented by an additional
dimension to be added to ordinary space and time. This only works if the
additional dimension plays no active role in the dynamics in a sense that
we make precise below. In some ways it is like a constraint as opposed to
one of the applied dimensions.
We shall see that the analogous mechanism for decision process
theory is to add a player-specific (internal) strategy for each player,
Dynamics of Decision Processes 65
which we call inactive from the usual perspective of strategies. The usual
strategies we shall call active. This approach, we believe, resolves the
conceptual problems and provides the necessary bridge. We use this as
the framework for decision process theory.
It is thus our view that decisions are carried out jointly by one or
more subjects who act based on their personal knowledge of past actions.
There is some attribute or behavior of each player or agent that is
persistent; that does not change with time. The notion of persistence is
tied with our notion above that the player-specific strategy is inactive. In
contrast to previous authors, we view that the players’ active behaviors
or strategic choices need not be persistent, rational or necessarily
converge to some equilibrium flow. Their choices in general are variable
precisely because they learn from past outcomes and are affected by
network relationships.
Following this idea, we add inactive dimensions that we associate
with the players in the decision, to the active strategy dimensions
associated with the pure strategies. In a coordinate basis, the most
general form of the line element Eq. (2.22) with hidden dimensions j
is:
We indicate active strategies by the indices a, b, and the inactive or
hidden strategies by the indices j, k , . We define hidden to mean that
the inactive metric components jk , payoff potentials Aaj and active
metric components gab , are independent of the inactive dimensions j .
For the purpose of this discussion time is active. The hidden or inactive
dimensions play a restricted dynamic role.
In differential geometry, when there is a frame in which the metric
elements have the form Eq. (2.42) and are independent of the dimension
j , then there is an isometry associated with that dimension: the metric
does not depend on that dimension. As part of decision process theory
we postulate that there is at least one isometry for each player in a
decision. To the extent that players may exhibit schizophrenic
tendencies, there may be more than one isometry for each player.
66 Geometry, Language and Strategy—Vol. 2
U j d j Aaj dx a ,
by which we mean dU j 0 :
Ua dxa
. (2.43)
U j d j Aaj dxa
j j j x
Aaj Aaj a j .
To identify the payoff potential with the payoff matrix that occurs in
the flow Eq. (1.13) suggested by game theory we compute the behavior
of geodesic curves defined in terms of the (vector) flows:
dx a
Va
d
d j
Vj
d . (2.44)
a
DV
V ;bV V ;kV 0
a b a k
DV j
V j ;aV a V j ;kV k 0
U j U bj kj Abj
U U ka a
U k U b 0 ba
U kj U bk jk Abk
U U a
1
U j U ba 0 ba
. (2.45)
j U j j
a U a a Aak k
ds 2 jk U j U k g ab U a U b jk U j U k g ab U a U b . (2.46)
dU a 0
dU j F j 1 2 C j ab U a U b
C a 0 . (2.47)
C j kl C j ka 0
C j ab Fabj
In other words the active components are exact and the inactive
components are orthogonal to the active components, but internally are
not orthonormal. So the active orientation potentials are determined by
the active metric. We still need to verify that the payoff field that appears
here is also the one associated with the flow equation. To do this we need
the orientation potentials that appear in Eq. (2.44).
The orientation potentials that have one or more inactive components
are determined algebraically from Eq. (2.32):
3
Greek indices cover both active and inactive dimensions.
Dynamics of Decision Processes 69
jka 1 2 Δ j ka Δ k aj Δ a jk 1 2 a jk
jab 1 2 C jab Cabj Cbja 1 2 jk Fabk
ajb 1 2 Cajb C jba Cbaj 1 2 jk Fabk . (2.48)
abj 1 2 Cabj Cbja C jab 1 2 jk Fabk
abc 1 2 a g bc b g ca c g ab
DV a dxb d j
V a ;b V a; j V a;bV b V a ; jV j 0
d d
dV a
a cbV cV b a jbV jV b a bjV bV j a kjV kV j 0 . (2.49)
d
dV b
g ab abcV bV c Vk Fabk V b 1 2 V jVk a jk
d
We see that the payoff fields have two distinct contributions based on
the decomposition of the payoff field into its valuation and negotiation
fields. There is one force (acceleration) along the valuation field
direction and is proportional to the utility density. There is a second force
due to the negotiation field. Unless the flow is in the two-dimensional
plane defined by the negotiation strategies, this force is zero. The force
is magnetic: it is orthogonal to the flow and to the normal to the plane.
Negotiation leads to change when one of the strategies is along the
direction you want to go and the other is an independent alternative with
a non-zero payment.
We expect additional contributions both from the orientation
potentials in the active space abc as well as from the scalar gradients of
the inactive metric. We see that the inactive space though hidden
contributes to the dynamics, albeit in a restricted way. As an example
from the physical domain, we note that the rotation of the earth is also
hidden because we think we are at rest; the Coriolis force is the evidence
from the orientation potentials that we are spinning.
The inactive flow appears as a “charge” density4 for player k , which
we can call the interest flow for player k . We now extend game theory
by exploring the properties of the player interest flow. We demonstrate
that its value stays the same (is conserved) along a geodesic. An attribute
of a spinning body is the conservation of angular momentum. There is an
analogy to decision theory. The geodesics along the hidden directions
can be computed from Eq. (2.26):
DV j dV j
j cbV cV b j kbV kV b j
V cV k 0
d
ck
dV j
d
jl lk V k 0 . (2.50)
d d
dVk
0
d
4
To be more precise, the inactive flow is the ratio of the charge to the mass.
Dynamics of Decision Processes 71
2.12 Outcomes
that we have not dealt with, which we call inertial forces. They are
constraints and play a necessary role in the realm of decision making and
are so far absent in our theory. In the next chapter we explore such
inertial forces using as a learning example, the behavior of fluids and
suggesting how those effects may be included in the theory of decisions.
We here summarize the chapter from the standpoint of using this
volume in a course. By the end of the chapter, the student should be able
to do the following using this framework:
Understand global behaviors of the theory
For specialized cases be able to apply global theory to realistic
decision processes
Be able to identify and use the underlying connections of physical
theory to gain insights into decision processes
Be able to classify general decision processes in terms of a standard
taxonomy.
The attainment of these outcomes will be facilitated by doing the
exercises in the next section. As a help, the detailed outcomes expected
of the student are listed below.
Understand units that are used in decision process theory—Sec. 2.1.
Understand the concepts of geometric forms—Sec. 2.2.
From Sec. 2.3 understand the source-free Maxwell equations as
expressed in the language of forms.
From Sec. 2.4 understand the Maxwell equations that arise from
sources as expressed in integral form.
Be able to translate these equations into the language of game
theory—Sec. 2.5.
The student should learn the essential point that aspects of decisions
are deterministic, generate actions and are influenced by actions—
Sec. 2.6.
From Secs. 2.7-2.8, be able to distinguish between transformations
that are holonomic and those that are not. Furthermore, actions are
related by cause and effect, which propagate at a finite speed.
The geometric properties of these effects are determined by the
common sense properties that transformations are torsion free and
measure independent, and these properties determine the orientation
potential in terms of the metric and structure constants—Sec. 2.9.
Dynamics of Decision Processes 73
From Sec. 2.10, the student will have learned that the concept of
covariance is that the rules of the theory are expressed in terms of
geometric structures that maintain their property over time and space.
The key such properties are scalar fields, vector fields and tensor
fields. A consequence of the theory is that cause and effect are carried
by the orientation flux field, called the curvature tensor in differential
geometry. It measures the degree to which global space-time is not
flat.
From Sec. 2.11, the student will have learned that hidden dimensions
provide the mechanism to create decision process theory. Like the
Coriolis acceleration, payoffs reflect accelerations associated with
hidden dimensions and provide a game theory mechanism for
strategic behavior in decision process theory. They also provide a
persistent conserved quantity analogous to charge conservation in
physics.
2.13 Exercises
zone with a bend to the right [Crawford, 1978]. Show that Coriolis
forces arise because of the spin of the earth and can be expressed as
a non-diagonal metric term gt tt A , [Ryder, 2009]. The vector
potential A gives a characteristic payoff z F and a velocity
( v ) force term ω v . Assume points on a sphere are measured in
latitude and longitude . Show that this follows from the
conservation law Eq. (2.50) with time as the inactive coordinate. If
there is azimuthal symmetry, show that this symmetry produces a
centripetal acceleration r .
(6) Show that the earth’s rotation produces forces that represent three
dimensional cyclonic flows that rise (increase the distance r from
the center of the earth), and move for example from the equator
towards the pole, fall and then travel back to the equator. Such
flows divide the earth into atmospheric cells that play a critical role
in global weather formations [Crawford, 1978]. They generate
frame orientation effects that would be unexplained if we were
unaware of the rotation.
(7) To obtain a geometric visualization of Eq. (2.19), imagine a small
area of space in which a thin and short hyper-cylinder of current is
along the z axis transverse to the cylinder plane with radius r .
The surface of the cylinder is a n 2 dimensional sphere. In this
region, show that the payoff flux components F rz are the only
contributions (approximately) to Eq. (2.19) and that they are
multiplied by the hyper-length S rz . Discuss the meaning of these
flux components and the components of Frz in Eqs. (2.10)
and (2.13).
Chapter 3
Inertial Behaviors
75
76 Geometry, Language and Strategy—Vol. 2
included in the kinematics of the last chapter. Other behaviors are based
on the constraints not included. Newtonian mechanics views all
background forces as being generated by characteristic masses and their
accelerations. In more general theories such as Maxwell and Einstein
theories, the background forces are characterized by their energy
densities and energy flows. In this way we take into account background
forces. We call these the inertial attributes and incorporate them into our
theory.
Looking purely at the kinematic or non-inertial aspects, we provided
in the last chapter a motivation for incorporating the payoff field of each
player into decision process theory. We borrow the term “inertial” from
the physical sciences: an inertial force reflects the resistance of a body at
rest to move or a body in uniform motion to change direction. In
Newtonian physics it is the product of the mass and the acceleration. By
extension, an inertial frame is one in which these attributes of
Newtonian physics hold. By contrast, a non-inertial frame is one in
which these attributes don’t hold. Two common examples are bodies in a
gravitational field and the Coriolis Effect seen by an observer on a
spinning globe. A slightly less obvious example is an observer on the
inside of the spinning globe feeling the effect of the centripetal
acceleration. We argue that vorticity or spin also plays a role in decision
making and so we anticipate the importance of non-inertial frames.
We say gravitation is non-inertial because the acceleration of an
object is independent of its mass: it reflects the equality of inertial and
gravitational mass. In this language, the effects of the payoff field are
non-inertial and are due to looking at the problem in a non-inertial
frame: these are non-inertial forces. In decision process theory, a system
subject only to such non-inertial forces follows a geodesic, Eqs. (2.49)
and (2.50). It is the generalization of Lagrange’s Eq. (1.9) with only the
kinetic term L T . In this chapter, we expand decision process theory to
include inertial forces.
Our reason for expanding the theory is the belief that in its current
form, the theory though already an extension of game theory, does not
yet describe important effects that are clearly part of realistic decision
processes. In the business world we are aware of inertial forces that
reflect how hard it is to move some organizations to become more
Inertial Behaviors 77
S Ldt .
path
(3.1)
There is one path that is special, the one in which the action is an
extremum. To find this path imposes a condition on the Lagrangian,
which leads identically to Eq. (1.9) and so we are justified in stating that
all mechanical systems obey the following principle of least action:
S Ldt 0 .
path
(3.2)
Inertial Behaviors 79
S Lcell dt . (3.3)
cells
The smaller the cells, the better will be the approximation. The
Lagrangian Lcell for each cell is determined by the kinetic energy Tcell
minus the potential energy Vcell . It is convenient though not mandatory to
consider the cells to be cubes and use continuous variables to describe
their essential behaviors.
We do this, starting with the kinetic energy, which is determined by
the product of the mass density, the square of the velocity of the cell and
the volume of the cell:
This gives the total force along the x axis due to the forces being exerted
from all six faces.
There are similar expressions for the other two components of force,
all of which can be compactly summarized as follows, where there is a
sum over repeated indices that represent the three coordinate axes:
1
Space can always be partitioned into cubes. What is important is the partitioning, not
the exact shape. If other shapes are considered, it may be necessary to consider more than
one such shape to “fill” space.
Inertial Behaviors 81
f j k t jk . (3.6)
d
dt
g j dxdydz f j dxdydz k t jk dxdydz . (3.7)
g j
k Vk g j k t jk . (3.8)
t
For a continuous media, these are Euler’s equations which represent the
conservation of momentum: the rate of change of the flow times the
matter density is the transverse gradient of the stress.
d
jVj Vj j jVj 0 . (3.9)
dt t
This result reflects one attribute of the resistance to change of inertial
systems. It generalizes to inertial media by extending the summation to
all the spatial directions.
The momentum density and matter density are thus determined by
two coupled equations:
jgj 0
t
. (3.10)
g j
k Vk g j k t jk
t
These two equations generalized to inertial media represent the major
result for this section, though not yet written in the most illuminating
form. We will write these results using the principle of least action. To
incorporate these results into decision process theory, we will further
rewrite them in a fashion consistent with the terminology from the last
chapter.
To achieve these goals, we put the equations into a form that
combines both the space and time aspects together in a covariant form.
We introduce the absolute stress:
For a frame in which the media is moving, the stress tensor t jk will in
general not be symmetric; however the absolute stress will be symmetric
(Tolman, 1987, p. 65ff).
We next rewrite Eq. (3.10) in terms of these absolute stress
components:
g j
k p jk 0
t . (3.12)
jgj 0
t
Inertial Behaviors 83
The second equation shows that the increase in matter is due to the
amount of momentum that flows into the cell; the first equation shows
that the increase in momentum is due to the amount of stress that flows
into the cell.
As an aside, a good idealization for many common materials is that of
a perfect fluid, one in which the absolute stress is determined by a single
quantity typically called the pressure: p jk p jk , where the “delta”
notation indicates that the value is unity whenever the indices are the
same and zero otherwise. The above equations then become:
V j
k V jVk p jk
t . (3.13)
d
jV j 0
dt
At low velocities, the first equation is Euler’s equation that the rate of
change of momentum is minus the gradient of the pressure. When there
are no applied forces, a system in constant motion remains in motion; a
system at rest remains at rest. These are the expected effects for inertial
forces.
For decision process theory if we adopt the concept that inertial
effects are characterized by a density that is a continuous function of
position and time, we expect to get equations of this type as well. The
conservation law (second equation) in Eq. (3.13) expresses the
conservation law for inertia, namely the net flow of inertia into a cell
determines the rate at which the amount of inertia inside that cell
increases.
Returning to the general case, we represent these results with a matrix
that adds to space a new dimension proportional to time: x0 ct . The
proportionality constant has units of a velocity so that the new dimension
has units of distance. We define the components of a tensor, a field with
two indices, Tab as follows:
84 Geometry, Language and Strategy—Vol. 2
T jk p jk
T0 j cg j
. (3.14)
T00 c 2
Tab Tba
The absolute stress, the momentum density and matter density combine
to form a symmetric energy-momentum stress tensor. The conservation
laws of momentum and matter in this notation are:
0T j 0 k T jk 0
. (3.15)
0T00 jT0 j 0
The component T00 has the units of energy per unit volume, an energy
density, since the proportionality constant is a velocity and mass time
velocity squared has units of energy. Similarly the component T0 j has
units of energy density since it is a momentum density times a velocity.
We now have a form for the inertial media expressed in the notation
from the previous chapter:
g bc bTac 0 . (3.16)
We will write a consistent principle of least action for inertial media (the
generalization of matter) and for payoff fields (the generalization of
electromagnetic fields). We saw in the last section that inertial media
communicate effects with a finite velocity, not instantaneously. This is
also a striking aspect of Maxwell’s equations and their payoff field
generalizations. Information is transmitted from one part of space to
another via the electric and magnetic fields at the speed of light as
indicated by the wave equations Eq. (1.16). We expect that this will be a
general consequence of the principle of least action for any type of
continuous media.
The action for the electromagnetic field and its interaction with
matter is:
The extremum for this action yields Maxwell’s Eq. (1.24) that includes
the interaction of the matter with the electromagnetic field. The major
difference between matter and radiation is that the electromagnetic
86 Geometry, Language and Strategy—Vol. 2
g bcbTac 0 . (3.19)
2
The energy-momentum tensor is also conserved in the presence of matter. We would
write the total energy-momentum tensor as the sum of the electromagnetic field and the
matter field. Each have an energy-momentum that is conserved separately.
3
An electrical engineer understands that energy is really stored in capacitors as the
electrical field and in inductors as the magnetic field. These fields are observed to be real
and often quite powerful.
Inertial Behaviors 87
We recall that the Newtonian view is based on our common sense that
certain forces, such as gravity and charge are instantaneous. We showed
that this view, however, is not consistent with Maxwell’s Eqs. (1.20) and
(1.24), which show that the Coulomb force Eq. (1.17) propagates with
the velocity of light. The Newtonian view also maintains the common
sense view of the existence of rigid bodies, which have the same defect
of propagating effects instantly from one part of the body to another.
However, we have suggested that real bodies are continuous and
communicate from one part of the body to another at a finite speed. We
were thus led to describe inertial effects by means of inertial media.
Einstein unified these viewpoints into a theory of relativity in which no
instantaneous transmission of forces exists and in which there are no
rigid bodies. We use his new common sense view as basis for our general
decision process theory.
3.4 Distance
x A xB y A yB z A z B .
2 2 2
s (3.20)
know that the charts lose accuracy away from that point, so when we
travel we need a number of charts in order to navigate around the earth.
The charts have to match on their common areas. It is possible to
describe the sphere in this fashion as a series of two-dimensional charts.
The collection of all the charts will then be our atlas. The structure
defined in this way is termed a differential geometry by mathematicians
and used by physicists such as Hawking & Ellis [1973]. It is also studied
in a branch of mathematics called fiber bundles [Steenrod, 1951].
It provides a precise language with which to describe all the attributes of
such structures.
The differential geometry makes only the assumptions at the point
that reflect our direct experience. At the given point we know that any
rotation of the two dimensions yields an equivalent set of charts. In order
to navigate between points we need to know the symmetry structure at
each point and we need to know how to glue the charts together, which is
tantamount to our knowing how a path appears on each chart as it moves
over the surface of the earth so that the charts provide a faithful
representation of the physical geometry. To apply this to any system, we
have to identify the dimensions in which we live, our topological space,
identify the local symmetries that are present for all dynamics and
identify the charts that are needed to describe motion. For decision
process theory, we generalize Einstein’s approach to identify the
topology as consisting of strategic space and time.
Einstein takes the implied form of the metric from Maxwell’s
equations. Consider an electromagnetic wave that satisfies the wave
Eq. (1.16). The waves are real numbers, but are usefully thought of as a
superposition of complex numbers:
2 c 2k k 0 . (3.22)
This expresses the fact that the wave travels with a constant speed c
called the speed of light. Einstein identifies a 4-vector as one in which
the time component is the frequency and the space components are
the wave numbers ck . With this identification the length of the 4-vector
is determined by the metric we suggested earlier, Eq. (1.23).
Though the Newtonian view and Maxwellian views are different, it is
possible that both are correct in their domain of application. However
there are phenomena that require both theories as part of the explanation.
The speed of light from a moving object should increase or decrease its
speed according to the Newtonian view; the energy (frequency) of the
light should increase, but the speed should stay the same according to the
Maxwellian view. Michelson and Morley demonstrated that the speed in
fact stays the same, showing that the Maxwellian view is more accurate.
The effect they observed was very small, so that no known deviations to
Newtonian mechanics would have been observed. Einstein’s method was
to provide a theory of mechanics consistent with the Maxwellian view.
He took as the observational fact the form of the metric as used in
Maxwell’s equation. Based on what was understood about
electromagnetic phenomena, he postulated that the descriptions of all
physical phenomena would have the same form in any frame of reference
as long as one understands that measurements are done using this metric,
called a Lorentz metric in physics after the physicist who first articulated
its importance:
s c 2 tb ta xb xa yb ya zb za .
2 2 2 2
(3.23)
a b
distance between points gab dx dx as the product of the metric and the
product of two coordinate distances, summed over all such product
possibilities. The continuous metric (media) replaces the rigid measuring
rod. Two observers moving relative to each other will use different
metrics to describe the same physical process. At any point in space and
any instant of time, it is always possible however to find a frame of
reference in which the metric is flat, Eq. (3.23); however at some other
point and some other point in time, this observer will in general not see
the same flat metric.
We conclude from the previous section the foundational results that unify
inertial fields with the payoff processes of game theory. Since rigid
bodies do not exist in the Einstein view, we say that matter (inertia) is
described in terms of fields. Such inertial fields are determined by the
Lagrange principle of least action in terms of the Lagrangian density,
where we integrate over space and time, analogous to the examples we
have given earlier for the Maxwell fields:
S L g dxdydzdt . (3.24)
This form for the Lagrangian can be used to specify a form for matter
that is continuous and consistent with Maxwell’s equations and our new
common sense. For decision process theory, this suggests that we take
seriously not only the principle that flow is influenced by the payoff
fields Eq. (2.49) but that we apply the generalized version of Maxwell’s
Eq. (1.24) to decision processes: decision flows are the sources for the
payoff fields. We say more about the exact form of the generalization in
the next section. Here we clarify the relationship between the inertial
fields and the action.
We require the form for the Lagrangian in Eq. (3.24) to be covariant
in terms of the metric. First we assume that the Lagrangian density L is
a scalar, i.e. it has the same value in every frame of reference. The action
is a scalar, so we require the unit of volume to be a scalar. So for
94 Geometry, Language and Strategy—Vol. 2
g bcTab;c 0 . (3.26)
Thus we recover and generalize the description given earlier for the fluid
example of continuous matter, we obtain a description that is consistent
with the observations of Maxwell for charged continuous matter and we
provide a covariant description whose equations are generated by the
principle of least action. We thus accomplish our goal of incorporating
inertial resistance and inertial cohesive forces into decision process
theory through the least action principle Eq. (3.25). We specify possible
models by specifying possible inertial media contributions to the energy
momentum tensors Tab .
Inertial Behaviors 95
R dω ω ω . (3.27)
Stokes’ theorem relates the total field through a bounded area to the sum
of differential quantities around the boundary. For a general group, in
which some elements don’t commute, the total field through an area
bounded by a closed loop equals the covariant differential of the
potential integrated around the loop. At a deep mathematical level, the
orientation flux field above is this covariant differential.
For the electromagnetic field the second term of Eq. (3.27) is zero.
This is not the case for the group of frames, the symmetry group
associated with the orientation potentials. Ordinarily, the orientation flux
field is called the curvature tensor of the space and is written as R . The
notation Eq. (3.27) is compact: the curvature is a matrix of 2-forms that
measures the flow or change of the orientation of a frame through a 2-
surface into the dual of a second 2-surface:
R ab Rabcd Ec Ed . (3.28)
The components, the tensor R a bcd are antisymmetric in the last two
indices. The tensor is also antisymmetric in the first two indices. By Eq.
(3.27), the curvature tensor components are determined by the
orientation potentials.
Since we have an equation of action for the electromagnetic field in
terms of the field strength F , we expect a similar principle to hold for
the curvature. Of the many forms that have been found and are
Inertial Behaviors 97
S R ab * Ea Eb . (3.29)
In evaluating the least action, the full curvature tensor does not appear,
only two of its contracted forms, Rab Rdabc g cd and R g ab Rab . The
98 Geometry, Language and Strategy—Vol. 2
g bcTab;c 0 . (3.32)
Compare this with our general expressions for the stress tensor for
matter, Eq. (3.19). Here we have the covariance principle that in
describing rates of change, ordinary derivatives are replaced by covariant
derivatives Eq. (2.26).
We have the last major piece of decision process theory from the
consequence of Eq. (3.31). The unified applications of the principle of
least action to all of the fields that participate in the decision process
bring together the orientation flux fields and the inertial media fields.
The inertial media fields reflect both inertial and application forces. All
of these fields carry energy and momentum. All of these fields are real.
All of these fields obey the conservation laws of energy and momentum
Eq. (3.32). Our last result for this chapter will be to investigate useful
models for the inertial media field, which we do in the next section. First
however we summarize a few results for reference later in the book.
We consider some of the consequences of the general form of the
Einstein’s Eq. (3.31). In general, the space will not be flat relative to a
given set of orientation potentials. Energy and momentum create the
source for the orientation potentials, which themselves create a flux that
carries energy and momentum. So although we argue that the space is
locally flat, the existence of non-trivial fields changes how individual
charts are pasted together to describe a more complex global structure.
We have argued that this is no different from looking at the earth in
terms of flat charts locally, yet nevertheless the space is curved. The
curvature is a consequence of how the charts must be glued together; it is
a consequence of the atlas. We measure effects of local curvature when
we look at second order rates of change. For example, we envision this
Inertial Behaviors 99
a
process by looking at a small square with sides dx a and dy . We take a
vector V a and displace it first along the direction dx a and then along
dy a ; alternatively we go first along dy a and then along dx a . The
difference of the two calculations can be evaluated in terms of the
covariant derivatives:
V a
;bc V a;cb dxb dy c Ra dcbV d dxb dy c . (3.33)
We make the reasonable statement that the difference between the two
a
directions will be a set of numbers R bcd that provide the proportionality
constants multiplying the area of the square and proportional to the size
of the vector being translated. This is again a statement of Stokes’
theorem and is a restatement of the arguments that led to Eq. (2.41), so
that the numbers R a bcd are the components of the orientation flux tensor.
The theorem from geometry is that a necessary and sufficient condition
for the space to be flat is that the curvature components vanish
everywhere.
For detailed calculations, we start with the components of the
curvature tensor given in terms of the orientation potentials, Eq. (3.27):
contracting the tensor with the metric are the quantities that determine
the equations of motion from the action Eq. (3.29). The curvature tensor
R ab can be evaluated in terms of the orientation potentials. For the
special case of a holonomic coordinate basis, the orientation potentials
are determined by the metric using Eq. (2.33) and the reduced curvature
tensor is:
100 Geometry, Language and Strategy—Vol. 2
Rab a b ln g cab
c
ab
c
c ln g acd bdc . (3.36)
The decision process theory includes the active forces due to competition
and cooperation of players as well as constraint forces described by the
energy momentum tensor Tab . We will provisionally call the constraint
forces the inertial forces. They lead to inertia and mass. We have
suggested in Vol. 1 that such forces lead to organizational dynamics
with such forces behaving like a fluid in physical systems.
In the study of weather, the atmosphere is modeled as a fluid with
flow, energy density and pressure. One resultant behavior key to
atmospheric physics is that wind flows from high pressure to low
pressure. It means that energy flows away from the concentrations of
energy. In a purely fair system, we would expect the flow to be purely
isotropic: no particular direction is favored. When applied to
organizational dynamics, it is the condition that all parts of strategic
space are equally populated in the absence of the active competitive and
cooperative forces. Any concentration of decision making in one area
should dissipate. This is a principle of homogeneity. All the choices are
equal and if they are not equal they tend to revert to that average.
The organizing sources that we have in mind are many. Decisions
are not made in a vacuum but in a context. They are made within a
family, a community, a society, not to mention in a physical
environment. They must include both competition and cooperation. Any
combination of these might form the relevant system. Even if it were
possible to include all of these effects, most of the details generated
would not be of interest. We want to extract only the effects of the
constraints on the variables of interest. The effects are captured for the
Inertial Behaviors 101
g abV aV b 1 . (3.38)
In the co-moving basis this determines the time component of the vector
field in terms of g00 . From our previous work, the stress components
p 00 and p 0b are zero. The time component of the energy momentum
tensor is the energy density T00 V0V0 .
If the basis is not only co-moving but Minkowski, the energy density
is equal to the time component of the stress tensor. In this same basis,
there are no momentum components and the space components of the
energy momentum tensor are given by p ab . Therefore we call the latter
the stress tensor components and in a general frame impose the
conditions:
pab pba
. (3.39)
pabV b 0
102 Geometry, Language and Strategy—Vol. 2
We have defined a fairly general model for the inertial constraint forces.
The energy momentum is a symmetric tensor, which will in general have
real eigenvalues at each point, one of which will have a positive length
; the remaining will have negative values. The eigenvector
corresponding to the positive eigenvalue defines the vector field V a .
We can then look at the difference Tab VaVb that we define as
pab . It has no positive eigenvalue, but only negative eigenvalues that
are projected onto the space orthogonal to the flow by the projection
operator:
We recover the general form Eq. (3.37). We have a general form for the
inertial sources to be included in decision process theory. We define
pressure as the average stress, with n space components:
p 1 n h ab pab . (3.41)
We remain consistent with our unified principle of least action Eq. (3.31)
as long as our energy momentum tensor components are conserved
according to Eq. (3.32). We provide the result of the conservation laws:
DVb
h ab
p ab
h a c hb d p cd ;b
. (3.42)
d
V a ;a Va ;b p ab 0
d
We have the inertial and orientation flux fields expressed in Eq. (3.42).
The right-hand side represents the inertial fields. The left-hand side
includes the orientation flux fields through the geometry, along with the
flow of energy. The gradient of the stresses provides a force; the second
equation is the matter conservation law for the inertial matter density .
Let us return now to the ideas of organizational dynamics. In the
simplest case, we expect the inertial stress tensor to be isotropic and the
form to correspond to the elastic perfect fluid Eq. (3.14) written in
covariant form:
Inertial Behaviors 103
In this case, the stress tensor elements are equal to the pressure:
p ab phab . (3.44)
DV a
p h ab p;b
. (3.45)
d
V a;a p 0
d
The first equation is Euler’s equation and the second is the conservation
of inertial media, defining the inertial media density in terms of the
energy density and pressure:
d d
. (3.46)
p
4
This and the next section are advanced. Though their results set the context for our
theoretical foundation, they are not needed for engineering calculations of decisions.
104 Geometry, Language and Strategy—Vol. 2
Tdt 0 . (3.47)
The integral provides the historical definition of action5. Since the kinetic
energy is always a quadratic form in terms of the generalized
coordinates, T m jk q j qk , this variation principle can be written in
terms of a “path length” ds 2 m jk dq j dqk in the geometry of the
active coordinates:
2mT ds 0 . (3.48)
nds 0 . (3.49)
5
Goldstein [1959, p. 231] notes that this principle of least action is associated with the
name of Maupertuis, though the objective formulation of the principle is due to Lagrange
and Euler.
Inertial Behaviors 105
light bends when it enters water, because the index of refraction of water
is different from that of air. It provides the theoretical foundations for the
construction of optical equipment. Hamilton’s principle of least action
has the same mathematical form as geometric optics with an index of
refraction 2mT determined by the kinetic energy. If there are no
external forces (other than possible constraint forces), then Eq. (3.48)
shows that the path will be a geodesic (extremum) in the space of
generalized coordinates. Equation (3.47) shows that the path will be the
one that takes the least time. In order to relate these ideas to action we
have to go deeper into the equations of motion.
Both Hamilton and Lagrange hoped to write equations of motion that
contained only the essential ingredients required for the solution. The
constraints were one thing that could be eliminated. The other thing to
identify was the set of variables that were constant in time: these are
associated with symmetries of the problem. Even though it might be true
that motion along a surface depends on the position, the force might not
depend on that position. The earth is a sphere to a high degree of
approximation. The gravitational force does not depend on the position
on the surface, just on the height. Hamilton’s approach to this problem
was to consider a function that he could identify with the total energy of
the system, and that depends only on the momentum and position. So for
example if there were only a single independent coordinate, Hamilton
introduced the “Hamiltonian” that is related to the Lagrangian (Sec. 1.6):
H H
p , q . (3.51)
q p
L L
p , p . (3.52)
q q
S S S
p , Q , K H . (3.53)
q P t
Inertial Behaviors 107
K K
P , Q . (3.54)
Q P
S S
K Q, P, t H , q, t 0. (3.55)
q t
dS
L. (3.56)
dt
The equation that determines the transformed Hamiltonian is a partial
differential equation for the action.
Ordinarily, we look only at the paths of least action, and not on the
value that action takes on other paths. What Hamilton discovered was
that if you looked at the action as a function determined from the above
partial differential equation, you would see that it describes a
(mathematical) wave front determined by surfaces of constant action. To
see this, the equation is separated into an energy term and a term defined
by Hamilton’s “characteristic” function W to determine the wave
velocity u of the front:
S W Et . (3.57)
W
H qk , E. (3.58)
qk
108 Geometry, Language and Strategy—Vol. 2
dW Edt kW ds
ds E E . (3.59)
u
dt kW 2mT
ds
0. (3.60)
u
The result is general, holding in the presence of external forces.
This deep dive into physics illustrates that the principle of least action
may be an approximation. The concept of geometric optics, which
Hamilton showed describes mechanics, is known to be an approximation.
We know that if the wavelength of light is large compared to the optical
device the wave approach must be taken into account to get an accurate
description. Hamilton noted the same thing; if Eq. (3.55) had a small
term on the right-hand side, then the classical solutions would deviate
from Newtonian mechanics, and the paths would no longer be paths of
least action in terms of the classical variables. The result is related to
decision process theory for decisions in that we must also view the
principle of least action as approximate. We alluded to this problem in
Sec. 2.1 in our discussion of units of time. We argued that we must
consider times large compared to the characteristics of the process in
order to insure that a decision was in fact made.
We suggest that the way to set units of time is to identify a quantum
of action that describes that decision process. Knowing the scale set by
Inertial Behaviors 109
the quantum of action would set the units of measuring time and would
augment the natural units defined in Sec. 2.1. In the next section we
indicate how quantum effects are introduced in physics, which suggests
how one might extend our theoretical foundation in the region in which
the principle of least action breaks down.
2
i k k V . (3.61)
t 2m
The most important aspect of this equation is Planck’s constant, ,
which has dimensions of action and sets the scale distinguishing classical
behavior from quantum behavior. For decision process we suggest
introducing a similar constant.
The equation has other attributes worth noting. The potential energy
V specifies the forces acting on the electron. Typically it has the same
form as the Newtonian potential for a given type of force: Coulomb
attraction will be a potential whose strength is inversely proportional to
the distance; thus the gradient will give the inverse square law. The
electron is described by a continuous field called the wave function that
is a complex function of space and time and has a behavior that follows
our expectations for a continuous matter distribution. Recall that a
6
This and the previous section are advanced. Though their results set the context for our
theoretical foundation, they are not needed for engineering calculations of decisions.
110 Geometry, Language and Strategy—Vol. 2
iS
exp .
1
2
(3.62)
We show that the phase is the classical action in the limit that the
parameter is small. We do this by separating out the real and
imaginary contributions of the resultant equation. The real part provides
the flow of matter and represents its conservation:
k vk 0. (3.63)
t
This is precisely the same form we obtained for continuous matter
distributions Eq. (3.9). Even in the quantum domain, there is a concept of
a continuum distribution that obeys the laws of continuity. In making this
comparison, the velocity of the electron is defined in terms of the
momentum p k m v k k S . The second equation is determined by the
imaginary part:
Inertial Behaviors 111
S 1 2 12
k S k S V k k 2 .
1
(3.64)
t 2m 2m
In the limit that Planck’s constant goes to zero, this is the Hamilton
Jacobi’s Eq. (3.55) and demonstrates that the phase S is the classical
action. It also demonstrates that the paths will be those of least action,
whenever the left hand size is significantly greater than the right hand
size, which is proportional to 2 .
The action described by S corresponds to the classical “rigid” body
we might call an electron. The action of the wave however can be
associated with the electron viewed as a continuous field defined over
space, in the same way we have defined the electromagnetic field. We
can determine Schrödinger’s equation through a principle of least action
from the field perspective (Goldstein, 1959):
2
S k k V
8 m
2
* *
2
* * dxdydzdt . (3.65)
S q 1 2 mq 2 dt . (3.66)
ta
i
K qbtb ; qa ta exp S q . (3.67)
all paths
The expression depends only on the initial and final positions and times.
The sum can be carried out by considering a finite number of equally
spaced time intervals with the interval of time determined from the time
difference and the number of intervals t tb ta N 1 , and then
taking the limit N :
12
2 i tb ta im qb qa 2
K qb tb ; qa ta exp . (3.68)
m 2 tb ta
Inertial Behaviors 113
3.10 Outcomes
7
I am indebted to Mr. K. Kane for bringing this essay to my attention.
Inertial Behaviors 115
3.11 Exercises
(9) In a coordinate basis Eq. (2.33), use Eq. (3.34) to demonstrate the
following symmetry relationships for the curvature tensor
components, which by covariance must hold in any basis:
(10) In a coordinate basis Eq. (2.33), use Eq. (3.34) to demonstrate the
following symmetry relationship for the curvature tensor
components, which by covariance must hold in any basis:
and the cyclic symmetry condition Eq. (3.71), which follows from
the second Bianchi identity [Gockeler & Schucker, 1987] in a
general coordinate system, derive the symmetry relation:
(12) Using the full set of symmetries Eq. (3.70), show that we don’t get
the cyclic symmetry Eq. (3.71) but the cyclic tensor below is totally
antisymmetric in all indices:
(13) Use the results of the previous exercises to show that the total
number of independent curvature components in a space-time of D
dimensions is
1
2 1
2 D D 1 1 2 D D 1 1
D!
4! D 4 !
(14) Using the average pressure Eq. (3.41) demonstrate the following
form for the contracted curvature tensor, where n is the number of
spatial dimensions:
p p
1
Rab VaVb p pab hab p hab . (3.74)
n 1 n 1
118 Geometry, Language and Strategy—Vol. 2
(15) Derive the conservation laws Eq. (3.45) from the form of the energy
momentum tensor Eq. (3.43).
(16) Show that the conservation of the energy momentum tensor Eq.
(3.32) is a direct consequence of the conservation of the Einstein
tensor G ab Rab 1 2 Rg ab , which follows from the definitions of the
curvature tensor and its contractions:
g bcGab;c 0 . (3.75)
(17) In Sec. 6.3, we will provide arguments for adopting the conductivity
model, in which there is a charge current proportional to the electric
field in a frame in which the medium is at rest. This can be modeled
in a covariant way by adding a viscosity contribution to the energy
momentum tensor. With the following form, show that one obtains
the conductivity model, where the viscosity is the conductivity:
V ; p ,
DV
p 1
2
p; h n2n2 ; h ; 2 h V ; h
1 0. (3.77)
Persistent Behaviors
119
120 Geometry, Language and Strategy—Vol. 2
1
By active we mean both time and the active strategies.
Persistent Behaviors 121
that highlights the vorticity aspects of the theory and may be solved using
known current numerical techniques. It satisfies the centrally co-moving
hypothesis.
We use such numerical results in later chapters to illustrate how
decision process theory provides a common ground for discussion of
economic issues. The model allows us to extend, using an electrical
engineering analogy, behaviors that are like DC circuits, to behaviors
that are like AC circuits.
2
The idea in physics is that simple systems are integrable. More complex systems are
those with constraints, but if the constraint equations are each integrable, the remaining
variables will also be integrable. For decisions we envision that the active and inactive
variables are those that remain after all the constraints have been imposed. The
assumption is that the remaining variables will be integrable and correspond to variables
that are exact as defined in the following paragraphs.
Persistent Behaviors 123
literature, e.g. [Hawking & Ellis, 1973], to say that the spaces are
manifolds, which means we include in the definition all of the local
properties and charts, as well as a prescription for how the charts relate to
each other when they overlap. We then say that two manifolds are the
same if there is a mapping that takes one into the other and vice versa in
such a way that the manifold properties are the same: they have the same
topology and geometric structure. Such mappings are called
diffeomorphisms. Our linear transformations may or may not be
diffeomorphisms. Thus different coordinate bases may not represent
equivalent manifolds. Nevertheless it is useful to use general linear
transformations to explore different models. Often, different models have
a common embedded manifold that can be identified by a suitable linear
transformation. We find that to be the case with the co-moving
orthonormal coordinate basis defined below, Sec. 4.6.
For coordinates that are locally holonomic, we expect certain
mathematical properties. Each constant surface that corresponds to a
holonomic coordinate has the attributes of a potential field. Potential
fields generate a unique vector at each point on the surface that is normal
to the surface. In general, parallel surfaces generate the same vector
fields. To gain more insight in potential fields, you may recall potentials
whose gradient a determines the properties of the vector field that
is normal to the surface of constant potential. Surfaces of constant
potential are used in engineering and physics. For example the potential
surfaces within a capacitor of arbitrary shape determine the electric field.
The gravitational potential in Newtonian mechanics defines surfaces that
map the gravitational forces. The potentials provide a convenient method
to discuss physical effects associated with vector fields. The vector fields
are typically those that more directly determine dynamics, not the
potentials. Moreover, not every vector field that determines physical
behaviors can be written as the gradient of a potential.
Although locally holonomic coordinate bases provide an intuitive
meaning for measurements, there are also advantages of looking at
global coordinate systems that are not holonomic. They may better
illuminate curvature effects. As a simple example, recall that we live on a
rotating sphere whose holonomic coordinate system would be fixed.
However, a rotating coordinate system has advantages because it
124 Geometry, Language and Strategy—Vol. 2
U E . (4.2)
U ; U ; . (4.3)
You might think that the most convenient frame of reference would
always be a coordinate or holonomic basis. Such bases lead to
differential equations that have been well-studied and provide necessary
information about existence and properties of solutions. However, such
local bases don’t necessarily provide the best global view, especially if
there are symmetries. For example, in decision process theory,
persistency determines the property of the distance measure, Eq. (2.42):
126 Geometry, Language and Strategy—Vol. 2
U a dx a
. (4.6)
U j d j Aaj dx a
det jk
. (4.7)
g det g ab
Noting that these two sets are orthogonal to each other, the determinant
of the full metric for the space is the product g , which can be either
positive or negative. We indicate the absolute value of a quantity in the
usual way, for example is the absolute value or magnitude of the
inactive determinant.
In the normal-form coordinate basis, the payoff matrix is determined
by the differential of Eq. (4.6) that is manifestly covariant:
dU j F j 1 2 F j ab U a Ub
. (4.8)
F j ab Abj;a Aaj;b
Fj ab jk g ac g bd Fcdk . (4.10)
This highlights that there are no non-zero metric components that mix
active and inactive strategies in the normal-form coordinate basis. As a
further physical aside, we note that the Coriolis Effect and centripetal
acceleration on the earth in which the inactive coordinate is time,
respectively would be the “payoff” tensor for the Coriolis Effect and the
gradient of the inactive metric for the centripetal acceleration, identified
as gravitation.
appropriate to flat space not curved space. For this reason, we go against
this view and allow gauge arbitrariness in the choice of pure strategies
and time, which allows for dynamics that reflect a topological and
geometric structure that will be constrained by its success in matching
observational data. The gauge freedom is analogous to the freedom of
using “charts” for navigation: each chart treats the earth as being flat or
Euclidean. Because they describe an object that is round, they don’t
exactly match up. The mapping that matches them is the gauge
transformation. The gauge freedom is that at any point, we are free but
not required to consider the space (our chart) to be flat. Though there is
gauge freedom, it is not a freedom that can be removed as an attribute of
the theory: topological objects with curvature such as the sphere require
multiple charts. We assert that decision processes generate curvature and
require theories with this gauge property.
To solve the local field equations, we can pick any gauge. The
solution will be valid in some region around our initial conditions. Our
choice is the harmonic gauge described in the literature, [Wald, 1984].
We adapt an argument from that literature to apply harmonic
coordinates to the normal-form coordinate basis. We start with the field
Eq. (4.1) for decision process theory that is the consequence of the
principle of least action. We write these equations in the normal-form
coordinate basis. The argument assumes we have a set of coordinates,
x a , which are scalar functions of y a in another basis in which the metric
g ab y is known in terms of these coordinates y a .
We define the active coordinates x a in the normal-form coordinate
basis using the following differential equation, where the Greek indices
, span both the active and inactive dimensions of the normal-form
coordinates:
g x a ; 0 . (4.11)
1
g
b
g g bc c x a 0 . (4.12)
1
g
b
g g bc c x a 0
1
g
b
g g ba 0 . (4.13)
1
2 g bd b g ac jk F k cd Tj a . (4.14)
g ab b Aaj 0 . (4.15)
2 . (4.18)
This is Poisson’s equation and reflects that matter attracts other matter.
In the dimensions of 3+1 space-time, this is an inverse square attraction.
In a strategic space-time with n spatial dimensions, this force law is
generalized.
More generally, we can see that if time is an isometry as defined in
the next section, there will be two effects corresponding to the Coriolis
Effect and centripetal acceleration. In general relativity the “centripetal
acceleration” is determined by the time component of the metric and the
corresponding “Coriolis Effect” is gravitomagnetism [Ryder, 2009, p.
180ff] and can be measured with gyroscopes and clocks by means of
satellites. It is interesting that there is a connection between payoffs,
magnetism and Coriolis. The source of the connection is the underlying
symmetry or hidden variable. To make the connection complete for
electromagnetism, we use [Kaluza, 1921] and [Klein, 1956] for the
Kaluza-Klein theory and the analogy to their proposed hidden fifth
dimension.
It is worth emphasizing that these effects do not arise from a physics
analogy but from the mathematical structure of the proposed decision
process theory and general principles of differential geometry. They all
share the same underlying mathematics. We find it significant that the
gradients of the orientation potentials provide a basis for an attraction
that exists independently of the payoffs. In whatever frame we work, an
important effect is that of acceleration. This is the symptom of the force
134 Geometry, Language and Strategy—Vol. 2
Tabinactive 1
2 jk jk ;ab 1 4 a jk b jk 1 2 g ab jk g cd jk ;cd
1 8 g ab g cd mn jk mn ;c jk ;d 3g cd jk
;c jk ;d . (4.20)
strong gravitational sources. The existence of this field was not foreseen
by our initial analysis in Sec. 1.7. It is not uncommon however in
theories based on those of Kaluza [1921] and Klein [1956].
The last set of field equations obeys the massless wave equation,
which determines the cooperative payoffs in terms of sources:
1
2 g ab a jl b lk T j k 1 4 F j ab Fk ab . (4.21)
Again, the first term arises from the organizational fields restricted to
inactive components. These terms are for now arbitrary, since they don’t
have a clear analogy from physical models other than the perfect fluid.
This may not be a good guide. The second term is non-zero only if there
is a non-zero overlap between players’ payoffs: there must be common
cause.
In the absence of all sources, Eq. (4.21) is a massless wave equation
(a Klein-Gordon field in physics texts). These fields (along with other
possible fields) in Einstein’s theory could describe gravitational waves.
Thus even in empty space there may be influences that propagate with
the speed of light.
In general, the active and inactive field equations determine metric
potentials that are not Minkowski. Thus there is no a priori justification
for requiring the active and inactive metrics to be flat. Even in spaces
with no organizational sources, the payoffs contribute as sources and
provide the source for curvature. Thus given the equations, the next
logical step are to apply the decision process theory using these
equations for decision problems, solve the equations for such problems
(numerically), analyze the results and based on the results, and refine the
theory. In the process we expect to grow a quantitative understanding of
the decision process.
With suitable computing power, we have sufficient information to
apply the decision process theory to a wide variety of problems. As
anyone in engineering knows however, it is not just the theoretical ability
to solve problems that is of singular importance, it is also the practical
ability to find sets of problems that can be solved and from them, the
knowledge that gives the ability to build structures (bridges, circuits,
etc.) that do desired things.
136 Geometry, Language and Strategy—Vol. 2
K ; K ; 0 . (4.22)
K g K g ,
K ; g ,
K ; K ; g g 2 ,
K ; K ; g g g g g ,
K ; K ; g 0 . (4.23)
Therefore given the condition Eq. (4.22), there is a frame in which the
metric is independent of the dimension .
Conversely if there is a frame in which the metric is independent of
the dimension , we pick K to be the vector field that is unity along
that direction and zero otherwise. The above argument again holds and
we deduce Eq. (4.22). Since it is a covariant relation that holds in one
frame, it therefore holds in all frames. This relationship is attributed to
Killing.
If there are two isometries, there will be two Killing vectors K and
L . From two vector fields we create a new vector field called their
commutator or Lie product Eq. (4.4):
This has far reaching consequences. The set of all Killing vectors, which
includes their commutators, form a Lie Algebra, with the Lie product.
The isometries of the theory therefore generate a local symmetry group
(i.e. a symmetry group at each point in space), which has the same
structure at every point. This provides substance to the notion that
isometries are persistent and reflect the local symmetry group.
Isometries reflect properties of what we term an internal group,
which gives substance to invariance and persistency associated with the
inactive strategies. In particular, we see that the concept of independent
players has in fact been framed as a group theoretical statement that there
138 Geometry, Language and Strategy—Vol. 2
Field equations of the type Eq. (4.1) and their solutions have been
extensively studied, for example [Wald, 1984, p. 252ff], and [Hawking
& Ellis, 1973]. The essential point for us is that solutions to these field
equations exist and have reasonable properties if the metric, orientation
potentials, matter fields and gradients of these are specified initially on a
surface “orthogonal to time”. The field equations then determine the
evolution of the metric, orientation potentials, matter fields and flows on
all surfaces that are “later”, Sec. 6.6. Our goal is to use these solutions to
decision process theory to better understand the decision process. We
believe this understanding must be based on a quantitative understanding
of the consequences of these field equations. To this end it is natural for
us to adopt and adapt the approach to solving equations of this type that
has proven useful in the past (Cf. Sec. 4.3).
For these reasons, we look at the equations in a frame that is not
necessarily diffeomorphic to the normal-form coordinate basis, but is
one that is obtained from a linear transformation. The frame we want is
one in which one direction is along the energy flow vector that arises
naturally in considering the inertial fields Eq. (3.37). The energy flow
direction is a natural “time” direction. In this frame we are co-moving so
the vector has only a time component and the remaining directions are
orthogonal unit vectors with respect to the metric. We further look for a
system in which this metric is globally Minkowski. Because the metric is
constant, in general the coordinates will not be holonomic.
We call this frame the co-moving orthonormal coordinate basis or if
there is no ambiguity, the co-moving basis. Note that in general, it may
not be possible to physically transform to this frame simultaneously at
Persistent Behaviors 139
3
Later, we introduce the player fixed frame model in which we make further distinctions
between some of these orthonormal coordinates, which we associate with “proper” active
strategies and some we associate with “proper” inactive strategies (players).
140 Geometry, Language and Strategy—Vol. 2
X E X
X E X
. (4.26)
E E
E E
dU j 1 2 F j ab Ua Ub . (4.28)
U j E j Eα
. (4.29)
dU j E j | E α E β
Comparing the two expressions we find that each player field E j acts as
the vector potential for the payoff in the co-moving basis:
Persistent Behaviors 141
F j ab E a E b E j | E j | f j
. (4.30)
We can use Eq. (4.31) to determine the orientation potential in one frame
from the covariant derivative.
An alternate form is the inverse:
E | E E E E . (4.32)
Here we consider the (inverse) transformation matrix E to be a vector
field (timelike and spacelike) in the co-moving orthonormal coordinate
basis labeled by the normal-form dimensions . As noted, these
expressions reflect the transformation properties of the acceleration
potentials:
E E E E E E . (4.33)
E E E E E E . (4.34)
142 Geometry, Language and Strategy—Vol. 2
g ab E a ;b jk E j ; k
. (4.35)
1
g
a
g g ab E b
g ab a E b . (4.36)
a E j 0
. (4.37)
a Eb b Ea
g ab a b y . (4.38)
m E a | a bc g bc a kl kl
1
g
b
g g ab 0
. (4.39)
m E j
| j
bc g
bc j
kl 0
kl
E a | E a | . (4.40)
We have a general decision process theory that leads to field Eq. (4.1)
that allow us to investigate decision processes using the scientific method
to expand our understanding and our approach. We have players
identified in a persistent way with inactive strategies along with the
decisions they make that are active. We treat time as an active
dimension. We have argued that there will always be a timelike unit
vector that describes the flow of energy as well as a metric tensor such
that these vectors and tensors depend only on the active strategies
available to each agent. Decisions occur in this space-time x a that ranges
over time and the active strategies. In this normal-form coordinate basis,
none of the tensors depend on the inactive space components . In this
j
frame, the metric components between the active and inactive space are
zero. In the last section, we extended this general discussion to the co-
moving orthonormal coordinate basis.
The vector Va in Eq. (3.37) representing the flow of energy occupies
a central role in the theory. Some of the orientation potentials related to
the flow in the co-moving frame have large scale geometric significance
that makes it easier to understand the anticipated behaviors. We take one
of the orthonormal coordinate basis vectors to be the flow of energy
E V . We indicate this dimension by the Greek letter .
Persistent Behaviors 145
E ; V ; E E E E q E V
. (4.41)
q
0
DV
V ; V q E . (4.42)
Because the orientation potentials are antisymmetric in the first two
indices, these same parameters determine , so that an expansion of a
vector transverse to the flow is:
DF E DE
qV E . (4.44)
146 Geometry, Language and Strategy—Vol. 2
DF X DX DV DV
X V V X . (4.45)
This rate of change, which is defined in terms of the covariant derivative,
has the following useful properties that help us understand the geometric
meaning of the orientation potentials. Any pair of vector fields that have
zero Fermi derivative will maintain lengths and angle. The Fermi
derivative measures the amount of rotation of the basis. Using the Fermi
derivative, we deduce that the significance of the antisymmetric
orientation potential coefficients is that they specify
the frame rotation along the path.
We note that transformation to the flow vector has both active and
inactive components V a and V j , which is a transformation to a vector
field whose 1-form does not in general vanish. In other words, the flow
corresponds in general to a 1-form that is not exact. In Einstein’s theory
of relativity, this is stated as the twin paradox: twins traveling along
different paths may age differently as a result of their paths seeing
different accelerations: the paths are not inertial. This is true in this
theory as well. This makes sense because we expect the flow to reflect
the influence of sources that generate acceleration. We address the
question about whether the transverse components of the orthonormal set
needs to be exact in the next section.
E j 0
. (4.46)
a Eb b Ea
dE υ
Eυ Eο E υ Eυ
E Eα Eυ Eα
Eα Eβ 0 . (4.47)
We have expanded the wedge product using the notation from the
previous section, where we separate the behaviors along the energy flow
direction, the proper-active strategy directions and the remaining proper-
inactive strategy directions.
We convert the consequence of exactness to conditions on the
orientation potentials. We write these orientation potentials using the
decompositions Eqs. (4.41) and (4.43):
0
0
. (4.48)
0
With this notation the five distinct terms give five sets of conditions,
which can be simplified:
0
0
. (4.49)
0
These conditions along with the additional conditions Eq. (4.46) on the
frame implies that there is a coordinate potential y that is a scalar
function of the harmonic coordinates and Ea a y .
We add the following additional assumptions to the player fixed
frame model that we have found by trial and error to be consistent and
Persistent Behaviors 149
q 0
0 . (4.50)
0
E a e E a . (4.52)
Substituting this into the expression Eq. (4.51), we get a projection of the
flow onto the inactive directions:
Ea E j e E j 0 . (4.53)
Because at least one component of the timelike flow is not zero, the
implication is:
E j e E j . (4.54)
E a a e . (4.55)
Persistent Behaviors 151
We call the variation along the flow, the variation along the proper-
time coordinate. We also see that the operator j is zero based on Eq.
(4.54), as it should be on the space of functions of interest:
j E j E a E j e E a a 0 . (4.56)
g ab a b y q
. (4.57)
g ab a E b 0
Here we make use of the assumptions above concerning the values of the
orientation potentials. We apply Eq. (4.52) to the gauge condition Eq.
(4.36), and get:
g ab b e Ea 0
e g ab b Ea g ab Ea b e 0
. (4.59)
e g ab b Ea e 0
e 0
The result follows directly from the second line in Eq. (4.57).
Since the coordinates x a are exact, this coordinate vector E a in a
new coordinate system satisfies Eq. (4.40): E a | E a | . For the player
fixed frame model, we apply these conditions and find that there are two
distinct classes of equations:
E a E a q E a 2 e E a
. (4.60)
E a E a 2 2 e E a
152 Geometry, Language and Strategy—Vol. 2
This shows that in this basis, the vector fields are determined. The first
equation determines the “curl” of the vector field in terms of the
acceleration and what we show below is the electric field in the co-
moving frame. The second equation determines the “curl” in terms of
what we show later to be the payoff matrix in the co-moving frame.
We have determined equations for all the vector fields Ea associated
with the active components. We now turn to the vector fields E j
associated with the inactive components. In the normal-form coordinate
basis, the Killing vectors are the inactive vectors that lie along each of
the coordinate directions. So for example the Killing vector field K j
K j E K j E j .
(4.61)
E j | E j | 0
E j | E j | 0
E j | E j | 0
. (4.62)
E j | E j | 0
E j | E j | 0
E j | E j | 0
We summarize the result for each case in order, leaving to the exercises
at the end of the chapter the derivations.
(1) We start with the first case and write out the covariant derivative in
terms of the orientation potentials with the player fixed frame model
that the transformations depend only on the proper-time coordinate
and proper-active strategies:
Persistent Behaviors 153
e E j e E j 0 . (4.63)
(2) The second case provides the proper-time variation of the proper-
active strategy of the Killing vector field components:
(3) The third case provides the time variation of the proper-inactive
strategies of the Killing vector field components to be zero and
shows that Eq. (4.63) is identically satisfied:
E j 0 . (4.65)
(4) The fourth case has no time derivatives and is identically satisfied
based on Eq. (4.46):
E j E j 0 . (4.66)
(5) The fifth equation gives the spatial dependence of the proper-
inactive components:
(6) The sixth case is the time derivative of the component of the Killing
vectors along the energy flow:
E j q E j q E j 0 . (4.68)
(7) We can use the above case with Eq. (4.54) to determine the
“charge” gradient:
From the harmonic gauge conditions Eq. (4.39) and exactness conditions
a a
Eq. (4.60) we obtain the equations that determine E E :
1 e e E
a
E a q E a
. (4.71)
E a E a q E a 2 e E a
g ab 1 e e E a E b E a E b . (4.72)
The active metric may depend on proper-time. Even in the case that it is
independent of proper-time, it does depend on the acceleration and so
there will be effects due to acceleration in the solution.
We have pointed out previously, Eq. (4.30), that the co-moving
payoff is determined once the transformations are known. We can also
proceed as follows to determine the payoff. The payoff matrix is an
acceleration-potential in the normal-form coordinate basis, Eq. (2.48) as
seen for example in jab . Since the orientation potentials are related, we
use Eq. (4.32) to determine the payoff fields:
1
2 F j ab Ea Eb E j Ea Eb E j
q E j e E j Ea Eb Ea Eb
E j e E j Ea Eb Ea Eb . (4.73)
insight into the equations in the co-moving frame. From this form we
determine also the various co-moving payoff components:
f j
0,
f j
0,
f j
e f j
,
f j
2 q E j E j e E j 1 E k Ek
2 e E j
1 E k
Ek
,
f j
2 E j 2 E j . (4.74)
4.10 Outcomes
different space, time and phenomena. The student should learn that the
large scale behavior of the theory differs from the local behaviors
because of the possibility of acceleration: i.e. dynamic changes of
strategic choices. We illustrate the possible large scale behaviors by
introducing the player fixed frame model in both the normal-form
coordinate basis and the co-moving coordinate basis. This model refines
our notion of large scale persistency in a way that is consistent with and
follows from the local principles of the theory.
The attainment of the outcomes is facilitated by doing the exercises at
the end of this chapter. To further help the reader, we list more detailed
outcomes from this chapter by section.
From Sec. 4.1, the student will have learned that decision process
events at each point in time can be described locally in a coordinate
(holonomic) basis in which each coordinate is a surface of constant
potential for a potential field whose normal provides the coordinate
direction. This underlies the assumption that we can specify the
distance between any two decision process events at separate points in
time.
In Sec. 4.2, the student learns that in the normal-form coordinate
basis, gauge invariant results are directly obtained. This provides an
example of a “global” basis that has advantages over the “local”
coordinate basis. Though not holonomic, the basis is effectively
holonomic when transformations are restricted to the active subspace.
In Sec. 4.3, we frame the field equations in the normal-form
coordinate basis using the harmonic gauge that removes the gauge
degrees of freedom. These field equations form the basis for the
quantitative predictions of decision process theory.
In Sec. 4.4, the field equations that result are derived, supported by
results from the exercises. These equations provide the student a
guide to the behaviors expected in the theory and can observe the
similarities and differences between decision process theory and
physical theories.
In Sec. 4.5 the student will have learned that the Killing relations
from differential geometry provide a covariant articulation of
isometry, which is a common group structure at each point of space
Persistent Behaviors 157
and time. In decision process theory, the set of players form the basis
of this symmetry and so are persistent attributes of any solution.
In Sec. 4.6, the student will be introduced to the co-moving
orthonormal coordinate basis. The student should be able to
transform between this basis and the normal-form coordinate basis
and understand how to express the covariant derivatives, exactness
conditions and harmonic gauge conditions in each basis.
In Sec. 4.7 as preparation for creating models in the co-moving basis,
a geometric decomposition of the orientation potentials is given.
Certain of the orientation potentials describe how the frame rotates
when moving along particular directions.
In Sec. 4.8 the player fixed frame model is defined and the
consequence of exactness developed.
In Sec. 4.9, the consequences of persistency are developed, along with
the further constraints of the commutation constraints and harmonic
gauge constraints. The student should understand the basis of this
class of models and the defining equations of this model in the theory.
4.11 Exercises
Ua , Ub Ua , U b j j F j ab j ,
Eμ , E ν E .
(4.75)
(5) Using Eq. (4.75) show that the Lie product of the coordinate
directions vanishes if and only if the orientation potentials are
symmetric in the last two indices. Using the inverse transformation
below, what do you conclude about the Lie product of the
transformed coordinate directions:
Εα , Eβ E .
(4.76)
Rabjk 1
4 F g F jl km F g F
jl km
l
ac
cd m
db
l
bc
cd m
da
ml a jm b kl ml b jm a kl . (4.77)
R ij ab 1
4 lk a jk b il 1 4 lk a il b jk
(9) Using R j abc from Eq. (4.79) and R j abc R j bca R j cab 0 from Eq.
(3.71), show Eq. (1.20), which implies that the payoff matrix is
derivable from a potential.
(10) Using the explicit forms from the previous exercises for the
curvature tensor in the normal-form coordinate basis, show that the
Einstein Eq. (4.1) yields the generalization of Maxwell Eq. (4.14) in
the harmonic gauge. In a general gauge, show that the result is:
1
2
1
jk g bc Fabk ;c
T ja . (4.80)
(11) Using the harmonic gauge and the additional “covariant” gauge
condition Eq. (4.15) develop the wave equations for the player
vector potentials Eq. (4.16) from Eq. (4.80).
(12) Using the explicit forms from the previous exercises for the
curvature tensor in the normal-form coordinate basis, show that the
Einstein’s Eq. (4.1) yields the field Eq. (4.17) for the active
strategies.
(13) Using the explicit forms from the previous exercises for the
curvature tensor in the normal-form coordinate basis, show that the
160 Geometry, Language and Strategy—Vol. 2
Einstein Eq. (4.1) yields the field Eq. (4.21) for the inactive
strategies.
(14) Write and discuss the field equations in the absence of a matter field
and with inactive and active metric components equal to their
Minkowski (flat) values.
(15) Show that Poisson’s law, Eq. (4.18), results as a static, weak field
limit of Eq. (4.17).
(16) Demonstrate that the commutator (Lie product) of two Killing
vectors is again a Killing vector.
(17) Show that the results Eq. (4.49) follow from Eq. (4.48).
(18) Using the expression for the Lie product Eq. (4.76), for the player
fixed frame model and using the notation for that model from Sec.
4.8, show that the Lie product for co-moving basis proper-active
strategies are:
. (4.81)
Ευ , Eυ 2 E j 2 E j
j
(19) Using the expression for the Lie product Eq. (4.76), for the player
fixed frame model and using the notation for that model from Sec.
4.8, show that the Lie products for co-moving basis proper-active
strategies with the co-moving flow vector are:
Εο , Eυ q 2 e E a
a
. (4.82)
Εο , Eυ q E j 2 E j
j
(20) Using the expression for the Lie product Eq. (4.76), for the player
fixed frame model and using the notation for that model from Sec.
4.8, show that the Lie products for co-moving basis proper-active
strategies with the proper-inactive strategies are:
Ε υ , Eα e E a 2 E a
a
. (4.83)
Ε υ , Eα E j 2 E j
j
Persistent Behaviors 161
(21) Using the expression for the Lie product Eq. (4.76), for the player
fixed frame model and using the notation for that model from Sec.
4.8, show that the Lie products for co-moving basis proper-inactive
strategies and of these strategies with the co-moving flow are:
Εα , Eβ Εο , Eβ 0 . (4.84)
(22) Deduce the seven cases that are provided that follow from
Eq. (4.62).
(23) Derive the following for the gradients of the inactive metric:
a p 1
2 a jkV jV k 1 2 h jk Fabk bV j . (4.86)
p p
162 Geometry, Language and Strategy—Vol. 2
h jkV m
g
b aV k 1 2 ki a ilV l Vm g ab g 0. (4.87)
V aa
p
1
g
a
g V a 12
p
aV j h jk aV k . (4.88)
In this case the left hand side is zero if the flows are zero, whereas
the right hand side is not zero since the shear is not zero. What does
this imply? Can these equations be solved in a consistent fashion?
Chapter 5
163
164 Geometry, Language and Strategy—Vol. 2
The only step missing is the study of transient effects, which is also
illuminated by this approach.
The simplest situation for us however is not strictly analogous to an
electric circuit; however it does have a lot in common with transmission
lines and traveling waves. In both cases it makes sense to work in a
frame of reference in which the behavior is observed to be steady state. If
we travel along with the wave, then there will be no observed flow. In
this chapter we consider the case of decision processes which have that
property: it is possible to find a co-moving frame of reference in which
all of the dynamic attributes of the problem are steady state. Technically,
this means it should be possible to identify a frame of reference that is
co-moving, holonomic and one in which time and the inactive player
strategies are persistent and mutually commuting. We could then use the
principle of least action to provide us with the equations of motion. With
appropriate numerical techniques we could then solve these equations.
We then use these solutions to obtain the normal-form coordinate basis
solutions using the harmonic wave equation and the initial conditions.
There is a practical barrier however. There are a large number of
equations, which represent concepts that we have not explored. The
purpose of this chapter is to carry out this exploration and come up with
a set of equations that we can both solve and understand. To achieve that
goal, we find it useful to take a bottom-up approach rather than a top-
down approach. We start with a specific class of models that leads to our
goal, the player fixed frame models. We write the consequences of each
of the equations of motion and see what distinctions are implied. This is
a challenging task because it demands a fair amount of mathematical
juggling. To help the interested reader follow the thread, much of the
work is summarized in tables and the details are left as exercises. Toward
the end of the chapter we return to the goal stated here and demonstrate
that we have indeed solved the class of models of interest: models that
have steady state wave behaviors.
The detailed plan of the chapter is as follows. We first rewrite the
field equations in the co-moving orthonormal coordinate basis, or co-
moving basis for short. We then review the frame equations in this co-
moving basis, followed by the field equations making the assumptions of
the player fixed frame model. The basic results of the chapter are
Deconstructing the Theory 165
summarized in Tables 5.1, 5.4, 5.5 and 5.6. The expressions for the
frame are provided in Sec. 5.1. In Sec. 5.2 we show how we generated
the field equations in the co-moving frame and in Sec. 5.3, we simplify
the expressions using the player fixed frame model. To interpret these
results for decision process theory, in Sec. 5.4 we articulate the
distinctions that result from an examination of this model.
Before diving into the equations and the model, let’s recall how we got to
the equations. We have emphasized the importance of having a common
ground for the discussion of decision processes based on a theoretical
and scientific framework. The framework should be based on generally
agreed principles. We suggest the common ground be based on the
mathematics and empirical basis of physical theories as well as the
foundations of game theory and related works in economics. In decision
process theory, we take a unified and consistent view, which has led us
to replacing static theories and equilibrium frameworks with the physical
principle of least action articulated as the field equations, Eq. (4.1). We
have demonstrated that this mathematical expression, though compact, in
addition to providing the self-consistent framework summarizes a great
deal of information that we believe is commonly understood to be part of
the decision process. In Sec. 4.4, using the normal-form coordinate
basis, we expanded the field equations into three sets of equations and
sketched the information that might be derived from each set.
To make further progress in extracting predictions from decision
process theory, we consider models that have attributes in common with
static models; this is in the same sense that DC circuits provide insight
into AC circuits. We can use the same mathematics as long as we extend
the interpretation of what the mathematics represents. One main
departure is that we go from a static equilibrium or point to a
consideration of steady state behavior that extends throughout the
strategic space.
Our bottom-up approach is to proceed with the program set out in
Sec. 4.9 to find solvable models based on simplifications that are
166 Geometry, Language and Strategy—Vol. 2
suggest here, it is not clear what aspects can be usefully carried over. It is
also worth noting that discussions in other disciplines assume the reader
already understands a good deal, so there is some advantage in creating a
self-contained discussion of the solutions.
It is important therefore to specify a program in our domain that
identifies what is relatively simple from what is relatively complex. We
start that project. Our goal is at least to get to the point that we can see
that it is possible to have a complete dynamical numerical solution for
problems in which there are any number of strategies and any number of
players. We should be able to carry out this program for one, two, three
or four active strategies. As numerical techniques improve, it should be
possible to extend to a larger number of strategies: such approaches have
been used in other fields of study.
The key idea behind the player fixed frame model from Sec. 4.8 is that
in the co-moving frame of reference, the active strategy variables are
exact, which imposes dynamic constraints. We made additional dynamic
assumptions Eq. (4.50) to those of exactness, Eqs. (4.46) and (4.49). We
did not motivate those assumptions, a task we take up here:
q 0
0 . (5.1)
0
The first equation makes the plausible statement that the energy flow has
no acceleration along the proper-inactive directions. The player exercises
some persistency in his or her decision process.
In the second and third set of equations the interpretation is that the
co-moving frame has a fixed orientation with respect to the players and
that the only player bonding will be along the active strategy directions
and will be captured by . In particular, all compression or expansion
coefficients such as and the symmetric part of are zero.
Along the flow direction, the frame rotation is zero and so the Fermi
derivative of the frame rotation also vanishes in Eq. (4.44).
168 Geometry, Language and Strategy—Vol. 2
DF X DX DV DV
X V V X
. (5.2)
DX
X ; E
DF E
E
. (5.3)
DFV
0
Not only is the Fermi derivative of the flow zero along the direction of
motion, it is zero along any of the orthonormal coordinates. The Fermi
derivative of any frame component along any other frame direction is a
rotation set by the orientation potential restricted to the transverse space.
We return now to our notation from the last chapter and state the
consequence of our model assumptions Eq. (5.1) for the variation of
E j :
Deconstructing the Theory 169
DF E j
E j E j 0
. (5.4)
DF E j
E j E j 0
Based on the player fixed frame model, the orthonormal vectors for the
inactive space (the Killing vectors) remain fixed when moved along any
of the transverse directions. In particular we see the value of having
0 and 0 .
We have already commented about the frame rotation being zero
along the flow direction. Our assumptions amount to the requirement that
the frames do not rotate along the flow or along any of the transverse
orthonormal axes. Our assumptions can be summarized as saying that
with this hypothesis, the co-moving orthonormal coordinate basis is a
player fixed frame model, justifying its name. The orientation of the
proper player does not change. This class of models enhances the notion
of player persistency and extends the notion of persistency to be an
attribute that applies to the large scale structure. We do expect to see
some aspects of game theory in our solutions. We will also see
significant differences. In particular we see strong vorticity effects
evidenced in the vorticity tensor components .
E j
e E 1 E
j k
Ek .
Deconstructing the Theory 171
E j E j e
e
.
e 0
e q e 2 2 e e e
R α ο dω α ο ω α β ω β ο
R α β dω α β ω α ο ω ο β ω α γ ω γ β . (5.5)
, ,
For the appropriate orientation potentials we use Eqs. (4.41) and (4.43):
q
. (5.6)
,
The orientation potential 1-forms that result from these definitions are:
ω α ο V Eβ q V Eβ
. (5.7)
ω α β V E γ V E γ
dV Eβ E γ q Eβ V
dEα Eβ E γ V Eβ . (5.8)
, ,
Based on the symmetry condition R R , Eq. (3.70), we get our
first result, the time evolution of the vorticity:
The double bar notation for the components of the covariant derivative is
a short hand for the terms that involve only the transverse components:
We have used Eq. (5.9) to eliminate the time dependence of the vorticity.
We return to Eq. (5.11) later, though we note that this is a useful result in
its own right.
We obtain a field equation by contracting the tidal force using Eq.
(3.74) with n space components that total the number of active and
inactive strategies. We obtain an expression in which only the energy
density and average pressure appear:
p
R R p . (5.12)
n 1
We combine this equation with the time evolution of the expansion to get
the evolution of the volume compression coefficient m :
174 Geometry, Language and Strategy—Vol. 2
q |
p
2 p . (5.13)
n 1
Not all equations that we get are time evolution equations. The
possibility of field equations being independent of time exists with
Maxwell’s equations. For example, Coulomb’s law provides a set of
equations that are independent of time, with only spatial partial
derivatives. For equations of the Einstein type, such relations are named
after Codacci. In a space of n space dimensions and one time
dimension, there is a theorem that there will be n 1 Codacci equations.
We are able to identify n of these equations here.
The mixed component R is p , which is zero. This contracted
curvature tensor is:
We use this with Eq. (5.67) from the exercises at the end of this chapter
to obtain the n Codacci equations, [Hawking & Ellis, 1973, Eq. (7.17)]:
These constraints are on the surface orthogonal to the flow and contain
no time derivatives. The constraints are only spatial partial differential
equations that must be satisfied on the surface transverse to the time
flow. Since there are n 1 Codacci relationships, we need to find one
more.
Deconstructing the Theory 175
The remaining set of equations (lines 3-5 of Table 5.2) are computed
from the transverse components of the orientation potential 1-forms ωα β ,
Eq. (5.7). From Ex. 8 at the end of the chapter, the differential 1-form
Eq. (5.68) contains no proper-time partial derivatives of the frame
rotation , suggesting that it is set by other constraints. It does not
appear in the Codacci constraints Eq. (5.15), suggesting that there are no
time evolution equations for the frame rotations . Indeed, we don’t
expect evolution equations for q or , since their proper-time
derivatives don’t occur in the curvature tensor components, though we
may expect contributions from q e q .
3
2 D 1 D 2
Gauge choice 1
(e.g. exactness)
6
Field Eq. (5.17) 1
2 D D 1
3
2 D 1 D 2
Symmetry R R 1
Eq. (5.9)
9
Symmetry R R 1
2 D 1 D 2
2
Eq. (5.71)
24
2 D D 1
Total equations Orientation potentials 1 2
of the vorticity, Eq. (5.9) and the time evolution of the transverse
orientation potentials , Eq. (5.71). We have yet to compute the
evolution equations for the expansion parameters in order to have a
complete set of evolution equations of the orientation potentials
expressed in terms of the other potentials and the sources.
We obtain the complete equation for the expansion parameters by
starting with the contracted curvature tensor in terms of the sources, Eq.
(3.74):
p
R R R p ph h . (5.16)
n 1
The result is based on the expression for the curvature tensor Eqs. (5.67)
and (5.70) with , , , :
The essential point is that these equations determine the time evolution
of the expansion parameters entirely in terms of the sources and the
orientation potentials and their derivatives along directions in the surface
normal to the flow.
1
2 q
|| q || , (5.18)
Deconstructing the Theory 177
b
q q
2q q , (5.19)
q || q q
p
2 p . (5.21)
D2
1 2 2 1 2 1 2
1 2 1 2 . (5.22)
This equation shows explicitly that the energy density is the source of
curvature since a non-zero value of the energy density requires non-zero
values of at least some of the orientation potentials.
The exposition of the theory is complete in the co-moving
orthonormal coordinate basis. We make the following observations,
summarized in Table 5.3, for a space-time of dimension D n 1 . The
other set of relationships are constraints, partial differential equations on
the surface orthogonal to the direction of proper-time and are listed in
Table 5.3.
178 Geometry, Language and Strategy—Vol. 2
Symmetry Eq. 3
R R (5.73) 1
8 D D 1 D 2 D 3
Bianchi identity1 0
R R R 0 Eq. (5.76) 1
24 D 1 D 2 D 3 D 4
Bianchi identity 1
R R R 0 Eq. (5.77) 1
6 D 1 D 2 D 3
It can be shown that the curvature tensor computed from these metric
potentials will then automatically satisfy all of the symmetry relations,
Sec. 3.11, Exs. 9-13.
The situation is quite different in the Co-Moving Coordinate Basis.
The metric is a constant, the Minkowski metric. The unknowns are the
orientation potentials in Table 5.2, and the symmetry relations are no
longer automatically satisfied. Structurally, the curvature tensor is
1
See [Gockeler & Schucker, 1987] and Cf. Sec. 5.6, Ex. 19.
Deconstructing the Theory 179
antisymmetric in the first two indices and the last two indices, so the total
number of components is
D 2 D 1 .
2
1
4
which is
1
6 D 2 D 1 D 2 .
We end this section with the conservation laws Eq. (3.42) in the co-
moving orthonormal frame. These conservation laws are imposed on the
sources because of the field equations, see Ex. 16 in Sec. 3.11, Eq.
(3.75). We have both a transverse and longitudinal part:
m
p q h p | h
d . (5.23)
V | V | p 0
d
m
p q p p p
. (5.24)
p 0
The first equation is algebraic and linear for q in which all the indices
are transverse to the flow. The second equation gives us the time
evolution of the energy density.
Deconstructing the Theory 181
We start with the first set of equations in Table 5.2, the field Eqs. (5.17)
for , the expansion coefficients. Based on the model
assumptions, the expansion coefficients are zero in both the active and
inactive spaces, but not necessarily the mixed. The time variations for the
first two must be set to zero in Eq. (5.17) based on these model
assumptions:
0
. (5.25)
0
When both components are both active, the first equation simplifies, with
, , , :
182 Geometry, Language and Strategy—Vol. 2
We have covered the first two. The third and sixth are identically
satisfied in the player fixed frame model. The fourth is:
The fifth gives the “curl” of the payoff in the co-moving basis, after the
usual simplifications:
184 Geometry, Language and Strategy—Vol. 2
To complete the analysis of the time evolutions Eq. (5.17) for the player
fixed frame model, we write the result, after simplifications, for the
expansion coefficient , which in general provides the time evolution:
2
Electric and magnetic fields depend on the frame of reference. Our terminology in this
section is specific to the co-moving frame of reference.
Deconstructing the Theory 185
We tackle item two next in Table 5.3, which provides the consequences
of the field equations that result from R 0 that are named the
Codacci equations, Eq. (5.15). After simplification, in the player fixed
frame model, the proper inactive components give the divergence of the
mixed sum of expansion and vorticity components:
p
p ph h
n 1
2 q 3 2
3 2 1 2 1 2 0 . (5.38)
We turn to row four of the time evolutions, Table 5.2, which gives the
time evolution of the vorticity components, Eq. (5.9) that result from the
symmetry R R . We write out these conditions for the various
cases, , starting with the first case that both indices are
active. After simplification the result for the player fixed frame model is:
1 2 e q . (5.40)
The third case gives the time evolution for the inactive components,
which after simplification is:
0 . (5.41)
The last orientation potentials to analyze from Table 5.2 are the time
evolutions of the last row. There are six cases to consider:
We start with the first case and simplify Eq. (5.71) using the model
assumptions:
If this starts at zero it stays at zero over time. The symmetric part of
Eq. (5.43) is:
This assumes that the antisymmetric part of is in fact zero. It then
follows from this and Eq. (5.32) that:
Sticking with the same elements of Table 5.2, we consider the second
potential in Eq. (5.42). In the model, after simplification it yields:
188 Geometry, Language and Strategy—Vol. 2
It is identically zero (i.e. no new constraint) using Eqs. (5.46) and (5.40).
The fifth term computes to an identically zero time evolution:
0 . (5.51)
In this case the result is not automatically zero, so that we obtain the
“curl” equation for . Coupling this equation with the time variation
Eq. (5.48), we achieve the following simpler expression:
Deconstructing the Theory 189
Comparing this with Maxwell’s Eq. (1.14), these equations are the
generalization of Faraday’s Law to decision process theory.
We have now computed all of the items in Table 5.2; we have not
explicitly checked Eq. (3.71), but leave to an exercise to check that they
are satisfied. The remaining equations result from the conservation laws
on the inertial sources, Eq. (5.24). The longitudinal conservation law in
Eq. (5.24) simplifies for the player fixed frame model:
The results from the last section are summarized in Tables 5.4, 5.5 and
5.6 in this section. We discuss the distinctions that arise from these
results. In thinking about the results from the previous sections, it is
tempting to base distinctions solely on concepts that arise in game
190 Geometry, Language and Strategy—Vol. 2
dV b
gab abcV bV c Vk Fabk V b 1 2 V jV k a jk 0 . (5.57)
d
This shows that the acceleration (the first term) is determined by the
active orientation potentials abc , the electromagnetic forces Vk Fabk V b
and additional forces due to the charges and the gradient of the inactive
Deconstructing the Theory 191
metric. It is worth noting that this latter term can be written in terms of
the co-moving frame orientation potentials:
q e e 1 Ee e
2
1 2 V jV k a jk e e a
2
E a
2 e e e . (5.58)
1 e e
2
5.4.1 Distinctions—Persistency
,
. (5.59)
f j
E j | E j | F j ab E a Eb
Magnetic field Eq. (5.33) gives the “curl” of the co-moving
magnetic field. The time evolution is Eq. (5.53)
generalizes Faraday’s law Eq. (1.14):
Player Time evolution is Eq. (5.45). The divergence Eq. (5.27).
bonding The matrix commutator and curl are determined by Eq.
(5.30), since the time dependence of the magnetic field is
determined through Eq. (5.53) by the electric field;
Tidal Eq. (5.36) gives the divergence, Eq. (5.49) gives the
magnetic “curl” and Eq. (5.39) gives the time evolution.
f j
2 E j 2 E j . (5.60)
The dynamics of the tidal magnetic field differs from as seen in
Table 5.5. For example we see that the time variation depends on the curl
of the acceleration and distinguishes the electric field from the tidal
player interest. These effects need more detailed study. We may hope to
learn more from these dynamics in decision processes than what we
currently have learned from applications in physics.
What effects do these tidal strains and tidal spins represent in decision
processes? We have been able to draw little from game theory as these
effects would be absent at equilibrium. A preliminary study by Thomas
& Kane [2010] suggests that the player interest provides the possibility
for players to demonstrate interdependent and independent behaviors.
We will review that work in Chap. 7, updating it to conform to the
decision process theory outlined here using the player fixed frame model.
The orientation potentials describe tidal spins and tidal strains, the
orientation flux fields, which relate directly to the geometry of the
behavior. Decision process theory provides not only for strains but for
the organizational system components such as p of the stress tensor
that, along with the energy density and energy flow, characterize the
energy momentum of organizational fields. The orientation flux fields
are keys to decision process theory. The effects of frames that change
with the orientation flux fields provide the basis for an understanding of
how energy attracts or interacts with energy. Since we take energy to be
more fundamental than utility or value, we are in fact hoping to provide a
deeper understanding of economic value. We see in Table 5.6 that the
orientation flux field is directly related to the components p of the
stress tensor along the active directions. With the physics analogy as a
guide, we see that the metric components g 00 , Eq. (4.72), in the time
direction will be determined in part by the equations for the
transformations E a , which are determined by the exactness conditions
Eq. (4.60) and found to depend critically on the orientation flux field.
Deconstructing the Theory 199
p
q p 2 2
n 1
q q q .
(5.61)
p ph
p ph . (5.62)
p ph 0
p. (5.63)
1
2 q q q q 2
1 2 1 2
2
2
p
p ph h .
n 1
Pressure p 1
n p
p Algebraic relation between pressure p , which is the
energy average stress and the stress tensor components; Eq.
density (5.38) is an alternate to Eq. (5.37) for the energy
density; Eq. (5.54) provides the time evolution:
p
p ph h 1 2
n 1
1 2 2
q 3 2 3 2 0,
2 p .
Stress p Eqs. (5.56) and (5.55) must be satisfied by the sources.
tensor p The stress components internal to the players are not
p constrained.
5.5 Outcomes
5.6 Exercises
(1) Show that the Fermi derivative as defined in Eq. (5.2) leaves the co-
moving orthonormal metric (which is Minkowski) unchanged.
(2) Provide the frame equations in the player fixed frame model starting
from Eq. (4.62).
(3) Derive the time dependence of the charge, Eq. (4.59).
(4) Derive the space dependence of the player interest, Eq. (4.69).
(5) Demonstrate that the differential of the mixed 1-form potential is,
with , , , (the flow is in the direction indicated by the
Greek letter omicron ‘ ’)
dω α ο
q q q V Eβ
b V Eβ
Eβ E γ
q Eβ E γ . (5.65)
(6) Use Eq. (5.65) to show that the mixed curvature 2-form is, with
, , , :
Deconstructing the Theory 205
R αο
q q q q V Eβ
b V E β
b V Eβ
Eβ Eγ
q Eβ Eγ . (5.66)
(7) Extract the following curvature tensor components from Eq. (5.66),
with , , , :
(8) Use Eq. (5.7) to show that the transverse components of the 1-form
potential have the following differential 2-form, with , , , :
dω α β
V Eγ
q V Eγ
E E γ . (5.68)
(9) Use Eq. (5.68) to show that the transverse curvature 2-form is
( , , , ):
206 Geometry, Language and Strategy—Vol. 2
R α β d ω α β ω α ο ω ο β ω α ω β ,
R αβ
V Eγ
q q V Eγ
q V E γ
E E γ
E E γ . (5.69)
(10) Extract the following curvature tensor components from Eq. (5.69),
with , , , :
(11) Use Eqs. (5.70), (5.67) and the symmetry relations R R to
establish the time evolution of the transverse orientation potentials
with , , , :
(12) Use the symmetry relation R R and Eq. (5.70) to obtain the
constraints that are spatial partial differential equations for the
orientation potentials, with , , , :
Deconstructing the Theory 207
2 2 2
2 2 0. (5.73)
(14) From the form of the energy momentum tensor Eq. (3.74)
demonstrate that:
(15) Use Eq. (5.74) and the contracted forms for the curvature tensor to
show with , , , :
1 2 2 1 2
1 2 1 2
1 2 . (5.75)
(16) Use Eq. (5.75) to show that Eq. (5.13) and the “trace” of Eq. (5.17)
are consistent.
(17) Derive Eq. (5.24) from Eq. (5.23).
(18) Prove that if the “curl” and “divergence” are known, then the fields
are determined. Use the Cauchy-Kowalewsky existence theorem
[Courant & Hilbert, 1962, p. 39].
(19) Check to see whether the conditions Eq. (3.71) in the co-moving
orthonormal coordinate basis provide any new restrictions. Since we
have already checked each of the symmetry conditions Eq. (3.72),
use the fact that we only need to check the cyclic tensor Eq. (3.73)
208 Geometry, Language and Strategy—Vol. 2
and derive the forms below to show that no new constraints result,
with , , , , :
1
2 X q q q
. (5.77)
Our starting point for decision process theory is Eq. (4.1). If we adopt the
centrally co-moving hypothesis from Vol. 1, then in the (holonomic)
normal-form coordinate basis, there is a commuting Killing vector
whose active1 vector field components K a V a corresponding to the
time isometry are proportional to the energy flow; the inactive strategy
components are zero K j 0 . This means the centrally co-moving frame
is holonomic and the partial differential equations will be independent
of time.
We can solve these equations and then use the harmonic gauge
equations to return to the normal-form coordinate basis and enforce the
initial boundary conditions. An alternate, more restrictive and more
explicit approach is based on the player fixed frame model in Chap. 5.
We obtained partial differential equations, many but not all of which
were independent of proper-time in the orthonormal co-moving frame. If
we impose the additional condition that in this frame, all the orientation
potentials are proper-time independent, we can show that we adhere to
the centrally co-moving hypothesis. We suggest that this hypothesis may
be of longer term value in identifying interesting solutions of the theory.
As we suggested in the last chapter, significant insight is obtained in
the study of electromagnetic fields by starting with a study of electro-
static and magneto-static phenomena, in which all of the fields are steady
state or independent of time. The motivation is that real world systems
often have this property. The success of game theory using equilibrium
solutions suggests that decisions may share the same property. Therefore
1
By active we mean both time and the active strategies.
209
210 Geometry, Language and Strategy—Vol. 2
it makes sense to add this requirement to our player fixed frame model.
Unless stated explicitly otherwise, we make that assumption in the
remainder of this volume.
Because there are currents and magnetic fields, the approximation is
steady state. We emphasize that we are not assuming that the flows are
zero, just steady state. There are currents for example, associated with
each player. This is the analog of AC behavior in electrical circuits as
opposed to DC circuits in which there are no currents. These electro-
static and magneto-static solutions can be extended to cover features that
appear in fully dynamic systems. Wave phenomena are added by
extending these steady state solutions to harmonic solutions (phasors)
that can be solved using the same techniques as used for DC solutions. In
this way, more realistic dynamic solutions are built as superpositions of
harmonics.
The resultant solutions will be for any number of active or inactive
strategies and any number of players in the player fixed frame model in
which the orientation potentials in the co-moving frame are steady state.
The resultant harmonics might be termed electro-gravitational waves, as
the energy can move back and forth from the electromagnetic field to the
gravitational field.
The models we obtain are expressed as coupled partial differential
equations, albeit with a rather large number (potentially hundreds) of
variables, which admit complete numerical solutions. We are able to
obtain solutions if we meet certain criteria. It is a general theorem
[Courant & Hilbert, 1962] that when these criteria are met, there are
well-behaved solutions. That being the case, we can then in later chapters
solve these equations for interesting decision processes. Since
mathematical existence proofs are not the same as practical methods for
solutions, we have to demonstrate both that we meet the appropriate
criteria and that we can indeed carry out this program in practice. We
have to carefully specify all of the equations and constraints and evaluate
whether solutions are not only feasible but practical.
This chapter derives a harmonic wave equation from the player fixed
frame model that allows us to obtain solutions in the normal-form
coordinate basis. With an eye to finding a consistent set of solutions, the
model is modified, Sec. 6.1, so that the orientation potentials are steady
Steady State Harmonics 211
state, though still allowing for the possibility that the harmonic wave
equation produces active coordinates that are not. In Sec. 6.2 we consider
streamline solutions.
We discuss in detail in Sec. 6.3 a model for the player currents that
represent the energy momentum stress tensor for the constraint effects
that may drive solution behaviors. We say more about the inactive
components in Sec. 6.4 and the active components in Sec. 6.5. We then
obtain the frame-wave equation in Sec. 6.6 in the Co-Moving
Orthonormal Basis. We establish in Sec. 6.7 that there is a central co-
moving holonomic frame in which the metric is independent of time. In
that frame, the central time and hence the class of models considered
satisfies the centrally co-moving hypothesis of Vol. 1. For the fixed
frame model, in the central holonomic frame the active flow components
are zero, Ex. 19, Eq. (6.70).
For reference, we provide a special case of the model when there is a
single active strategy, Sec. 6.8. We use this model for two interesting
examples later in the book: see the Prisoner’s Dilemma (Chaps. 7-9) and
Robinson Crusoe economics (Chap. 10).
that are relevant. Not every equation is a differential equation; some are
algebraic. The algebraic equations require special attention, which we do
in Secs. 6.3-6.4, which leads to the assumption that the shear components
are zero (which we equate to very large conductivity) and the player
bonding tensors are diagonal in the inactive indices.
We start by deciding whether the equations have solutions at all. With
this in mind, in this section for Table 5.1, we temporarily assume that
the active transformations are independent of time. The exactness
conditions Eqs. (4.60) and (4.71) reduce to:
E a q 2 e E a
E a q E a . (6.1)
E a E a 2 e E a
q 2 p . (6.2)
This is more complicated than Kirchhoff’s law since the curl is not zero.
The curl depends on the acceleration as well as the player bonding
tensors. Because we assume all fields are independent of proper-time, the
time dependent Eq. (5.46) is satisfied identically. This is an example of
the self-consistency issue we raised: we have to satisfy all of the
Steady State Harmonics 213
q q 4 . (6.4)
This equation along with Eq. (5.26), Table 5.6 determines the differential
equations for the acceleration along the active directions.
There is one equation that is not identically satisfied, Eq. (5.30):
x a , X n a n . (6.7)
n
dxa
Z a x s . (6.8)
ds
These coupled first order equations have a unique solution and provide a
definition of the coordinate s associated with the vector field. The
coordinate s by its definition moves along the streamline and thus
provides a path dependent definition of proper-time. We refer to this
simply as the proper-time, with the understanding that the path is along
the streamline.
We apply this concept to the orthonormal set of vectors and conclude
that at each point there will be integral curves associated with each
orthonormal vector. In particular there will be an integral curve
associated with the flow that defines the proper-time (a streamline):
Steady State Harmonics 215
dxa
x E b x
a b
E a x .
a
(6.9)
d
The scalar field is the measure defined along the integral curve with
1 (Cf. Exs. 5-6).
We compute the tangent vectors from the holonomic scalar function
x a y , using the result of Ex. 6 in terms of the characteristic vector
potential (or simply characteristic potential) a :
E a x a
. (6.10)
E a x a 2a x a
E t t 0 a . (6.11)
0
The characteristic potential thus provides the initial values of this frame
transformation. If the “curl” of the vector field is given (next paragraph),
then different frames would be distinguished only by the divergence of
the vector field:
a a . (6.12)
a a 0 . (6.13)
2
The potential here is proportional to the analog of the gravitomagnetic potential of
general relativity.
216 Geometry, Language and Strategy—Vol. 2
q
3
As a subtle reminder, the exponential function is written as e using the “function”
font to distinguish it from the variables e that represent the player charges.
Steady State Harmonics 217
0 . (6.15)
We note that our argument does not logically require that the stress
components (player current) p be zero. We insure that the energy
density is steady state Eq. (5.54), if we require the shear components to
be small (very large conductivity). The corresponding stress components
need not vanish. For each (proper) player , the stress components p
represent their view of the player passion to exercise strategy . The
“divergence” of the payoff for this player, Eq. (6.2), shows that the
player passion is the source of their view of the player payoff in the same
way that the electric current is the source of the magnetic field in
physics:
For a single active strategy, every term but the last depends on the
payoffs, which must vanish since the payoffs are antisymmetric in
the active proper strategies, of which there is only one. This implies for a
single active strategy that
p 0 . (6.17)
For two or more active strategies however, it is possible for the player
passion components to be non-zero.
The next question is to determine the appropriate equations of motion
for the player passion. One possibility is the following player ownership
rule for player passions (Cf. Chap. 10): each player is accountable and
hence owns only his or her own strategies S , where S is the set of
strategies owned by . By this we mean there can be no player passion
for a strategy that is not owned by player , or equivalently p 0 for
every strategy not owned S . We may have players that own no
strategies: we call them dependent players. A player that owns at least
one strategy is therefore a non-dependent player. To insure that these
distinctions carry a meaning over the space and time of the decision
process, a sufficient assumption is that the “curl” of the player passion is
zero. Then there would be a coordinate surface whose normal is aligned
with each owned strategy. All such surfaces would be in the subspace the
player does not own.
Steady State Harmonics 219
To get small electric field we must go farther and say that the
conductivity is very large. If the conductivity is not constant there will be
an additional term in the second equation.
We see that the “curl” X need not vanish, though it might be
instructive to consider a model in which it does. We will denote this as
the ownership model, loosely related to the discussion above. There is no
assumption that the electric field is proportional to the player current: we
simply assert that X 0 . This model does not make the conductivity
assumption.
For the conductivity model, there also exists a potential (Ex. 14) that
can be used to define coordinate surfaces without assuming that X is
identically zero. The potential is found by identifying an integrating
factor for the player passion.
This leaves us with two quite distinct and useful models for the
player passion, both defined in terms of potential fields: the ownership
model and the conductivity model. We will study both. We assert that in
each case these potentials provide a unique and well-defined set of
definitions of player ownership based on the coordinate surfaces defined
by their potentials. The strength of ownership is a measure of the passion
the player brings to the decision. This description includes the
conductivity model where the “curl” does not vanish (Cf. Ex. 14).
We see the connection to ownership as follows. For the pure
strategies the player has control over, she is indifferent to which one
should be picked. The choice can be made on the basis of what makes
the most strategic sense: in game theory this choice is based on an
optimization strategy. We think of the surfaces of constant player passion
in the same way: they describe choices for which the player is
indifferent. The passion vector for each player is normal to her
indifference surface. The verbiage makes some sense because the
opposite of passion would be indifference.
0
. (6.19)
The first line is the “curl” equation and says that the player bonding
matrices are derivable from a potential. The second line consists of
algebraic equations stating that symmetric matrices mutually commute.
Symmetric matrices that mutually commute can be put into diagonal
form with a common transformation.
We identify the class of solutions that result from our steady state
hypothesis in which we must have a diagonal form for the player
bonding with a condition on the stresses:
Since the bond strain matrices have zero “curl”, they can be derived
from a potential, . We obtain diagonal player bonding
matrices with a strategic viscosity that is arbitrary and a reduced
pressure tensor that is diagonal. The divergence condition Eq.
(5.27) then sets the diagonal components:
p
q ph h . (6.21)
n 1
For steady state tensors, we use Ex. 1, Eq. (6.53) that implies we can use
ordinary derivatives. The resultant bond strain matrices are diagonal and
are derived from a potential so both equations in Eq. (6.19) are satisfied.
The result can be generalized by transforming the equations using a
constant non-singular transformation. If this same transformation is
applied to the initial coordinate basis, then we can align each inactive
strategy j with a corresponding proper inactive strategy j . The
222 Geometry, Language and Strategy—Vol. 2
equations for the frame transformations Table 5.1 then maintain that
choice:
E j E j . (6.22)
The inactive metric represents the degree of cooperation between the two
players and is equal to the overlap of the player interest flows (player
flows or charges). This is after making an appropriate rotation of frames
to identify the “diagonal” players.
We achieve additional insight by looking at the player bonding
shear equation defined by subtracting from the player bonding tensor a
multiple of the diagonal matrix so that the resultant bond shear tensor
has zero trace (Sec. 8.10, Ex. 3):
q 1
ni h . (6.24)
h
h 1 h p
n 1
q 2 1 2
1
We are free to choose the viscosity vector and the reduced pressure
components . We compute the average pressure from Eq. (6.26) and
224 Geometry, Language and Strategy—Vol. 2
the player current p from Eq. (6.18). We show in this section that the
field equations determine the remaining stress components, which are the
active stresses p . We use the fact that the acceleration vector q has
zero “curl” (Sec. 6.3) so the vector field is the gradient of a scalar field
q q . Furthermore, the divergence is determined by the scalar fields
and the average pressure:
In addition to the steady state scalar equations based on the steady state
hypothesis, we extract a time dependent wave equation that applies to the
player fixed frame model. We believe the result justifies the effort. With
the assumption of zero electric field, the coefficients of the expansion
(6.7) are determined by the harmonic gauge condition Eq. (4.71):
1 e e E
a
E a q E a 0 . (6.29)
1 e e E
a
1 e e X s 2 a s 2 s 1 s ,
s 0
q
E a
q ,
E a X q X
n 0
s
a s
s0
s
a
s s
2a X s 1 s 1
a s
s 0
q X s a s s q q X s a s 2 s
s 0 s 0
2a q X s 1 s 1 s
a 2
s 0
2q a X s 1a s 1 s s
s 0
We see that the second term in the series is determined by the first two
terms: the position and the initial value of E a . Though more
complicated than solving differential equations with constant
coefficients, it is clear that the same pattern maintains. The coefficient
X a s 2 will depend on the functions X a s , X a s 1 and their gradients. The
general case is Ex. 16. This demonstrates that given arbitrary fields
X a 0 X a1 , all other fields X a n are determined.
The power series demonstrates that there are solutions to the frame-
wave equation Eq. (6.29) based on given initial conditions. We expected
this based on general theorems on elliptic partial differential equations
(Courant & Hilbert, 1962). To summarize, we write the wave equation as
a partial differential equation (Ex. 17):
1 e e
4a a 4a q q q 2 2 x a
x a 2 2a q x a
2 a 2 a 4a q q x a
q 2 q q x a q x a 0. (6.33)
This differs from the usual wave equation because of the time and
strategy dependence of the coefficients. These differences make it
impossible to find solutions that factor into a function of proper-time and
a function of position. We address this problem in the next section.
Nevertheless, we can construct general solutions from linear
combinations of specific solutions. As a step to identify specific
solutions that might be of interest, note that any solution will be
Steady State Harmonics 227
x a U 0 a V0 a . (6.34)
q q 2 q q V
0
a
0. (6.35)
i i
N s s
These power series set the values of the coefficients on the surface 0
by requiring X N 1a X N 2 a 0 for polynomials PN a , , Ex. (18). So,
on the hypersurface z 0 , we have:
i i
s s
X a
s 0 V a
0 Re U a
0 Im
s! s!
Y a
s 0 X s 0
a
. (6.37)
i i
s s
X a s 0 V a 0 Re U a 0 Im
s! s!
This wave equation has the feature that linear combinations of solutions
are again solutions: the equations are linear in x a .
In general, solutions to these equations will reflect oscillations as well
as attenuation (and growth). Such behaviors can be studied by
considering phasor solutions that correspond to a fixed frequency
x a e i , Ex. 33. We call such solutions frame-waves. Even though the
differential equations are equivalent, these bounded solutions will in
general not be equivalent to the harmonic polynomial solutions, which
can lead to unbounded behaviors for the frame transformations.
We can then solve Eq. (6.40) with harmonics (phasors) that are
independent of central-time, Exs. 33-34, where x a x a ei :
2 e2 q 1 e e 4a a 2i eq a q x a 0 . (6.41)
1
a 0 . (6.43)
q
These equations insure that the commutation rules Eq. (4.60) are
satisfied.
We write the partial differential equations that result from the
harmonic wave Eq. (6.33) for the components of Eq. (6.7), where in the
normal-form coordinate basis, t is time, u is the active strategy
direction4 and a t , u :
4
A similar discussion can be made in the central holonomic frame if we use Eq. (6.40).
See Sec. 6.10, Ex. 51.
Steady State Harmonics 235
1 e e
q q 2 2 x a x a 2 q x a
q
q 2 q q x a q x a 0 . (6.44)
The single strategy model is ideal for understanding in better detail how
we obtain the harmonic coordinate behaviors from these
transformations. We expect to gain insight by using the harmonic
polynomials PN a , , Eq. (6.36) and the coefficient equations (Ex. 18):
X N 1a X N 2 a 0,
X N 2 N 1 q
a
XN a
N q q q q N 1 X N a 0,
X N 1a 2 N 1 q X N 1a
N 1 q q Nq q X N 1a 0,
s N 2, N 3,,0 :
X s a 2 s 1 q X s a
s q q q q s 1 X s a
s 2 s 1 1 e e X s 2 a 0. (6.45)
E j E j ,
E j q E j 2 E j ,
e q e 2 e ,
f j
f j
0, f j
0,
f j
2 q E j E j e E j e E j 1 E k Ek . (6.46)
236 Geometry, Language and Strategy—Vol. 2
q 2
p
p ph h . (6.48)
n 1
The inertial forces (using Eq. (6.93) from Ex. 19) determine the active
diagonal stress component:
We are free to pick the energy density scalar and inactive stress
scalars p . The average pressure is determined from the diagonal
stresses, Eq. (3.41).
The inertial forces also determine the divergence of the acceleration,
Eq. (5.61):
p
q q q q 2 p . (6.50)
n 1
1 e e
q q 2 2 x a x a 2 q x a
p
q q 2 p x
a
n 1
q x a 0. (6.51)
x a 2 e 2 q 1 e e x a q x a 0 . (6.52)
6.9 Outcomes
R 1 2 g R T .
The student is expected to see that the various distinctions and their
related field equations of Part 1 are brought together in this chapter
and provide a basis for harmonic solutions along streamlines. These
solutions are electro-gravitational waves of decision process theory.
On the transverse surface, the resultant equations are partial
differential equations that are mathematically similar to those of
electromagnetic theory. The student should gain an appreciation of
how these equations might be solved.
6.10 Exercises
(1) For any scalar field , prove that in the co-moving basis,
| | and hence for the player fixed frame model:
(2) For any scalar field , prove that in the co-moving basis, | |
and hence for the player fixed frame model:
(3) Show that for any scalar function that is a function of the
proper-active strategies y only, that such a function is independent
of proper-time. In particular demonstrate for each coordinate that
y 0 . (6.55)
y . (6.56)
. (6.58)
2a
xa X ,
n 0
n
a n
E a X n
n 0
n
a n 1
,
E a X
n 0
n
a n
X n a n n 2a X n a n n 1.
n 0 n0
(6.59)
(8) Use Ex. 1 to show the first commutation rule of Eq. (4.60) is
satisfied:
E a E a x a x a q 2 e E a . (6.60)
(9) Use Ex. 1 with the scalar and the assumption that the scalar fields
are static to show that a is constrained by the magnetic and tidal
magnetic fields and that the curl of must be zero:
Steady State Harmonics 241
(10) Use Ex. 1 to show the second commutation rule of Eq. (4.60) is
satisfied with the vector field a satisfying Eq. (6.61):
E a E a x a x a 2 e E a . (6.62)
(11) Show that the co-moving coordinate basis field equations from Sec.
5.4 for the player fixed frame model satisfy the normal-form
coordinate basis longitudinal flows, Eqs. (2.49) and (2.50),
modified to include the inertial acceleration:
(12) Show that imposing a gauge condition, such as the following, on the
vector field a in Eq. (6.61) corresponds to a frame transformation,
taking the most general form of to be q in terms of a
scalar field:
a e q 0 . (6.64)
(13) Show that the “curl” of the new seasonal vector potential field
A 2 e q afrom Eq. (6.61) simplifies and determines the
seasonal payoff and acceleration fields. Show that the field
equations for the “curl” of the magnetic fields satisfy these
equations and further, show that with the gauge from the previous
exercise, the vector field depends on the player current in the
second order differential equation shown below:
242 Geometry, Language and Strategy—Vol. 2
eq a eq a eq e
f ,
f f f 0,
e q eq a e p e q
2 e
2 q 2 e .
(6.65)
p P p
P q P
. (6.66)
λ P exp qI φ
p 2 p
(15) Discuss the significance of the source term and using Ex. 14, show
that the passion source term is a scalar multiple of the gradient
p . So in a sense, this conductivity model gradient is as related to
ownership as is the one in the ownership model.
(16) Verify Eqs. (6.30) and (6.31). Use these equations to determine the
general form of the power series in terms of the initial functions
X a 0 and X a1 :
Steady State Harmonics 243
s 1 s 2 1 e e 4a a X s 2 a
X s a 2 s 1 q X s a
s s 1 q q q q X s a
2 s 1 2 s 1 a q a a X s 1a
4 s 1 a X s 1a 0. (6.67)
1 e e
4a a 4a q q q 2 2 x a
x a 2 2a q x a
2 a 6a q
q q 2 q q x a q x a 0. (6.68)
(18) Use the power series solution to show that the wave Eq. (6.68) has
polynomial solutions PN a , with X N 1a X N 2 a 0 and show
that the order of solving for the coefficients starts with the equation
for X N and then equation for X N 1a using this solution and so
forth:
X N a 2 N 1 q X N a
N q q q q N 1 X N a 0,
X N 1a 2 N 1 q X N 1a
N 1 q q Nq q X N 1a
4 Na X N a 2 N a 2 N 1 a q X N a 0 ,
244 Geometry, Language and Strategy—Vol. 2
s N 2,,0 :
X s a 2 s 1 q X s a
s q q s 1 q q X s a
4 s 1 a X s 1a 2 s 1 2 s 3 q a X s 1a
s 2 s 1 1 e e 4a a X s 2 a 0 . (6.69)
W eq W e q e W 2a e q
W 0 W 0 W . (6.70)
W k Ek W k Ek W k 0
(20) Show that the transformation W W W from the central
holonomic frame to the co-moving orthonormal frame has the
following values and is the inverse of the transformation Eq. (6.70):
e q e q e
W W W 0
1 e e 1 e e
2a 2a
W W e W . (6.71)
1 e e 1 e e
Wk Ek Wk Ek Wk 0
Steady State Harmonics 245
(21) Using the frame transformations from Ex. 19, show that the
following provide the frame transformation E from the normal-
form coordinate basis to the central holonomic frame basis. In
addition show that the inverse frame transformation E is as
indicated below.
E a e q E a E a E a 2a E a E ka 0
E j 0 E j 0 E kj kj
. (6.72)
Ea 1 e e e q Ea 2a e q Ea Ea Ea Ea k 0
E j 0 E j 0 E j k kj
(22) In the central holonomic frame, show that the inactive metric
elements jk jk are equal to the corresponding inactive metric
elements in the normal-form coordinate frame, show that the mixed
metric elements g kb 0 are zero and show that the active
contravariant metric elements g ab have the values given below.
Furthermore, show that the determinant of the active metric
elements are as indicated:
g 1 e e 4a a e 2 q g 2a eq
g 2a eq g . (6.73)
det g ab 1 e e e2 q
(23) In the central holonomic frame, show that the active covariant
metric elements are:
e 2 q 2a e q
g g
1 e e 1 e e
. (6.74)
2a e q 4a a
g g
1 e e 1 e e
(24) Using the fact from Ex. (23) that the metric elements in the central
holonomic frame are independent of central time , show that
246 Geometry, Language and Strategy—Vol. 2
(25) Use the frame transformation to obtain the player payoff matrices in
the central holonomic frame, considered as tensors in the space of
time and active and inactive strategies:
F j e q f j
F f
j j
2a f j
2a f j
. (6.75)
F k F k F
j j j
ik
0
(26) Based on the form of the payoff matrix Eq. (6.75) in the central
holonomic frame and using the definition Eq. (4.30) transformed to
the central holonomic frame, show that the payoff matrix for the
active components is the curl F abj a Abj b Aaj of some potential
function Aaj by showing:
(27) To determine the potential function Aaj implied by Ex. 26, both
the curl and divergence need to be specified. Show that the gauge
condition g ab a A j b 0 in the normal form coordinate frame, which
we call the harmonic gauge for the vector potential, Vol. 1, can be
expressed in the central holonomic frame by establishing the
following relations, where the covariant derivatives are here
restricted to active components only:
Steady State Harmonics 247
1
g
b
g g ab A j a g ab b A j a 0
g ab A j a ;b
1
g
b g ab g A ja . (6.77)
ab
g A j
a ;b 1
2 g A a b ln 0
ab j
1
g
b
g g ab Aaj g ab b Aaj
1
g
b
g g ab Aaj 0
(28) Show that in the central holonomic frame the active flow
components are V W e q , V W 0 and the inactive
flow components are V k W k V k . If we call the covariant-flows
the “momenta”, show that these momenta are
e q 2a
V
, V
1 e e 1 e e
from the sum of the three contributions. In the process show that the
competitive acceleration and cooperative acceleration have the
forms provided in the last two equations. You will also have
demonstrated that the three acceleration contributions are each
independent of proper-time as is the active geometry acceleration in
the central holonomic frame.
Qa g ab V c c V b abc V b V c
Qa Vk Fabk V b 1 2 V j V k a jk qa
Q 0
q 2 e e e
Q
1 e e
2
. (6.78)
2q e e 2 e e 1 e 2 e e
Ek fk
1 e e
2
1 e e
2 2
(30) Using Ex. 29 for the active geometry acceleration in the central
holonomic frame and the transformation properties Ex. 21, show
that the active geometry acceleration in the normal-form coordinate
basis is Qa Ea Q . So in particular the time dependence is
provided entirely by the transformation Ea . Furthermore, show
that the covariant magnitude of this acceleration g ab Qa Qb is
Q Q and so a scalar independent of proper-time: like the
pressure p , energy density and cooperation potentials jk , the
covariant magnitude is constant along a streamline. Show that for
any scalar that is constant along the streamline, the gradient is
a Ea and so show that in the normal-form coordinate basis
such scalars will in general be time dependent if there are non-trivial
harmonics.
(31) In our numerical analysis in Vol. 1, in addition to an exactly solved
single strategy model, we computed a variety of examples in which
Steady State Harmonics 249
1 e e
4a a e 2 q 2 x a x a
4a e q x a 2e q a q x a
q x a 0. (6.79)
(33) Show that one can apply the phasor approach directly to the partial
differential equation in Ex. 32, namely one can superpose solutions
of the type x a x a ei where the phasor component x a satisfies
the following equation that is independent of the central time:
250 Geometry, Language and Strategy—Vol. 2
2 e2 q 1 e e 4a a 2i eq a q x a 0 . (6.80)
(34) Show that the equation in Ex. 32 also has linear solutions:
x a U 0 a V0 a ,
U 0 a q U 0a
2 e q a q V0 a 4a e q V0 a 0,
V0 a q V0 a 0 . (6.81)
x a U a cos V a sin ,
U a 2 e 2 q 1 e e 4a a U a
q U a
2 e q a q V a 4 e q a V a 0,
V a 2 e 2 q 1 e e 4a a V a
q V a
2 e q a q U a 4 e q a U a 0. (6.82)
p 2 z q z p 2 1 e e 0
. (6.84)
2 z q z i 1 e e 4 z q z
2
1 2 1
(38) In the central holonomic frame the player payoff fields are
determined by a current, Eq. (4.14). Show that the current
components are given in terms of the following expressions
involving the co-moving orthonormal scalars:
p j E j p
. (6.85)
p j eq E j p e 2eq E j a p
(39) Show that the gauge condition implicitly chosen in the central
holonomic frame is
252 Geometry, Language and Strategy—Vol. 2
b g ab 0 . (6.86)
Also show that in the central holonomic frame, you can implement
the player independence hypothesis from Vol. 1 by explicitly
defining the player ownership structures to be the unit diagonals
Q , which are unity when the strategy (active or inactive) is owned
by that player and zero otherwise. This gives an alternative to the
strong ownership concept in Sec. 11.5.2. Show that different
boundary conditions for the flows are addressed by the
transformations below and the harmonic wave equation.
k
k k
V k
0
W k
,
0 1 0
0 0
Steady State Harmonics 253
k
k kk V k 0
W 1
0 1 0
. (6.87)
0 0
A e q e p e q
2 e 2 q 2 e . (6.88)
(42) Assume that we have the assumptions stated in Ex. 41 and that z is
a coordinate direction that is constant on the initial surface. Show
that the gauge condition z 0 and the equations of motion leads
to the following equations, which form an elliptic partial differential
equation on the initial surface.
z A 0 z z Az z A 0 ,
z
A A
z z z A ,
z
e q Az z A
z
e p z z e z q
dx dx dx 0
B b b
B
B b b dx dx dx 0 . (6.90)
B B 0 λ B exp φ
q 2 p
(44) Continuing with the vector potentials from Ex. 43, show that the
equations of motion Eq. (6.16) can be determined with gauge
conditions b 0 and 0 defined on the surface:
(45) Show that for Ex. 44 and an initial surface defined by a coordinate
axis z being constant, the elliptic partial differential equations on
the surface for the gauge condition are:
b z z b
z
q h 2 b z
b z 0 . (6.92)
(46) For the single active strategy streamline solutions of Sec. 6.8, show
that the inertial forces, Table 5.6, provides the diagonal components
of stress, Eq. (5.38):
Steady State Harmonics 255
p
q p ph h
n 1
1 2 1 2 0. (6.93)
(47) For the single active strategy streamline solutions of Sec. 6.8, show
that the inertial forces, Table 5.6, provide the divergence of the
acceleration field given by Eq. (5.61).
(48) For the single active strategy streamline solutions of Sec. 6.8, show
that Eq. (6.48) can be used to determine the divergence of the player
bonding:
(49) For the single active strategy streamline solutions of Sec. 6.8, using
the divergence of the acceleration Eq. (6.50), show that the
harmonic wave Eq. (6.44) can be expressed as:
1 e e
q q 2 2 x a x a 2 q x a
p a
q q 2 p x
n 1
q x a 0. (6.95)
(50) For the single active strategy streamline solutions of Sec. 6.8, using
the divergence of the acceleration Eq. (6.50), show that the
harmonic polynomials Eq. (6.45) can be expressed as:
X N 1a X N 2 a 0,
X N 2 N 1 q
a
XN a
p a
N Nq q 2 p XN 0,
n 1
X N 1a 2 N 1 q X N 1a
p
N 1 N 1 q q 2 p X N 1
a
0,
n 1
256 Geometry, Language and Strategy—Vol. 2
s N 2, N 3,,0 :
p a
X s a s sq q 2 p Xs
n 1
2 s 1 q X s a s 2 s 1 1 e e X s 2 a 0 . (6.96)
(51) For the single strategy model, show that the wave equation (Ex. 32)
simplifies to:
x a 2 e 2 q 1 e e x a q x a 0 . (6.97)
g z g z 0
. (6.98)
z g z z g z z g zz 0
(53) For the model and steady state gauge in Ex. 52, show that for the
metric, a reasonable set of starting values on the initial surface
would be:
g
. (6.99)
z g 0
(54) Show that the player fixed frame model solutions in the centrally
co-moving frame not only satisfy the seasonal gauge in Ex. 52 but
also the gauge Ex. 39. Show that these two conditions however are
not equivalent and that in general, there is no reason to believe that
the gauge for the player fixed frame model will continue to hold for
more general models. Start with the seasonal gauge and on the
initial surface z 0 , impose the additional constraints as written
below. Discuss whether these constraints will be maintained by the
field equations. We have demonstrated in Ex. 52 that the last
equation is maintained, but did not draw any conclusions about the
other equations.
g z g z 0 g 0
g z g z 0 g 0 . (6.100)
z
g z g 0 z g 0
zz zz
(55) With the assumptions of Ex. 52 for the seasonal gauge, show in
particular that the curvature components Rzz Rza Rab , Eq.
(3.36), that enter into the field Eqs. (3.31) for the transverse active
strategies ,, z, are given below. The notation is that the
metric splits into a block diagonal form with a z -block g zz and a
non- z -block g ab : a , b , . Do these equations imply that the
seasonal gauge conditions are maintained?
258 Geometry, Language and Strategy—Vol. 2
g az z g az z g zz 0,
g zz g zz , g ab g ab , z ,
Rzz 1
2 g g zz 1 4 g zz g g zz g zz
1 4 g g zz ln g
1 2 z z ln g 1 4 g de g cf z g ce z g df , (6.101)
Raz 1 2 gb z g ab 1 4 g zz gb g zz z g ab
1 4 gb z g ab ln g 1 2 a z ln g
1 4 g zz a g zz z ln g 1 4 g cf g de a g ce z g df , (6.102)
Rab 1
2 a b ln g ab
1 2 ab ln g acd bdc
1 2 ab
g zz g zz 1 4 g zz g zz a g zz b g zz
1 2 g zz a b g zz 1 2 g zz z z g ab
1 4 g zz z g ab z ln g 1 2 g zz g cd z g ac z g bd . (6.103)
(56) Given the results from Ex. 55, show that an appropriate gauge for
the vector potential associated with the time isometry is the
following, which can be satisfied by starting with a z z az 0 on
the hypersurface z 0 . Show that the gauge will be maintained by
the field equations at all other z .
g a g
g z a z g . (6.104)
az 0
Chapter 7
261
262 Geometry, Language and Strategy—Vol. 2
that can move along the transmission line. Thus there are two parts to the
story.
First there is the co-moving behavior that provides the relationship
between the frequency of the driving amplitude and the wavelength of
the standing wave. Second there is the circuit like relationships between
the amplitude and phase of the travellng wave at the source and the
amplitude and phase of the wave at the load. The difference in decision
process theory is that the co-moving behavior is specified by a detailed
steady state surface geometry, some aspects of which are suggested by
game theory. In addition the circuit relationships are replaced by
complex relationships between conditions on this (hyper)-surface and
equations of motion that determine the behaviors away from that (hyper)-
surface. These decision engineering relationships we explore in Part 2.
The steady state surface geometry is characterized by four major
categories of steady state scenarios: codes of conduct, competition,
cooperation and inertial flow. We illustrate these in a variety of ways,
starting with the prisoner’s dilemma with a single active strategy and
three additional codes of conduct. In Chap. 7, the codes of conduct
provide a new and interesting perspective not often covered in game
theory. This perspective helps to resolve what is sometimes considered a
dilemma when game theory is applied to this problem. In Chap. 8, the
steady state geometry is examined in more detail. In Chap. 9, dynamic
scenarios including determinism and chaos provide further insight into
the dynamic mechanisms.
In Chap. 10, we consider an even simpler game with a single active
strategy and no codes of conduct: Robinson Crusoe economics. We
discuss a taxonomy or organization for the theory. In Chap. 11 we set the
context of these examples with a general review of game theory and
recent economic approaches. A key difference with our approach and
game theory approaches is the way we deal with risk, which is covered in
Chap. 12. Our approach is similar to the systems dynamics approach
[Forrester, 1961], which we cover in Chap. 13. In both our approach and
the systems dynamics approach, it is important to take a global
perspective where we consider and close all of the feedback loops,
especially those that are often missed: those occurring over long time
intervals or distant strategic distances.
Prisoner’s Dilemma: A Code of Conduct Application 263
7.1 Introduction
The results in this section extend previous work by Thomas & Kane,
[2008] and Thomas & Kane [2010], who applied a single strategy player
fixed frame model to the prisoner’s dilemma. They modified the forms
for the payoffs introduced below in Eqs. (7.3) and (7.4), to this model.
We shall see that reducing the number of strategies in this way to a single
active strategy is equivalent to choosing a code of conduct (Sec. 11.2). In
this case, public-interest supplements self-interest.
These authors found similarities to the decision process theory with
the incomplete games proposed by Harsanyi [1967-1968], in his trilogy
on game theory. Decision process theory goes one step further than
Harsanyi in proposing that actions are characterized not only by
competition as measured by payoffs, but by inertia and charge. As with
physical systems, inertia implies that there must be enough force to
overcome inertia in changing a system from a given course. Ordinary
competition-forces—i.e., the forces that arise from one’s beliefs and past
experiences—may not be sufficient. The second is the property of
charge. In their model of public goods games, Eshel, Samuelson,
& Shaked [1998] described players (subjects) as either egoists (who
maximize self-interest) or altruists (who maximize other-interest).
In contrast, Thomas & Kane [2010] suggested that Eshel et al.’s
egoist/altruist distinction corresponds roughly to a well-known and
empirically-tested distinction made in psychology regarding independent
266 Geometry, Language and Strategy—Vol. 2
the game theory Nash equilibrium result. To the extent they change we
obtain the deviations of decision process theory from game theory.
game theory. For two players for a non-zero sum game, an analogous
Nash equilibrium determines the mix of strategies. For the prisoner’s
dilemma it is that both players choose not to confess. They optimize their
self-interest by acting defensively.
To articulate the difference, we don’t need to work directly with the
individual payoffs such as Eq. (7.1) but can construct our argument using
the symmetric game Eq. (1.12):
Fab1 N2 C2 N1 C1 t
N2 0 0 1
10 0 0
C2 0 0 1 9
10 9
10 0
. (7.3)
N1 110 1 0 0 1
0
C1 0 9 10 0 0 9
10 0
t 0 9
10 0
1
0
9
10 0
0
We identify the “hedge” strategy with time t and take this payoff to be
the value at an initial value of time and at an initial strategic point.
We construct the similar payoff for player 2, again evaluated at the
same point:
Fab2 N2 C2 N1 C1 t
N2 0 0 110 1 1
0
C2 0 0 0 9 10 9
10 0
. (7.4)
N1 1
10 0 0 0 0
C1 1 9
10 0 0 9
10 0
t 1
0
9
10 0
0 9
10 0
0
These forms for player 1 and player 2 define new payoff matrices that
are each symmetric. In each case, a two-person game is defined in which
each player has the same set of strategies. The max-min of each
symmetric game is the same as the original game from which it was
constructed and so it represents exactly the same strategic possibilities as
the original game.
Prisoner’s Dilemma: A Code of Conduct Application 271
DVa
qa q Ea
. (7.5)
dV b
gab abcV bV c Vk Fabk V b 1 2V jVk a jk
qa
d
We see that the flow may be in the null space of each of the payoff
matrices and yet the rate of change of the flow is not zero.
Conversely, even if the flow is not in the null space of the payoffs,
the other terms, including the covariant acceleration term qa , can
conspire to make the flow steady state, dV a d 0 . We thus
acknowledge the importance of self-interest, to the extent we apply that
loose concept to the forces derived directly from the payoffs. However
we must also acknowledge the potential importance of other effects. In
order to estimate the size of such other effects we must appeal to model
calculations of the various conspiracy effects in Eq. (7.5).
An important example of such effects consists of resonant
scenarios, which arise when only the acceleration and payoff terms
contribute:
dV b
g ab Vk Fabk V b 0 . (7.6)
d
This special case arises when the metric and payoffs are constant along
the streamline, the cooperative and gradient terms are negligible and the
orientation components are zero. Such harmonic behaviors were
described in Vol. 1. It is worth emphasizing that we expect these
272 Geometry, Language and Strategy—Vol. 2
behaviors in the active space and characterize the driving force in terms
of the active payoffs. Typically these harmonic behaviors oscillate
around the Nash equilibrium. We analyze the competitive behavior
further in Sec. 15.1. In physics, the analogous situation is a charge
particle in a magnetic field oscillating around the direction of the field.
Thus in moving away from game theory to decision process theory
with the steady state hypothesis, we expect a much richer geometry or
landscape. In particular, in the centrally co-moving frame, all of the
terms in Eq. (7.5) will be independent of the central time . It will be in
this frame and the closely related co-moving orthonormal frame that we
investigate the steady state scenarios. Consider the acceleration values
q in the co-moving orthonormal frame. An important attribute of non-
trivial steady state behavior is that there will be acceleration. For the
player fixed frame model, this is found for the active components q .
We see that in the game theory approximation, with the initial
condition that the acceleration is zero, the acceleration stays zero at all
other values of time and space. We break this approximation in the
player fixed frame model, even when the orientation potentials are static.
This is because the variation of the acceleration with distance is
determined by the divergence Eq. (5.61):
p
q p 2 q q q . (7.7)
n 1
Whatever the origin of the conspiracy effects, they are not caused
directly by the payoffs. The results are governed by the inertial, player
interest and player bonding effects.
This suggests that we may learn a significant amount by considering
even the simplest single strategy model in which the payoffs (for the
effective players of the model) are greatly simplified. We then focus our
study of the effects of Eq. (7.7), further simplifying it since with a single
active strategy there are no tidal magnetic effects 0 .
Prisoner’s Dilemma: A Code of Conduct Application 273
m . (7.8)
This changes our thinking of the strategies as being summed to unity for
each player, which focused only on the relative strategic values and
274 Geometry, Language and Strategy—Vol. 2
1
0
a
a 2
j
j 2
. (7.10)
The value of beta depends on the active and inactive flow components.
For the prisoner’s dilemma, the active strategy space has four
dimensions. Since there are two players, the inactive strategy space has
two dimensions. We choose the initial inactive flows to be zero:
j 0 0 .
The total strategic space dimensionality is n 6 . We next set the
active flow values. For numerical work, if we provide a value for ,
we obtain 0 from Eq. (7.10), defined by the initial strategy flow. The
decision mass is then obtained from Eq. (7.8). The Nash equilibrium
for the prisoner’s dilemmas is that both players confess:
Nash b {0,1,0,1, m} , where m can be chosen arbitrarily at this point.
This determines beta.
Nothing in game theory depends on the game value, since it does not
influence the choice of Nash equilibria. Typically, one can add to each
game a constant amount leaving these strategic possibilities invariant. In
the symmetric form this constant is the matrix:
0 0 v v v
0
0 0 v v v
0
c v v v 0 0 v 0 . (7.11)
v v 0 0 v 0
v v v v 0
0 0 0 0
will eliminate all of the payoff time components (elements in the row or
column labeled by time t ). We change this for our example.
Our argument is that in decision process theory, the time components
of the payoff reflect the biases that we believe accurately represents the
notion of the game value from game theory. Not every game needs to be
fair, nor do we require the time components for each player to be equal.
We see that is true for the constant Eq. (7.11) but not for the assumed
forms we have chosen for the prisoner’s dilemma. This suggests another
equivalent form that does not change the game theory strategic content,
but may represent different dynamics in decision process theory.
It should also be possible to frame the prisoner’s dilemma in a way
that it too could be put into the form of a fair game where the time
components for each player are equal. We achieve this by adding
internal payoffs to F 1ab so that the product F 1ab Nash b 0 of the payoff
and the starting strategy Nash b are zero:
0 9
10 45 9 10 0 0 0
9
10 0 0 0 1 0
1
10
45 10 0
1 1
10 0 0 0 . (7.12)
9
10 0 110 0 0 1 0
0 0 0
0 0 0 0
This payoff plus the constant matrix c 9 10 from Eq. (7.11) reproduces
the same payoffs as the submatrix Eq. (7.1). There is an internal payoff
or utility for player 1 to choose to not confess of 110 . For example,
without this bias, the return for player 1 for the equilibrium strategy
would be negative instead of zero. We thus obtain the same strategic
behavior as the symmetrized game Eq. (7.3).
If both players choose to cooperate and not confess ( N1 N 2 ), then
player 1 loses 110 . If player 1 chooses to cooperate and not confess and
player 2 chooses to confess ( N1C2 ), then player 1 loses 1 unit. If player 1
chooses to confess and player 2 chooses to cooperate and not confess
( C1N2 ), player 1 sees maximum benefit and receives 0. Finally if player
1 and player 2 both choose to confess ( C1C2 ), then player 1 loses 9 10 .
276 Geometry, Language and Strategy—Vol. 2
The Nash equilibrium point is the strategy whose product with the payoff
is zero.
In a similar fashion, we choose the payoff for player 2, so that the
product F 2 ab Nash b 0 is also zero:
0 1
10
4
5 110 0 0 0
1
10 0 0 0 1 0
9
10
45 9 10 0 9
10 0 0 0 . (7.13)
1
10 0 9 10 0 0 1 0
0 0 0
0 0 0 0
This payoff plus the constant matrix c 9 10 from Eq. (7.11) reproduces
the same payoffs as the submatrix Eq. (7.2). We thus obtain the same
strategic behavior as the symmetrized game Eq. (7.4).
We have defined some of the initial conditions that are needed to
specify a complete solution to the field equations. Before continuing, it is
worth reiterating the multiplicity of payoffs that are possible, which
seemingly represent the same strategic situation. We look at the
elementary aspects of game theory for example and see that the strategic
content of a game in normal form does not change if the payoffs are all
multiplied by a common factor. Similarly the strategic content doesn’t
change if a constant is added to all of the payoff elements. Finally,
internal payoffs offset by the appropriate time components don’t change
the strategic result. Adjusting the time components also allows scale
changes for the equilibrium strategies of each player.
The max-min rule discussed in Sec. 1.7 yields the same equilibrium
strategies. If the game is fair, there is no difference between these games.
If the game is not fair, the only difference is the size of the payoff, which
will be modified by the same multiplicative factor or additive constant
that modifies the individual payoffs. The important question is whether
this behavior squares with our common sense. It is reasonable to imagine
that the strategic choices will in fact be impacted significantly if a game
with low stakes is made into a high stakes game. In decision process
theory, we in fact expect there to be differences.
Prisoner’s Dilemma: A Code of Conduct Application 277
The reason for this expectation in decision process theory is that the
payoffs fields carry energy. The larger the field values the more energy
these fields carry. We noted the acceleration flow depends on the inertial
and orientation flux fields, Eq. (7.7). The acceleration flow is a direct
measure of changes to the strategic values. The magnetic components in
the co-moving basis play no direct role, whereas the inertial and player
interest terms do. We thus have at least one difference between decision
process theory and game theory models that can be put to the test. As we
develop the player fixed frame model for the prisoner’s dilemma, we will
study the consequences of different game values by changing the initial
conditions with additions of constant matrices Eq. (7.11). In the next
section we introduce codes of conduct and their relationship to inactive
strategies.
The most general approach to the prisoner’s dilemma (Sec. 7.3) is to start
with the four active strategies associated with our normal form payoffs
Eqs. (7.12) and (7.13). To these strategies we add two inactive strategies,
one for each of the two prisoners. We use the normal form payoffs plus
the appropriate constant matrix c v , Eq. (7.11) as the initial payoff
field value. We use an initial flow value that represents the strategic
choice we want to study. For numerical solutions, we specialize to the
player fixed frame model that has streamline solutions that can be found
using the equations from Chap. 6 based on the steady state hypothesis.
We expect solutions that are both steady state and non-steady state.
To break out of the prisoner’s paradox, we impose a player code of
conduct in which we go from two players to five players, three of which
collectively choose a single common active strategy. In these solutions,
there are no active payoffs; nevertheless we can maintain the same initial
conditions for the payoffs and strategies. We still have solutions that are
every bit as stable and correct as the Nash equilibrium solutions when all
strategies are active. Moreover, our approach is not without precedent.
The code of conduct is similar in concept to the standard of behavior of
Von Neumann & Morgenstern [1944, p. 31ff]. They noted that such
278 Geometry, Language and Strategy—Vol. 2
F j xb b Axj . (7.14)
dVk
q E j q E j 0 . (7.15)
d
In particular, for the player fixed frame model, the flow is conserved.
This is distinct from the behavior expected for active strategies that will
be influenced by the rotational effects of the payoff matrix seen in
Eq. (7.5). Multiple numerical examples are provided of this behavior in
Vol. 1.
Prisoner’s Dilemma: A Code of Conduct Application 279
The situation is different for two inactive strategies and a given set of
initial conditions. If two strategies x y are inactive, in addition to the
conservation behavior Eq. (7.15) of these strategies along the streamline,
there can be no payoff between them:
F j xy 0 . (7.16)
Looking at the payoffs for the prisoner’s dilemma, this is not true
simultaneously for our initial payoff values Eqs. (7.12) and (7.13). There
may however be a different basis in which this is true.
We start with a frame in which for each player one direction is
specified by confess (C) and the other not confess (N). An equally
complete choice would be the sum (C+N) and difference (C-N) of the
strategies. We label the coordinates
t 1 2 N 2 C2 N1 C1 ,
s1 1
N1 C1
2
s2 1 2 N 2 C 2
. (7.17)
r1 1 2 N1 C1
r2 1 2 N 2 C2
The sum of the strategies for each player, r1 and r2 , represent how often
each player plays. They represent the player preference scale. The
differences s1 and s2 represent the player relative preference to not
confess. In this basis, we still can’t choose more than one strategy to be
inactive.
Another equivalent set of strategies is the sum r and difference u ,
leaving all other variables unchanged:
280 Geometry, Language and Strategy—Vol. 2
r 1
2 r1 r2
. (7.18)
2 2
u 1 r r1
In this case the resultant payoffs now have a form that is consistent with
taking three of the strategies inactive.
With the coordinate basis order u r s1 s2 t , we have for
player 1 the initial payoff matrix:
0 45 1
10 2 9 10 2 0
4
5 0 0 0 0
F 1
cab 9 10 110 2 0 0 0
0 . (7.19)
ab
9 2 0 0 0 0
10
0 0 0 0 0
0 4
5
9
10 2 110 2 0
4
5 0 0 0 0
2
F ab cab 9 10 9 10 2 0 0 0
0 . (7.20)
1 2 0 0 0 0
10
0 0 0 0 0
To recover the original form of the payoffs for the prisoner’s dilemma
we transform the constant value matrix Eq. (7.11):
0 2v 0 0 2v
0
2v 0 0 0 0
c v 0 0 0 0 0 . (7.21)
0 0 0 0 0
2v 0 0 0 0
0
Prisoner’s Dilemma: A Code of Conduct Application 281
0 z w 12 x y z w 12 x y z w
0
z 0 w 12 x y z w 12 x y z w
0
w 2x y z w 12 x y z 0 y w 0 . (7.22)
1
w 1 x y z w 12 x y z y 0 w 0
2
w 0 w 0 w w 0
0 0
The transformed payoff is the most general form in which the three
strategies r s1 s2 can be inactive:
0 2 w x y 2 z 2 2w
0
2w x 0 0 0 0
y 2 0 0 0 0 . (7.23)
z 2 0 0 0 0
2w 0 0 0 0 0
Despite the fact that there are no active payoffs, there remain non-trivial
dynamics that are a consequence of the cooperative and inertial forces. A
consequence of the expanded scope of decision process theory is that we
specify more behaviors than just payoffs and active strategies. We have
introduced inactive player strategies as the mechanism that underlies the
notion of a player and payoffs. In addition there are orientation potentials
whose existence and behaviors follow from decision process theory. We
set these additional initial behaviors of the prisoner’s dilemma for the
player fixed frame model with a single active strategy.
We keep in mind the following as we proceed. First, we are looking
for solutions to the prisoner’s dilemma that have three additional
isometries. We are thus justified in calling such solutions symmetric.
These isometries are associated with a code of conduct the players agree
to adhere to. We are also justified in thinking of such solutions as just.
General solutions to the prisoner’s dilemma have much in common with
the utilitarianism school in ethics [Tavani, 2011], whereas symmetric
solutions have more in common with just utilitarianism.
Second, we must be mindful of what we mean by initial conditions.
Solutions to partial differential equations have unique solutions when the
form of the solution is specified on some initial boundary or bounding
surface. It may be at an initial point in time or space. If our intent was to
predict the future based on a known state of the prisoner’s knowledge
about each other and the situation, then the initial conditions would be
based on time. Our goal here is a little different in that we want to
284 Geometry, Language and Strategy—Vol. 2
dx
V . (7.26)
d
We have normalized the flow to unit length, Eq. (7.9) and Eq. (7.25):
dx dx
ˆ 1. (7.27)
d d
Based on an initial flow direction, there is a natural set of orthonormal
vectors that can be constructed. The set aligns proper-time along the flow
direction and constructs vectors transverse to the flow using the Gram-
Schmidt orthogonalization process, starting with the initial strategy
directions.
We sketch this process. The first step is to take E a V a and
E j V j . We can then choose the remaining vectors in any order, which
we take to be along 1 2 s1 s2 r u . These are independent of the
flow, though not in general transverse to the flow. We take a unit
direction along the first strategy 1 , which defines the direction U1 . We
construct a new orthonormal vector that is both orthogonal to the flow
and of unit length:
U 1 U 1 E E
E 1 . (7.28)
ˆ U 1 U 1E E U 1 U 1 E E
We go through each vector and use the same algorithm. Each new vector
must be orthogonal to all previous vectors and have unit length. This
process can be done in any frame of reference.
The last strategy will be the active relative player effort strategy, u .
In the symmetric normal-form coordinate frame, this strategy is
286 Geometry, Language and Strategy—Vol. 2
If there is a non-zero player interest , the player will develop a non-
zero coupling away from equilibrium. We think this provides support for
the streamline solution choice that the electric field is zero, Eq. (6.15).
Given the initial flow directions and the values of the complete
symmetric co-moving orthonormal set at some initial position, we use
Eq. (6.46) to determine the symmetric co-moving orthonormal frame at
all other positions. These equations provide coupled differential
equations in the proper relative player effort and can be uniquely
solved once the coefficients in the equations are determined. The
coefficients are the orientation potentials, which we deal with next.
p
q q q q 2 p ,
n 1
e q e 2 e ,
p 0 . (7.30)
We make model assumptions about the inertial components p and the
relationship between energy density and pressure p (Sec. 7.10). To
complete our specification of the numerical problem for the field
equations, we determine the initial values of the player interest and
player bonding components. These we show are related to the payoff
fields.
To relate the initial values of these orientation potentials to payoffs,
we start with the normalized flow Eq. (7.27). We note that this holds in
any frame of reference and provides the basis for discussing the
prisoner’s dilemma in the normal-form coordinate basis, the symmetric
normal-form coordinate basis, the holonomic basis or the symmetric co-
moving orthonormal coordinate basis:
ˆ jk jk
ˆ ja jk Aak . (7.31)
ˆab g ab jk Aaj Abk
The metric potentials ˆ in the holonomic basis tie these different
descriptions together. So for example, we have on the left the metric
potentials in the holonomic basis, whereas on the right we have the
potentials that describe behaviors in the normal-norm coordinate basis.
In addition to the strategic flows V a , we know the initial behaviors of
the payoffs, which are related to the holonomic metric potentials.
288 Geometry, Language and Strategy—Vol. 2
Assuming the transformed basis from Sec. 7.5 and further assuming that
the metric potentials are independent of all coordinates except u t ,
we relate the metric potentials to the normal-form coordinate basis
potentials:
j , k 1 , 2
m r , s1 , s2
a, b u , t . (7.32)
aˆ jm a jk A jk a A a jk A jk F
k
m
k
m
k
m
k
am
u ˆ jm m jk F um
k
t ˆ jm m jk F tm
k
0
. (7.33)
t ˆ ju 0
u ˆ jt m jk F utk
specified at all, such as uˆ jk . They represent influences that impact the
dynamic behaviors but are absent from game theory models. We must
get better insight into such behaviors.
For the symmetric normal-form coordinate basis, we have the same
holonomic metric potentials ˆ , which are expressed in a form similar
to Eq. (7.31), with the noted differences in what is active and what is
inactive:
a, b u , t
j , k 1 , 2 , s1 , s2 , r
ˆ jk jk . (7.34)
ˆ ja jk Ak a
ˆab g ab jk A j a Ak b
j , k 1 , 2
m r , s1 , s2
u jm m jk F um
k
(7.35)
t jm m jk F tm
k
0
t ˆ ju jk t Ak u 0
u ˆ jt jk F k ut jm F m ut m jk F utk F k ut F utk
We take the initial behaviors that we have determined and compute the
consequences in the symmetric co-moving orthonormal coordinate basis.
290 Geometry, Language and Strategy—Vol. 2
We start with the electric field f j components from Eq. (6.46), which
are determined by the game values:
f j
E t E u E u E t F tuj
. (7.36)
f j
2 q E j E j e E j e E j 1 E k Ek
To this we add the behavior for jm that can be derived from Table 5.1
(Cf. Sec. 4.11, Ex. 23, Eq. (4.85) specialized to the streamline solution),
which are set by the initial payoffs for the prisoner’s dilemma:
jm m jk E u F um
k
. (7.37)
jm 2q E j Em 2 E j Em E j Em 2 E j Em
j 1 , 2
j
. (7.38)
1
2 F tuj
j , k 1 , 2
m r , s1 , s2 . (7.39)
1 2 F Em 1 2 m jk F
u j m
j
tu
k
um
j, k 1 , 2
. (7.40)
u jk 2u j k
t 1 2 N2 C2 N1 C1
V . (7.42)
1.061 0. 0. 0.2500 0. 0.2500 0.
t 1 2 s1 s2 r u
V . (7.43)
1.061 0. 0. 0.1768 0.1768 0.2500 0.
With these values, for the inactive space 1 2 s1 s2 r the player
interest and player bonding components are chosen to agree with the
game value Eq. (7.36) and player payoffs Eq. (7.37) using a 1 3 ,
which leads to a decision mass of m 4 and hence 0 4.243 :
We arbitrarily set all the remaining player interest and player bonding
components to zero with the exception of the diagonal components
labeled x1 x2 x3 . We distinguish between prisoner 1, prisoner 2 and
the remaining inactive dimension. The pressure component Eq. (7.41),
with units 1 , is a function of three unknowns (Fig. 7.1):
It is clear that we must choose values of these unknowns that are large
enough or the pressure will be negative.
Prisoner’s Dilemma: A Code of Conduct Application 293
There are two disconnected surfaces, which are illustrated when all
three parameters are equal, x1 x2 x3 (Fig. 7.1). For our numerical
work, we pick a pressure component p 1 , which occurs when all
three parameters are equal to x1 0.4732 .
We can pick a value in which the pressure is uniformly zero.
However the gradient of the acceleration q does not vanish since it is
equal to the vorticity squared, Eq. (7.30). A non-zero inertial acceleration
implies the existence of an inertial stress or pressure. The pressure
changes value at other strategic points.
p. (7.47)
7.11 Outcomes
The student will have learned that the prisoner’s dilemma demonstrates
that a player code of conduct can be imposed in decision process theory
and such hypothesis leads to a “stable” set of solutions. The theory thus
extends the notion of solutions based on Nash equilibrium in game
theory, which provide another class of “stable” solutions. By adding
three codes of conduct to the prisoner’s dilemma, the player fixed frame
model for the prisoner’s behaviors reduces to a single active strategy.
Complete numerical solutions to these models are obtained that represent
both steady state and non-steady state dynamic behaviors. These
Prisoner’s Dilemma: A Code of Conduct Application 295
7.12 Exercises
t 1 2 s1 s2 r1 r2
t 1 0 0 0 0 0 0
1 0 1 0 0 0 0 0
2 0 0 1 0 0 0 0
S . (7.48)
N2 0 0 0 0 1
2
0 1
2
C2 0 0 0 0 1
2
0 1
2
N 0 0 0 1 0 1 0
1 2 2
C 0 0 0 1 0 1 0
1 2 2
(2) Show that the new basis coordinates x a and the new payoff fields
F abj are determined by the transformation S in Ex. 1, where for the
payoff fields we restrict the transformation to exclude the player
inactive strategies:
Prisoner’s Dilemma: A Code of Conduct Application 297
x S -1x
. (7.49)
F j S -1F jS
(3) Show that the transformation from the basis of Eq. (7.17) to that of
Eq. (7.18) is accomplished through the orthogonal transformation
T below, TT T I , with the transformations of the coordinates and
payoffs being expressed analogously to Eq. (7.49):
t 1 2 s1 s2 r u
t 1 0 0 0 0 0 0
1 0 1 0 0 0 0 0
2 0 0 1 0 0 0 0
T . (7.50)
s1 0 0 0 0 0 0 0
s2 0 0 0 0 0 0 0
r1 0 0 0 0 0 1 1 2
2
r 0 0 0 0 0 1 1
2 2 2
(4) In this section, we have chosen the code of conduct based on the
differences between the strategy to confess and the strategy to not
confess. Rewrite the model and find the appropriate transformations
if the code of conduct is to pick as inactive the strategy to confess
for player 1 and the corresponding strategy for player 2.
(5) Show that a model equivalent to the prisoner’s dilemma is that of
two players who have the choice to honor a contract or break the
contract. This scenario is significantly more general than the
prisoner’s dilemma and underlies the invisible hand of Smith
[1776].
Chapter 8
Prisoner’s Dilemma—Steady
State Geometry
299
300 Geometry, Language and Strategy—Vol. 2
q 2 e e e
Q .
1 e e
2
This shows that the steady state geometry determines the forces that
change the active decision flow. The four fields that govern that behavior
are: the player engagement e , the inertial acceleration q , the player
interest and the player bonding . The centrally active
acceleration Q vanishes only if these contributions conspire to cancel.
In this chapter we project how these behaviors impact decision making,
as well as introduce a number of related concepts that provide further
depth of understanding for these concepts.
In addition to the decision theory aspects of these concepts, it is
helpful to introduce more mechanical aspects of these concepts that
stress the analogy of this theory to physical theories. In material objects,
we are familiar with the idea that the stresses and strains of matter are
related to dynamic behavior. We expect different vibration
characteristics between a very brittle object and a very ductile object;
different transient behaviors, for example.
Typically stresses are proportional to strains as a first approximation.
Stresses and strains are also familiar in electromagnetic phenomena,
though they go by different names. In this case we associate currents
with stresses and strains with electric fields. The relationship between the
two is expressed as Ohm’s Law: the current is proportional to the voltage
(related to the electric field). In decision process theory, which has many
similarities to electromagnetic theory, we also have relationships
between stresses and strains, though the steady state geometry
relationships are more complex.
We explore these aspects of steady state geometry for the prisoner’s
dilemma. In Sec. 8.1, we focus on the stresses, which manifest as inertial
pressure and provide the input for computing the accelerations. We build
off the idea of the organizational system from Vol. 1. We show model
results in Sec. 8.2, with sensitivity analysis provided in Sec. 8.3. The
geometry is further characterized by the strengths of the codes of
conduct, expressed in terms of the orientation potentials, which can also
be characterized as strains. These behaviors are described in Secs. 8.4-
302 Geometry, Language and Strategy—Vol. 2
The general decision process theory Table 5.6 identifies in Eq. (3.42),
the change in inertial acceleration q , which is also the acceleration
304 Geometry, Language and Strategy—Vol. 2
q q . (8.1)
(Sec. 9.3), we shall see that the charge density (player engagement)
determines the initial shape. The charge density is indirectly determined
by the inertial effects. There are many types of effects, clearly indicating
that the inertial effects are not scalar but tensor forces.
At this point we observe that in the co-moving orthonormal frame,
the spatial components of the flow are zero, which leads us to Eq. (8.1).
This equation and the shape of the potential well, Fig. 8.3, support our
argument that inertial effects induce stability. We believe this is a crucial
aspect of decision making when there is a non-trivial code of conduct.
Decisions will collect together; where
they collect creates an equilibrium
point. Because they collect they attract
subsequent decisions towards the same
point in a way analogous to
gravitational attraction. This
equilibrium point is determined by the
code of conduct, not by considerations
Figure 8.3. Inertial potential q of utility, self-interest, or even public-
interest.
In decision process theory, the attribute that the pressure bunches
around the origin does not hold in general. For these symmetric solutions
of the prisoner’s dilemma, it holds for initial pressures down to a critical
pressure pc 0.02693 defined as the point at which the acceleration
gradient vanishes, which is proportional to the scalar formed from the
player interests, .
Below this pressure the equilibrium point moves away from the
symmetric center. The cause is traced to the presence of the centrifugal
effect of . In the next section, we investigate this in more detail, since
this provides further insight into the source of stable behaviors.
We consider several cases that modify the PDM (Sec. 8.2), each with an
initial pressure of 0.01, a value that is below the critical pressure. We
vary the input flow keeping the inertia fixed at the model value of 4. We
306 Geometry, Language and Strategy—Vol. 2
call the flow in which both prisoners choose not to confess the common
good solution. The Nash equilibrium flow where both prisoners confess
we call the Nash solution. We consider these two solutions, with and
without the non-zero game value identified in Sec. 7.5.
For the Nash solution, the shape of the pressure for zero game value,
Fig. 8.4, has a peak at the origin. This demonstrates that nothing prevents
the bunching of the energy density and pressure around zero, because the
Nash solution has player interests that are almost identically zero
when there is no game value. We find that the Nash solution is
structurally stable for zero game value in the simple sense of attraction.
Figure 8.4. Pressure for Nash solution: Figure 8.5. Pressure for common good
zero game value solution: zero game value
In contrast, the pressure for the common good solution, Eq. (7.42),
has a minimum at the origin, Fig. 8.5. It demonstrates that with the same
assumptions and no game value, the player interests lead to a stable
solution based on the pressure gradient. Again, extra energy is expended
to insure the steady state behavior. The corresponding potential well
Fig. 8.3 turns into a potential mountain. We conclude that there is not
enough inertial mass to balance out the centrifugal effect ascribed to the
game, Fig. 8.6, though there is sufficient pressure gradient. We thus find
evidence that this solution is not structurally stable.
Prisoner’s Dilemma—Steady State Geometry 307
We thus agree with the assessment that the common good solution is
possibly less structurally stable in the absence of inertial effects.
However, we showed that with sufficient initial energy density, the
instability can be overcome, Fig. 8.1. The source of the sign change is
clearly seen in Eq. (7.7). Whenever there is a player interest, the initial
pressure has to be large enough to overcome the effects or the origin will
be unstable.
Figure 8.6. Common good solution for Figure 8.7. Nash solution for
1 : zero game value 1 : non-zero game value
h
1
ni h . (8.2)
ni h
The bond shear components leave the volume unchanged but distort the
shape of the cube by sliding a face of the cube parallel to one of the other
faces in such a way that the volume does not change. It may help to think
of a cube of Jell-O as something with a fixed volume and consider
different distortions to this. These ideas give a physical picture of the
strategic strains on a small cube that is moved along an active strategy
direction in the co-moving orthonormal frame.
In the absence of player interests, such as we saw above for the Nash
solution with no game value, the player bonding depends only on the
energy density, acceleration and stress. For the numerical solution of the
Prisoner’s Dilemma—Steady State Geometry 311
prisoner’s dilemma, we start with the initial conditions, Eqs. (7.36) and
(7.37). This gives the value 0 2.366... . The shape of the player
bonding (Fig. 8.8) is modified only slightly by the player interests.
The player bonding gives direct evidence of changes in the
configuration as a consequence of the energy density and active stress
component. If we move a unit cube in the five inactive dimensions along
the active strategy, we find that the volume changes as it moves left or
right (Fig. 8.9). In particular we see that the size of the box goes to zero
as we go further left in the direction in which prisoner 1 is more active.
As we go to the right, the size of the box becomes constant. The
asymmetry of the volume corresponds to the asymmetry noted earlier
about the shape of the pressure (Sec. 8.2). For our numerical example,
we have chosen to focus on solutions in which prisoner 2 is typically
more active.
The idea that the unit of measure changes depending on where we
are, illustrates that in decision process theory, the concept of space and
its ordinal properties are distinct from the concept of distance and its
cardinal properties.
Figure 8.8. Player bonding gradient Figure 8.9. Player bonding volume as
(PDM) a function of strategy (PDM)
The initial sources of strain are the initial payoffs, Eq. (7.39). In addition
to the player bonding, these strains generate what we characterize as
player bonding shear in which the shape changes without rotation but the
312 Geometry, Language and Strategy—Vol. 2
e q
. (8.5)
e q 2 2
ni h p 1
ni h p
The form factor e q , (Fig. 8.10), depends only on the inertial terms. For
the case of the numerical example with a perfect fluid, the reduced shear
simplifies:
The reduced shear depends on the inertial factors through the form factor
e q . In particular, when the player interests vanish, the reduced shear is
constant.
Figure 8.10. Form factor e q with Figure 8.11. Reduced bond shear
initial value normalized to unity (PDM) 1q2 (PDM)
One measure of the shear effects is . It has a maximum at the
origin and is roughly constant over the whole range. In addition, shear
effects lead to changes in the payoff tensor. In other words, due to
successive plays of the decision process, each prisoner adjusts his choice.
For steady state behaviors, this will manifest itself as a spatial
dependence. For the values chosen here for the prisoner’s dilemma, the
changes are primarily those due to the change in the form factor,
Fig. 8.10, as opposed to changes in the value of the reduced shear, such
as Fig. 8.11.
There are 14 independent bond
shear components, of which six we
have identified as being related to the
payoffs of the prisoner’s dilemma.
That means there are eight
components that measure something
new, something that directly relates to
the strain configurations that result Figure 8.12. Reduced bond shear
(PDM)
from the assumed stresses. 1 2
314 Geometry, Language and Strategy—Vol. 2
For example, consider the mixed component 12 between the two
prisoners, Fig. 8.12. We think of the prisoners as interacting only by
means of their respective payoffs. However, decision process theory
provides influence through other means, such as through the bond shear
components 12 , Eq. (8.8). Though zero at zero relative player
effort, it shows a small but non-zero effect at other values of relative
player effort.
jk E j Ek E j h Ek . (8.9)
Figure 8.13. Player interest field Figure 8.14. Electric field f j
in the co-moving frame in the co-moving frame (PDM).
(PDM).
In Eq. (7.38), we argued that the game value sets the size of the player
interest. More precisely, the game value sets the size of the electric field
components f j , Eq. (7.36):
f j
E t E u E u E t F tuj
.
f j
2 q E j E j e E j e E j 1 E k Ek
f j
0.4242..., 0.4242...,0.1882...,0.1882...,0.2662... . (8.10)
We call player interests since from Table 5.1, Eq. (4.69), they
determine the behaviors of the player engagement e , Fig. 8.15. The
player engagements determine the mixed metric attributes, Eq. (4.51)
that determines the relationship E j e E j between persistency
components, Eq. (4.54). This relationship shows that the player
engagements are persistency attributes and demonstrates that the initial
values of the player engagements are set by the initial values of the
orthonormal set. Our choice of zero flow for the two prisoner directions
translates to their corresponding player engagements being zero at the
origin. The other player engagements are not zero, as seen in Fig. 8.15.
The player engagements also contribute to the active metric, Eq. (4.72),
which we deal with in Chap. 9.
318 Geometry, Language and Strategy—Vol. 2
In Sec. 8.5 we showed that the player bonding volume decreased as the
active strategy becomes negative in the co-moving frame. We can
show that this is also true in the normal form coordinate basis by looking
at the behavior of the diagonal components of the inactive metric
components jj using Eq. (8.9), Fig. 8.16. As before, j has the
directions 1 2 s1 s2 r . We see that the diagonal components are
all tending towards zero as we move left into the region that prisoner 1
becomes more active. The initial conditions were set assuming that the
co-moving frame and normal form frame were aligned at the origin.
The gradient of the metric potential determines the off-diagonal initial
values of the bond strain, Eq. (7.39) in terms of the known payoff values
of the prisoner’s dilemma:
0.9465... 0. 0.1414... 1.2727... 1.
0. 0.9465... 1.2727... 0.1414... 1.
jk 0.1414... 1.2727... 0.9761... 0.02958... 0.04183... . (8.11)
1.2727... 0.1414... 0.02958... 0.9761... 0.04183...
1. 1. 0.04183... 0.04183... 1.005733...
Figure 8.19. Prisoner 1 inactive Killing Figure 8.20. Prisoner 2 inactive Killing
vector components E1 (PDM) vector components E2 (PDM)
The behaviors of the metric potential are set by the persistent frame
transformation components in Eq. (8.9). We gain further insight by
looking in more detail at the behaviors of these persistency components.
The flow components V j E j as seen in the normal form coordinate
320 Geometry, Language and Strategy—Vol. 2
basis are determined, Table 5.1, Eq. (4.64), by the acceleration and the
player interests.
E j q E j 2 E j .
E j E j .
E j E0 j e .
If the charges are small, the second term dominates and the geometry is
determined by the diagonal components of the player bonding. If the
player bonding shear is small, the components are all equal and we
expect the diagonal components of the metric to be approximately equal
to 2 ni , which is obtained by solving . For the numerical
model, we provide the solution, Fig. 8.21. If we scale the curve by 2 5 ,
we see the qualitative behavior of Fig. 8.16. It is not exactly true since
we have ignored the effects of the charges E j .
We have already seen how the
player bonding effect decreases the
size of the transformation cube as we
go to the left. Some evidence of this
behavior is also seen in the
components E1 for prisoner 1 in
Fig. 8.19, where the size of the proper
direction for each player decreased in
Figure 8.21. Player bonding
the co-moving frame. Similarly in Fig.
(PDM)
8.20, we have the corresponding
components for prisoner 2. In addition to the player bonding effects, the
shear effects impose behaviors along the other directions.
Based on Fig. 8.21, we see that whenever gets large and negative
(the player bonding potential is large and positive), we enter a regime
in which the diagonal components of the inactive metric jj goes to zero.
322 Geometry, Language and Strategy—Vol. 2
These expectations are met from the full solution, Fig. 8.16. So, as
proposed in Sec. 8.8.3, if we think of these contributions as measuring
the utility between separated points, then the same utility occurs to points
with larger and larger spacing, which suggests the concept of inelasticity
from economics. Conversely, in the other direction the player bonding
changes sign and levels off, suggesting that the measure of utility has a
limiting value and depends directly on the point spacing. We think this
makes some sense if you interpret this to mean that the more players
bond, the less players will feel the urge to compete.
A second point worth noting is that the player bonding exists not only
between different players, but also for the same player as discussed
above. For the numerical example under consideration, there is not much
variation within or between different players. This is an artifact of the
initial values chosen. There are situations where these values are quite
different. Moreover, in general the player bonding will be different for
different strategic directions: in this model we have only a single active
direction. Thus decision makers may make different choices and assign
different utilities to their choices depending on where they are at in the
strategic space. We see evidence for this in Fig. 8.17.
det jk jk . (8.12)
Prisoner’s Dilemma—Steady State Geometry 323
This shows the same qualitative behavior, Fig. 8.22. The invariant
volume element is proportional to this
determinant. Since this determinant
goes to zero as we go left, it confirms
that the yardsticks in the inactive
space shrink in that direction.
It also shows that the invariant
volume shrinks if we go sufficiently
far to the right. This is not so obvious
Figure 8.22. Determinant of the from the behaviors that we have seen.
inactive metric jk (PDM) A more detailed analysis shows that
one of the eigenvalues of the metric is
tending towards zero. Furthermore, the stresses are going to zero as we
go far to the right.
8.9 Outcomes
q 2 e e e
Q . (8.13)
1 e e
2
From Sec. 8.4, the student should see that strains, which are the
consequences of the inertial stresses, determine the player bonding
. In the steady state hypothesis, these strains are steady state and
are determined from the known conditions. Player bonding
components and player interests result from payoffs and equilibrium
strategies that are not Nash equilibriums. Such components generate
inertial attractive effects similar to gravity as well as repulsive effects
due to pressure gradients. Shear effects are present as well, which in
the symmetric prisoner’s dilemma are attributes of the initial payoffs.
From Sec. 8.5, the student should see from the numerical results that a
given solution is not symmetric between the two players. For the case
we considered, prisoner 1 ( 0 ) is more constrained than prisoner 2.
The player bonding gradient is larger in magnitude on the left as well.
From Sec. 8.6, the student should see that the player bonding shear
effects are sensitive to the inertial stresses. For a perfect fluid, the
contributions are from the player interests.
From Sec. 8.7, the student should see that the player interests can be
quite large. When they are large, they drive the player engagement
gradients, moving zero engagement to non-zero engagements, such as
in Fig. 8.13.
From Sec. 8.8, the student will be able to trace the effects of the
stresses and strains in the theory to the persistency components
E j E j E j . The student will understand the mechanism that
generates these important new effects in decision process theory that
payoff values must change as a function of strategy. The student will
also understand the mechanism for player engagement or coupling
that determines the size of game theory effects in the theory. A result
that has possibly wider applicability is that an aggressive player
displays a greedy charge whereas the other player has an
accommodating charge. From Sec. 8.8.6 the student should see that
the player bonding directly determines the inactive metric jk in the
centrally co-moving frame, which is not changed if viewed in the
normal form coordinate frame. The student should see that the player
bonding potential determines the utility for player cooperation.
326 Geometry, Language and Strategy—Vol. 2
8.10 Exercises
E j
2 E j
E j
e f j
. (8.14)
f j
2 q E j E j e E j e E j 1 E k
Ek
(2) Show that for the general case, the player bonding component ,
Eq. (8.2), satisfies the following equation:
p
q ni p ni
n 1 . (8.15)
p 2 2
(3) Show that for the general case, the bond shear components ,
(8.2), satisfy the following equation:
(4) Show that for the streamline solutions of Chap. 6, the bond shear
and player bonding can be written in terms of potentials as follows:
(5) Using the substitution defining the reduced bond shear potential
below, show that the equation for the reduced bond shear is
given below for the streamline solutions of Chap. 6. In particular,
you see that when the reduced pressure is proportional to the
Prisoner’s Dilemma—Steady State Geometry 327
identity, h for any scalar field , then the source term on
the right is zero. What can you conclude about the reduced bond
shear? Show that this implies that the reduced bond shear matrix
is constant for a single strategy model.
q 1
ni h
e q . (8.18)
e q 1
ni h
Chapter 9
329
330 Geometry, Language and Strategy—Vol. 2
eq
. (9.1)
We indicate the centrally co-moving frame by a “bar” over the
coordinate variable. This transformation leads to coordinates that are in
the stationary line gauge Eq. (6.98), which can be transformed
diffeomorphically into the harmonic gauge using Eq. (6.81), with
x a , U 0 V a 0 and constant V a 0 . The equations determine U 0
with boundary conditions consistent with the centrally co-moving frame.
Prisoner’s Dilemma: Determinism and Chaos 331
In the centrally co-moving frame for the general case of more than
one active strategy, there will be two additional steady state geometry
tensors: the player payoffs F j related to the orthonormal orientation
potentials ; and the seasonal payoff f related to the orthonormal
potentials . These are all zero when there is only a single active
strategy.
There are three main categories of behaviors that we expect from our
shore-side perspective: travelling wave or harmonic behaviors as a
consequence of an initial harmonic force; resonant behaviors that are
harmonic and persist in the absence of an initial time dependent force;
and transient behaviors that die out in time, leaving one of the other
previous behaviors in place. Implicitly we considered harmonic
behaviors in Vol. 1, which are not expected here because there is a single
active strategy. We ignore for now the transient behaviors, so that in this
chapter we focus on the harmonic behaviors, which extend our notion of
steady state geometry1.
For a single active strategy, the behaviors are specified by Eq. (6.44)
in terms of harmonic polynomials, or equivalently by the central frame
wave equation that was derived in Sec. 6.8, Eq. (6.52) for any single
harmonic frequency for the coordinate x a x a ei :
x a 2 e 2 q 1 e e x a q x a 0 . (9.2)
Note that we need not assume that the harmonics are the same for each
coordinate, nor need we assume that there is a single harmonic: each
harmonic obeys an equation of the same form. For simplicity however,
we start with a single harmonic that is the same for all coordinates.
We see the steady state geometry contributions to the wave equation.
We see the effects of the player engagements, player bonding and
inertial acceleration in the resultant dynamic behaviors. In addition, we
should see that the initial conditions in effect generate a time dependent
source. It is for this reason that we argue that the resultant behaviors are
1
This is similar in concept to the way in which steady state AC behaviors extend the
concept of DC behaviors: the behaviors are independent of time, with the dynamics
captured by the specification of the driving frequency.
332 Geometry, Language and Strategy—Vol. 2
harmonic based on a time dependent driving force. Not only are the
flows dynamic in the sense that the steady state flows may exhibit
accelerations, they are dynamic in the sense that in the normal-form
coordinate basis, they explicitly depend on time.
As an introduction, we start with a general discussion of determinism
and chaos, Sec. 9.1. In Sec. 9.2 we focus on the flow of time as seen in
theory, which sets the stage for non-linear time dependences with
equations that appear linear. We provide evidence of a gravitational field
whose origin is from the inertial acceleration and from the player
engagements. We follow this with the behaviors of the strategy flows,
Sec. 9.3 showing that the harmonic time behavior generates a harmonic
spatial behavior. This provides evidence for behavior similar to other
physical systems. In Sec. 9.4, we compute the time and distance contours
from knowledge of the values of the active metric components g ab . To
explore possible sources of chaos and other interesting behaviors, we set
up a sensitivity analysis in Sec. 9.5 based on what the players consider to
be at stake in the decision process. We carry out this analysis in Sec. 9.6
for the baseline, Sec. 9.7 for low and high stakes and in Sec. 9.9 for the
normal form coordinate basis perspective. In Sec. 9.9, we examine
recurrence plots both in time and in strategy space in search of evidence
for chaotic behaviors in steady state phenomena.
0.05
t t , a t bt sin N
t , a t
. (9.3)
0.05
u u , a b
u u
sin N
u , a u
2
It may sound strange to talk of rotations. Yet payoffs, because they are anti-symmetric
tensors, specify and generate rotations. Rotations thus play an important role.
Prisoner’s Dilemma: Determinism and Chaos 339
g ab 1 e e E a E b E a E b
. (9.4)
gab 1 e e Ea Eb Ea Eb
g 1 e e e2 q
g 0 . (9.5)
g 1
e 2 q
g . (9.6)
1 e e
This is determined from the player engagements, Fig. 8.15 and the
inertial acceleration Fig. 8.3, both of which contribute potential wells
centered at zero. Thus the time bonding has a minimum where the inertia
is greatest and where the acceleration is zero, Fig. 9.1. This is where we
expect time to move more slowly.
We expect that in areas with
large inertia, any quantitative
measure of time slows down. Not all
parts of the topology can be reached
from a given starting point:
boundaries will be set by areas
where the components of the metric
field vanish. We can show that
steady state wave solutions exist,
Figure 9.1. Time Bonding g (PDM)
exhibiting curvatures determined by
these boundaries. See the exercises in Sec. 9.11.
340 Geometry, Language and Strategy—Vol. 2
The travelling wave behaviors are also visible in the contour plot for
the frame transformation E u , Fig. 9.5. Here we see the distortions that
reflect the difference between our wave equation, Eq. (9.2) and the usual
wave equation. We see that the null lines between the peaks and troughs
are not at right angles. Furthermore they bend at the boundaries at the
left, determined by the vanishing of the inactive metric components. The
direction of the travelling wave also bends due to the same boundary.
If we think of the peaks as being places where inertia collects, then the
motion of the inertial peak is the movement of energy density. This
motion should be less than the maximum speed possible and should form
an angle of greater than 45 degrees.
We have determined the time bonding gtt , which is formed from the
inverse of the metric components g ab . The result, Fig. 9.6, shows the
inertial acceleration and player engagement effects, which are both
independent of central time. On top of these two static effects is a ripple
in time due to the single harmonic. Because of the ripple, there will be
local minima in which inertia can be trapped. Except for the fact that we
might have too few dimensions (Cf. Sec. 6.7), we interpret the effect as a
gravity wave, since it is a wave in the metric. A more accurate
interpretation might be that it is a reflection of a frame-wave, Sec. 6.6.
We suggest that the gravity-source has the travelling wave behavior that
is reflected in the metric.
344 Geometry, Language and Strategy—Vol. 2
We see structure
in addition to the big
picture behavior of
the trapping field
formed from the
inertial acceleration
and player
engagements. Not
only is our notion of
time influenced by
the frame, so is our
notion of distance
along the strategic
Figure 9.6. Time bonding gtt , (PDM)
direction, Fig. 9.7,
where the main components come from the transformation E u and a
somewhat smaller contribution from E t .
As with the time flow, there is a static behavior on which there is a
small ripple effect due to the harmonic contribution. It is significant that
the wave properties one sees in physics are reflected here as well:
harmonic behavior in time leads to harmonic behavior in strategy space;
harmonic behavior in space leads to harmonic behavior in time. We
conclude that a localized event at an initial point in time and space will
radiate outwards in both time and space.
d 2 g ab dx a dx b 0
g tt 2 gtu c g uu c 2 0 . (9.7)
2
g g g
c tu tu tt
g uu g uu g uu
Prisoner’s Dilemma: Determinism and Chaos 347
Table 9.1. Player Stakes and Decision Choices for Sensitivity Analysis
We consider different values for the game beta and (equal) player
stakes 1 2 , which we choose from Table 9.1. These choices
have a major impact on the time component gtt of the metric that
determines the behavior of time intervals, Eq. (9.6). In a qualitative way
Prisoner’s Dilemma: Determinism and Chaos 349
g tt 1 e e E
1 t 2
. (9.8)
9.6 Baseline
Figure 9.13. Low stakes and low contour plot for the
metric gtt ,
Prisoner’s Dilemma: Determinism and Chaos 351
The choice region for the decision process, the allowed space of
strategy values , is bounded below at approximately 0.48... .
Much below this the equations become singular. We can go much higher
that 1 , though it requires progressively more harmonic polynomial
terms to get reliable results, or we must use the alternate approach with
Eq. (9.2). We obtain a lattice picture that is truncated by the choice
region. As in our previous dynamic calculations, we see the effects of the
wave equation.
To widen the choice region, we go to the case with low stakes and low
, where we expect little contributions from the strain, little
contributions from the player engagements (based on small ) and large
contributions from the harmonic contribution. We get the expected result
that the player interests are almost zero, the player engagements are
small and the first factor in Eq. (9.8) is almost unity. The potential well
comes almost entirely from the second factor E t , , the harmonic
contribution; note that there is a negligible contribution from the initial
shape of E t ,0 . This is clearly displayed in the contour plot, Fig.
9.13. The choice region for the decision
process is expanded from the baseline
case.
The contour plot shows peaks (and
valleys) that move along lines of 1 ,
such as the two that start at the origin.
The light color indicates peaks and the
dark color indicates valleys. For this case
we replicate the harmonic wave equation
and phasor solutions. The underlying
lattice structure is seen in the flow E t ,
Figure 9.14. Low stakes and low Fig. 9.14. We see a hexagonal packing
contour plot for the flow pattern. The light color areas indicate the
E t , path taken along the minimum of the
potential.
352 Geometry, Language and Strategy—Vol. 2
An alternative but
not equivalent way to
narrow the choice
region is to consider
high stakes and low
. In this case we
modify the shear
components, which
influence the time
structure, the lattice
pattern and narrow
the choice region,
Fig. 9.15. As we
Figure 9.15. High stakes and low time bonding move away from the
gtt ,
origin along the
strategic axis, there
are progressively deeper potential well pockets, indicating a degree of
instability. The energy generating these pockets comes from the payoffs
and the assumption of high stakes. The higher stakes along with
harmonics generates the effect. If we remove the harmonic contribution
we don’t expect the same structure. The striking feature of the contour
plot is the clover leaf pattern, an unexpected consequence of the partial
differential equation. This clover leaf pattern is not as clear in the
contour for E t .
The final case we consider is high stakes and high , which has
solutions for a slightly narrower range, Fig. 9.16. For high stakes and
high , the choice region is a little smaller with the same structure as
high stakes and low , Fig. 9.15.
These are significant consequences and they are based on the size of
the payoffs. They are in addition to the effects based on the relative
values and the benefit to the players on the expected game values. In
addition to the inertia effects associated with the energy density and
pressure, there are consequences based on the characteristic or speed
of the flow, which is a type of inertia.
One reflects a tendency each player may have to favor certain
strategies. This we associate with the energy density and pressure. A
Prisoner’s Dilemma: Determinism and Chaos 353
high inertia suggests that the players are reluctant to change their style of
play. For example, we might consider a version of chess in which there is
a timer. This changes the speed at which the game is played. A very high
speed version of chess would be a high stakes game, which introduces a
very different aspect of the inertial fields.
We speculate that
the latter definition of
inertia is significant in
physical theories of
cosmology. Early eras
of the universe might
be characterized
primarily as radiation,
so that this would
correspond to a high
era. Things don’t clump
together. In contrast,
the current era would
be a low era.
Galaxies have formed;
Figure 9.16. High stakes and high time bonding there is a lot of
gtt , clumping. These
characteristics hold in
decision processes. Thus we believe the effects we have identified in this
section are essential. Both the speed and stakes play a role that has
not been accounted for in the standard game theory approaches. They
determine the choice region and structure of the decision process.
In this and the last section, we have provided the time dependent
dynamic behaviors we expect for the prisoner’s dilemma in the co-
moving coordinate basis. In the next section we return to the normal-
form coordinate basis and examine the dynamic behaviors we expect for
the payoffs.
354 Geometry, Language and Strategy—Vol. 2
Recall from Sec. 7.6 that initially, the flow represents the strategy
choice of each prisoner, Fig. 9.17. One possibility is that this flow is
constant. We have explored the dynamic possibilities in our baseline
Nash solutions by superposing a harmonic onto this constant behavior.
We see the assumption that each player starts with the strategy to
confess. The time strategy reflects the baseline value for 1 3 . We see
the code of conduct imposed that the flow to not confess is zero and so
these flows oscillate around zero.
The harmonic disturbance propagates outward in space, Fig. 9.18, in
all appearance like an electromagnetic wave, as a reflection of the frame-
Prisoner’s Dilemma: Determinism and Chaos 355
wave behavior. In this case we see that the choice region narrows
causing the effects of
the harmonics to
damp out.
Furthermore we see
that at both extremes,
the prisoner chooses
some amount of the
confess strategy. We
expect to see these
effects in the payoffs
and other strain
parameters. It is
Figure 9.18. Baseline confess choice for prisoner 1 insightful if we
consider the transformed version of the space, Eq. (7.17), with Eq. (7.18)
for the behaviors that are internal to the decision process theory.
In this basis we have the player interest contributions F j tu , which
can be determined from the co-moving frame field f j , Table 5.1,
Eq. (4.30) using the frame transformations:
j
f
F j tu
. (9.9)
E t E u E u E t
As expected, the steady state behavior of the numerator for the prisoner’s
dilemma model parameters, Fig. 8.14, is modified by the time
(proper) distances than in the former case. However, the physical size of
the system is smaller, Fig. 9.22. In some sense the system is more brittle
when the stakes are high.
Since the player interest components reflect and replace the concept
of game value or expected payoffs, the brittleness reflects the risk
inherent in such high stakes decision processes.
Our focus in this chapter up until now has been on steady state
harmonics with a single harmonic. This focus is in keeping with our
study of steady state behaviors and the implications of the steady state
geometry to standing wave behaviors. Of course we have anticipated that
there will be non-linear effects and that they would manifest themselves
as changes of the magnitudes and phases of the flow as functions of
position and time. We saw evidence of this in Sec. 9.6 with the time
bonding contours, Fig. 9.12.
In this section we return to the concepts of determinism and chaos,
Sec. 9.1 and look for evidence in our steady state behaviors. Chaos in
deterministic systems is usually studied by starting a system in a known
position and following the causal behavior that results. For decision
358 Geometry, Language and Strategy—Vol. 2
process theory with a single strategy, that means using Eq. (9.2) and
imposing boundary conditions at an initial time t 0 , specifying
such things as the energy flow for all values of the active strategy u .
We use the differential equation to generate positions and flows at all
later values of time and at all strategy values. A simple explanation of
chaotic behavior is that small changes in initial values lead to
qualitatively different behaviors later. This sensitivity to the initial
conditions, suggestive of chaotic behavior, results from specific non-
linear behaviors of the physical processes.
The equations are complex and we have not completed an analysis of
the partial differential equations viewed as the initial value problem
suggested above. We suggest that we may indeed have such specific non-
linear behaviors and that they may be investigated in an alternative
manner. We can focus on a single harmonic and look for sensitivity of
structures in time as we change positions in space.
To identify structures and potential chaotic behaviors in both theory
and experiment, we use recurrence plots. A number of authors see
encouraging results using such plots, such as the study of chaotic
behaviors for heart beats by Sabelli [2005] and chaotic effects in
quantum wave functions and the early universe by Thomas et al. [2010].
The idea is that for any quantity of interest that depends on time, f t ,
you compare the quantity at different times and look at the two-
dimensional plot subject to the condition: f t f t . For
stochastic processes, there should be no pattern at all, since the
assumption is that future events are unaffected by past events. For
periodic behaviors, the two-dimensional plot will also be periodic. For
chaotic structures however, in contrast to stochastic processes, the
two-dimensional plot is expected to have non-repeating structures. For
the strategy and time flows we create such plots, Figs. 9.23 and 9.24.
Prisoner’s Dilemma: Determinism and Chaos 359
9.10 Outcomes
subject’s potential energy per unit charge and that in the prisoner’s
dilemma model; it is closely related to the prisoner’s expected game
value payoff. These inertial and payoff mechanisms are new and specific
to the gauge theory framework adopted. We see a new mechanism, for
example, that suggests that when game values are large, the value itself
influences the game behavior.
By studying the numerical models, the student learns that the results
are based on a causal evolution of behavior starting from an initial
surface at zero proper-time or equivalently, an initial surface at zero
strategic distance (from a presumed equilibrium for example). This
demonstrates that payoffs are functions of strategies that are in general
different from that initial surface. The allowed ranges are determined by
the causality principle and the principle of possible change.
The student should appreciate that with the steady state hypothesis of
Chap. 6, all strategies can be active. The methods for solving these
equations may require different numerical techniques [Courant &
Hilbert, 1962]. Mechanical and electrical engineering disciplines solve
problems of a similar nature [Bhatti, 2005] and their techniques may be
applied to the partial differential equations Eq. (6.33). Multiple active
strategies introduce new attributes a from the commutation rules Eq.
(6.14). Nevertheless, the same techniques of analyzing the solutions in
terms of harmonic polynomials, Eq. (6.36) is possible using Mathematica
and the numerical method of lines [Wolfram, 1992].
The attainment of the outcomes of this chapter is facilitated by doing
the exercises in the following section. Based on this investment, the
student should achieve the more detailed outcomes below based on
section.
From the discussion in Sec. 9.1, the student will have seen that for
small amplitude behaviors, systems are often linear. This means that
the amplitude of the oscillation plays no role in the outcome.
However, for non-linear systems and sufficiently large amplitudes,
the amplitudes do play a role, leading us to speculate that
deterministic behaviors have chaotic consequences. The student will
have seen some evidence of this in the numerical examples.
The student will have seen the effects of time bonding in Sec. 9.2.
This concept follows from differential geometry; it identifies the time
362 Geometry, Language and Strategy—Vol. 2
9.11 Exercises
(1) The fastest signal between two points is along the active null
geodesic direction g ab dx a dx b 0 . For equally spaced surfaces of
Prisoner’s Dilemma: Determinism and Chaos 363
(2) Use Eq. (9.10) in the previous exercise to show that the requirement
of a non-zero velocity along the null geodesic requires g uu 0 .
This is the principle of possible change that means any physical
point must be able to move in an open region, even if that region is
very small.
(3) Show that for streamlines that have speeds close to the null-
geodesic speed, the causality principle narrows the size of the
space: one gets a picture not unlike the physics picture of the big
bang with a singularity at some finite point in the past in which
space has zero size.
(4) Identify numerically the speed of the null geodesics and draw it in
the proper distance—proper-time plane. In the physical space, this
corresponds to going along the diagonals of the contour mesh.
364 Geometry, Language and Strategy—Vol. 2
(5) As an example of Ex. 4, take the model of Sec. 9.6 having high
stakes and low and deduce that the vector direction u along
du transformed to the co-
moving frame is given by
E u E u , Fig. 9.25.
(7) Show that another way to examine the behavior of high stakes and
low in Ex. 5 is to look at the timeline and spaceline contours,
Fig. 9.27, associated with the time vector E t E t , (“vertical”
Prisoner’s Dilemma: Determinism and Chaos 365
t , , t , ,
E a , E a , .
Figure 9.35. Strategy flow recurrences Figure 9.36. Time flow recurrences for
for single harmonic PDM single harmonic PDM
(14) Show that with the two frequency
PDM, not only are there non-trivial recurrence relations in time
along a streamline, we can have non-trivial recurrence relations in
368 Geometry, Language and Strategy—Vol. 2
369
370 Geometry, Language and Strategy—Vol. 2
Crusoe model in this chapter, and a general set of two-person games that
we will deal with in Chaps. 14-18.
We start with the practical steps needed to solve decision process
problems at every level, at least for steady state solutions (those
analogous to AC circuits). The steps are:
a, i p c : d ,
a
di d p dc .
Thus for the prisoner’s dilemma with a single active strategy, we have
a, i p c : 1, 5 2 3 . If we try to enumerate the different
possible cases, we rely on the separation between active and inactive
strategies. We realize that a given decision process may change
depending on how the participants identify the codes of conduct. We also
recognize that there is a difference between the codes of conduct and the
inactive strategies associated with a given player: a player has sole
control of his or her inactive strategy, whereas a code of conduct is
controlled by multiple players. We suggest this notation as being a useful
way to decompose the decision process.
Ordinarily, we imagine that each player controls their own inactive
strategy. They have a role in controlling the various codes of conduct.
They may also control one or more of the active strategies. If a player
controls no active strategy, we call that player dependent. To make this
concept more precise, we may need to establish how that concept enters
the theoretical description. This is currently an open question.
We use the prisoner’s dilemma as an example of how this general
decomposition provides useful insight. In Sec. 6.8 we introduced the
code of conduct as a1 symmetric decision process in which each inactive
strategy that is part of the code of conduct is associated with a new
distinct player. Thus we start with the most general solution of the
prisoner’s dilemma in the normal-form coordinate basis and consider
four active and two inactive strategies, a, i p c : 4, 2 2 0 .
1
We use the word symmetric to indicate a technical aspect of the theory, which is the
existence of a frame of reference in the differential geometry that treats inactive and
active dimensions on the same footing. See for example the discussion in Sec. 2.11.
Robinson Crusoe Economics 373
4 0 4, 1 3, 2 2, 3 1, 4 0 .
optimizes his or her choice. It might be thought that the game has little
strategic interest.
We shall keep an open mind. As part of keeping an open mind, it is
valuable and instructive to revisit the story on which this example rests.
Note that we might not be as sympathetic today towards Robinson
Crusoe as earlier readers.
The usual economic interpretation is based on the story by Daniel
Defoe in which Robinson Crusoe is stranded on an island. Before he
meets and rescues Friday, he survives through his “own efforts” of
hunting and farming. Here’s the plot summary from Wikipedia, [2008]:
“Crusoe leaves England setting sail from the Queens Dock in Hull on a sea voyage
in September, 1651, against the wishes of his parents. After a tumultuous journey that
sees his ship wrecked by a vicious storm, his lust for the sea remains so strong that he
sets out to sea again. This journey too ends in disaster as the ship is taken over by Salé
pirates and Crusoe becomes the slave of a Moor. He manages to escape with a boat and is
befriended by the Captain of a Portuguese ship off the western coast of Africa. The ship
is en route to Brazil. There with the help of the captain, Crusoe becomes owner of a
plantation.
“He joins an expedition to bring slaves from Africa, but he is shipwrecked in a storm
about forty miles out to sea on an island near the mouth of the Orinoco River on
September 30, 1659. His companions all die; he fetches arms, tools and other supplies
from the ship before it breaks apart and sinks. He then gets battered by huge waves as he
struggles to make it to an unknown island. He proceeds to build a fenced-in habitation
and cave. He keeps a calendar by making marks in a wooden cross he builds. He hunts,
grows corn, learns to make pottery, raises goats, etc. He reads the Bible and suddenly
becomes religious, thanking God for his fate in which nothing is missing but society.
“He discovers native cannibals occasionally visit the island to kill and eat prisoners.
At first he plans to kill the savages for their abomination, but then he realizes that he has
no right to do so as the cannibals have not attacked him and do not knowingly commit a
crime. He dreams of capturing one or two servants by freeing some prisoners and indeed,
when a prisoner manages to escape, Crusoe helps him, naming his new companion
"Friday" after the day of the week he appeared and teaches him English and converts him
to Christianity.
“After another party of natives arrives to partake in a grisly feast, Crusoe and Friday
manage to kill most of the natives and save two of the prisoners. One is Friday's father
and the other is a Spaniard, who informs Crusoe that there are other Spaniards
shipwrecked on the mainland. A plan is devised where the Spaniard would return with
Friday's father to the mainland and bring back the others, build a ship and sail to a
Spanish port.
“Before the Spaniards return, an English ship appears; mutineers have taken control
of the ship and intend to maroon their former captain on the island. The captain and
Crusoe manage to retake the ship. They leave for England, leaving behind three of the
mutineers to fend for themselves and inform the Spaniards what happened. Crusoe leaves
the island on December 19, 1686.”
Robinson Crusoe Economics 375
In this story, Robinson Crusoe is evidently not alone since he has the
help of Friday and others. Moreover, the outcome for him is dictated as
much by the morality of the times as the economic situation. His
worldview and in particular his self or independence, plays a key role and
must be considered along with the payoff aspects of his decision
processes. There are also external and historical constraints that come
with the story, which are normally ignored in an economic discussion.
This simplest economic situation is therefore one in which there is a
single agent and two strategies: a persistent internal inactive self-strategy
associated with the player’s interest flow and an active external active
strategy that through the player’s actions, produces outcomes. To this we
need to add the constraints that go beyond the statement of the number of
strategic choices. We will do this in decision process theory by a
specification of the stresses (pressure) and strains (bonds) that form the
boundary conditions of the time dependent problem.
We characterize the Robinson Crusoe story as one in which a subject
forages (hunts and farms), whose sole strategy is to determine how much
or how little of this he or she should do. The assumption is that there are
sufficiently abundant resources so that no choice needs to be made on
which approach to take (hunt or farm).
It makes sense that anybody stranded on a desert island starts out
with significant prior experiences. Just as in the Robinson Crusoe case, a
subject brings with them a set of fundamental beliefs, which is their
worldview (Cf. Sec. 11.2). Someone brought up to be interdependent
might be more likely to conserve even with super abundant resources;
they might save for a rainy day or a day in which someone else might be
stranded. They might treat Friday better. Someone brought up to be
independent and a little self-centered might strive only to optimize his or
her own pleasure. Thus there are behaviors of strategic interest even in
this simple example. We show how decision process theory handles such
behaviors and constraints, starting in the next section where we put this
story into normal form.
376 Geometry, Language and Strategy—Vol. 2
conserve what is there and take only the minimal necessary to insure
personal survival.
This strategy however is inactive; nothing material changes based on
the value of this strategy. What gives Crusoe pleasure or pain does not in
fact change the decisions that Crusoe makes, which are assumed to be
solely associated with foraging. However, because of these internal
desires, there exists a value or player interest that Crusoe can assign
to the outcome of the decision process. This value exists because of the
hidden variable of the inactive strategy, but does not depend on the value
of that inactive strategy, only the active strategy of foraging.
The dynamics will depend on this decision value along with the net
result of the constraints of being on the island. There will be stresses or
forces that may move Crusoe to take actions (invasions by other
islanders), that are not strictly part of the problem at hand. These will be
captured in terms of the pressure of the system, p . There are bonding
forces as well as the strength of engagement e that relate how
quickly or not Crusoe learns from his experiences. Most of these effects
are not important from a static view but dictate how strategies are forced
to change in time.
The normal form has a single active and single inactive strategy,
i : 1, 1 , corresponding to a 2 1 -dimensional space-time.
a ,
j
There is
a single player interest field with non-zero component F tu ; there are no
purely spatial components of the payoff and so no payoffs that
correspond to ordinary game theory. There is only the game value
represented by this time component. We use the single strategy model
and frame-wave solutions from Sec. 6.8.
Our goal is to separate the aspects of the behavior that are steady state
from those that are time dependent. Our method is to imagine that we
move in a frame in which the system appears to be at rest: the co-moving
coordinate basis. For the models we have been considering, the various
attributes of the theory will be independent of time in that frame.
The flow equations in the co-moving frame, Table 5.1, are expressed
in terms of the flow direction , the proper inactive strategy direction
and the proper active direction . The flow equations depend on the
acceleration q q , a single vorticity or player interest , the
strategy , the player bonding coefficient and no shear
components, Eq. (6.46):
E j E j
e 2 q e
E j q E j 2 E j E E j e . (10.1)
f j
2 jj E j q e 1 e e e 1 jj E j E j
jj E j E j E j E j E j E j 1 e e
q q 2 q 2 p p
2 q
3 2 2 . (10.2)
p
q
2
p 0
Robinson Crusoe Economics 379
The inactive pressure p and energy density are undetermined. For this
investigation we make the simplifying assumption:
p p p
. (10.3)
r p
q q 2 q . (10.4)
the extremes. These arguments are steady state arguments, not time
dependent arguments.
In addition to the steady state equations, we have the frame-wave
phasors, Eq. (6.52) after we eliminate the time dependence:
x a 2 e 2 q 1 e 2 x a q x a 0 . (10.5)
f j
2 E j q e 1 e e e 1 jj E j E j . (10.6)
between the player’s active strategy choice and their inactive strategy
choice.
If we take the boundary decision value field to be positive, then
Eq. (10.6) determines that we pick the player interest to be negative.
Note that there are two solutions to player interest so that we do have to
arbitrarily set what we mean as a positive for one of the fields.
We think it reasonable to assume that at the known position, there is
no player engagement. Based on the player interest being negative, this
implies that the player engagement is positive towards less foraging and
negative for more foraging.
The known values of the transformations at the origin are taken so
that the co-moving basis and normal-form coordinate basis are aligned at
the origin 0 :
E u E j 1
. (10.7)
E u e E u E j 0
If we take other values for the known flow at the origin, then we use the
orthogonalization process described in Sec. 7.6 to determine these
transformations.
If we look at Eq. (10.2), we note that we have not said much yet
about the compression or strategy bonding parameter . We have
found that large values of strategy bonding determine how quickly
energy transfers from motion to fields as we move along the strategy
direction. Thus this will also impact the relative strength of foraging
between the extremes. We have again two possible sign choices for the
bonding: we have chosen it to be positive, which yields a large available
space for what we term to be conserving. The equations are symmetric in
that there is another solution with bonding negative in which the larger
space is for foraging.
The remaining scalar acceleration, stress (pressure) and strain (player
bonding) q p are set to represent a solution that has a small
force towards more foraging, a small stress and a strain that yields a
negative vorticity from Eq. (10.2):
382 Geometry, Language and Strategy—Vol. 2
For the prisoner’s dilemma, our known behavior was for zero
acceleration and a maximum pressure. However we found in our
sensitivity analysis, other solutions that had zero acceleration and a
minimum pressure. We are reminded
of weather in which we have both
high pressure and low pressure points.
Based on the known behaviors
from Sec. 10.6, we find solutions in
which the pressure is a minimum, Fig.
10.1 and for the acceleration, which
goes to zero, Fig. 10.2. Recall that if
Figure 10.1. Pressure p we set the known value of the
acceleration to zero, it will be zero for all strategic foraging values.
A possible rationale for a small
starting pressure is that Robinson
Crusoe has been on the island for
some time and that there may be a
foraging choice that he has
determined where he has no particular
recollection of any historical or island
constraints (such as invasions) to
contend with. In other words, there is Figure 10.2. Acceleration q
very little constraint at this point. He is not engaged: e 0 . There may
Robinson Crusoe Economics 383
10.8 Frame-Waves
x aTWA E e
p
E x a 0 . (10.9)
p p i p 1 2 0 q 0 i 2 e 2 q 1 4 0 q 0
2
0
x a x a sin p p .
TWA p
E e (10.10)
Robinson Crusoe Economics 387
p 1 2 0 q 0 . (10.11)
cp . (10.12)
p
x a x a x a
SW
Thus we can ignore the second term. We expect that this wave does not
travel: each part moves up and down together, albeit with different
amplitudes.
In the following sections, we will investigate the behavior of both
travelling frame-waves and standing frame-waves. We start with a
discussion in the next section on the limits to choice.
1 e 2 . (10.14)
We term this the boundary difference since we see that it tends to zero at
the boundaries of the space, Fig. 10.11.
390 Geometry, Language and Strategy—Vol. 2
g 1 e 2 e2 q . (10.15)
time d . The wave components are those components that are left over
from the total number of field variables (based on the number of
independent metric components in a holonomic frame), 1 2 d d 1
minus the number of gauge conditions d and minus the number of field
equations that are independent of time, which can be shown to be d
based on the Codacci equations, Sec. 5.3.4.
On general grounds we see that for our specific case of a, i : 1, 1
active and inactive components, the dimension of space-time is d 3 so
there should be no gravitational radiation wave components in free
space. This is not the end of the story, however. Gauge theories generate
two distinct types of wave phenomena: what we may call near zone and
far zone waves. To pick an example, take the case of electromagnetic
waves.
There are Coulomb forces in the near zone that follow an inverse
square law. A charged fluid source that moves as a travelling wave will
generate an electromagnetic field based on Coulomb’s law in response to
that motion. That wave moves as a travelling wave with the same
characteristics. It will respond based on Coulomb’s law with a delay
equal to the speed of light, yet the wave will appear to move with a speed
characterized by the charged media. It will have additional components
as well, based on the Coulomb field.
There will be radiation fields generated anytime a charge accelerates
defined as fields that drop off as the inverse distance, more slowly than
the Coulomb field. They are present in the far zone, long after the near
zone waves have died out. These waves will appear as travelling waves
that are moving at the speed of light, not at the speed of the source
waves.
392 Geometry, Language and Strategy—Vol. 2
1 2 d a d i 1 d a d i 2 1 2 d d 3 . (10.16)
In particular, for the player fixed frame model in the co-moving frame,
we have eliminated the time behavior of the inactive metric, which in the
norm-form coordinate basis is independent of proper-time. We expect no
radiation waves in this model for these inactive components.
There are different possible wave behaviors for the payoff and player
value fields. There are d a 1 independent wave components expected,
after we account for gauge invariance and identify the wave equations
that depend on time. We expect these waves to be like electromagnetic
field waves and be present when there are two or more active strategies.
Similarly, we expect gravitational radiation waves when there are at
least three or more active strategies. Such waves will be in the far zone,
so the choice region has to be sufficiently large.
To recap, non-game theoretic aspects of the decision process dictate
the limits to choice for both the stationary aspects of the decision as well
as the travelling wave aspects. We have taken values so that the
travelling wave behaviors can be seen clearly. We investigate such
travelling wave solutions in the next section.
Robinson Crusoe Economics 393
structure than each of the component frame waves. Because the bonding
field carries energy and momentum, there are constraints on its values.
We see possible singular behaviors developing near the limits to choice.
We see a standing wave pattern for the strategy flow, Fig. 10.18.
With a little imagination, we make out a pattern that is not a square, but
starts to look like a hexagon. We think that such structures are not
accidental but reflect the underlying dynamics. We expect that
reflections of the non-linear dynamics will be present in each of the
frame transformations.
Robinson Crusoe Economics 397
also at the left. So in the middle region, the standing wave represents the
way in which the value of what Crusoe does propagates. In this case, we
see the result of a standing wave with equal propagation left and right.
10.14 Outcomes
co-moving coordinate basis. The problem has the same content in the
theory; each frame provides valuable insight into the problem. This
basis in particular helps isolate the steady state behaviors.
From Sec. 10.6, the student must identify all the variables that must
be initially specified and provide known values for these variables.
From Sec. 10.7, the student isolates the steady state behaviors of the
problem.
From Sec. 10.8, the student has learned how to decompose the
solutions to the frame-wave equation into travelling waves. Part of
this decomposition involves identifying the approximate solutions.
From Sec. 10.9, the student learns the importance of identifying the
limits to choice, and quantifying that as the choice region. The
approximate travelling wave solutions may help to characterize the
choice region.
From Sec. 10.10 the student will have learned the characteristics of
the travelling wave solutions.
From Sec. 10.11 the student will have learned the structures one
might expect for the standing wave solutions.
From Sec. 10.12 the student learns more about the boundary
behaviors, which can set time limitations as well as limitations to
choice. Long time frame events will have different dynamics than
short term events, which will be more similar to game theory
expectations.
From Sec. 10.13 the student learns more about the decision process
field.
10.15 Exercises
a, i p c : 1,
1 d p dc
a, i p c : 2, 1 d d
p c
a, i p c : 2, 2 d d
p c
a, i p c : 3, 1 d d .
p c
a, i p c : 3, 2 d d
p c
a, i p c : 3, 3 d d
p c
...
(4) Imagine a political game in which there are three “players”: there is
a player that chooses the political party (say A or B); there is the
player that chooses a financial payoff to the party in power (say high
or low); and there is the player that chooses the popular vote (say
for A or for B). Assuming the game has been defined already in
normal form, discuss the attributes of the game and suggest one or
more possible codes of conduct and the form the equivalent game
might take.
(5) With the numerical calculation based on Sec. 10.8, describe the
effects seen in the holonomic coordinates, Fig. 10.25.
(6) How do the effects seen in Fig. 10.25 compare to those in the
prisoner’s dilemma, Sec. 9.4?
406 Geometry, Language and Strategy—Vol. 2
(7) With the numerical calculation based on Sec. 10.12, describe the
effects seen in the holonomic coordinates, Fig. 10.26, the case
corresponding to high player bonding.
(8) We have computed the time bonding, Fig. 10.28, using the base
model of Sec. 10.7 with a lower frequency of 3 . Discuss its
similarities to the model in Fig. 10.21.
429
430 Geometry, Language and Strategy—Vol. 2
propose that the active players in the decision process cooperate through
a dynamic law of cooperation. Such cooperation is by the active players
and involves the active strategies. In Sec. 12.5, we propose a third
important law, the dynamic law of organization. This law deals with the
societal or organizational setting, which takes into account the
constraints of the context in which competition and cooperation occur.
With these laws, dynamic competition, code of conduct (Sec. 11.2),
and dynamic cooperation (Sec. 12.4) provide the dual notions of non-
cooperative and cooperative games in decision process theory.
Everything hinges on how we deal with uncertainty, which we turn to
now.
pure strategy that has the largest payoff would then be the best choice
from this defensive stance, called the max-min solution. In the chess
game example, the max-min solution for each player has a payoff. If the
two payoffs are the same, the game is strictly determined.
This is the max-min theorem and holds for all strictly determined
games. For strictly determined games, the outcome or payoff is the same
for every player and each player plays one of their pure strategies.
Furthermore, there is no other pure strategy that will give a better
outcome irrespective of what the other players might do. In general,
games are not strictly determined, so these general games need to be
dealt with separately.
For games that are not strictly determined, the max-min solution may
not be optimal. Though the intensive form exposes only the strategic
properties of the game, it fails to take into account an important aspect of
uncertainty for games and decisions. We know most about games if we
play them repetitively. As in scientific inquiry, we have a hard time
creating a theory for phenomena that occur once. The exception is for
strictly determined games, since it makes little difference if a game is
played once or multiple times. The max-min solution is still the optimal
solution.
For any repetitive game that is not strictly determined, any player that
consistently plays a single strategy can have their strategy discovered by
the other players. Other players may cease acting defensively in hopes of
improving their outcomes. Once a player discovers that their defensive
strategy choice starts leading to lower payoffs, they will be encouraged
to change to some other pure strategy. This in turn will lead the other
players to modify their choices. We discuss in more detail below whether
a player makes a choice of pure strategies based on a prediction of what
other players might do, or makes their choice based on past history of
what is no longer a good idea.
Either way, each player sees an advantage to diversify their portfolio
of choices and choose plays according to some frequency distribution of
pure strategies, called a mixed strategy. For players that pick mixed
Decisions and Determinism 433
“Probability has often been visualized as a subjective concept more or less in the
nature of estimation. Since we propose to use it in constructing an individual, numerical
estimation of utility, the above view of probability would not serve our purpose. The
simplest procedure is therefore to insist upon the alternative, perfectly well founded
interpretation of probability as frequency in long runs. This gives directly the necessary
foothold.”
“… For many years, extending over a century and a half, attempts were made to
extend the domain of the idea of probability to the deduction of inferences respecting
populations from assumptions (or observations) respecting samples. Such inferences are
usually distinguished under the heading of Inverse Probability and have at times gained
wide acceptance. This is not the place to enter into the subtleties of a prolonged
controversy; it will be sufficient in this general outline of the scope of Statistical Science
to reaffirm my personal conviction, which I have sustained elsewhere, that the theory of
inverse probability is founded upon an error and must be wholly rejected. Inferences
respecting populations, from which known samples have been drawn, cannot by this
Decisions and Determinism 435
“The importance of the social phenomena, the wealth and multiplicity of their
manifestations and the complexity of their structure, are at least equal to those of physics.
It is therefore to be expected—or feared—that mathematical discoveries of a stature
436 Geometry, Language and Strategy—Vol. 2
comparable to that of calculus will be needed in order to produce decisive success in this
field. (Incidentally, it is in this spirit that our present efforts must be discounted.) A
fortiori it is unlikely that a mere repetition of the tricks which served us so well in
physics will do for the social phenomena too. The probability is very slim indeed, since it
will be shown that we encounter in our discussions some mathematical problems which
are quite different from those which occur in physical science.
“These observations should be remembered in connection with the current
overemphasis on the use of calculus, differential equations, etc., as the main tools of
mathematical economics.”
In support of their view, we see value in the axiomatic tools that have
been applied that have provided much needed analysis and clarity of
distinctions about what constitutes decisions in general and economic
decisions in particular.
We believe the pendulum of this discussion has perhaps swung too
far. We are not in agreement that “tools of the physical sciences are
inappropriate”. We believe a number of advances have been made since
that statement was written, as well as a few setbacks. The particular
complexity, namely the existence of a stable standard of behavior,
suggested by Von Neumann & Morgenstern [1944] does not always exist
[Aumann, 1989, p. 13]:
“Von Neumann and Morgenstern were thus led to the following definition: A set K
of imputations is called stable if it is the set of all imputations not dominated by any
element K :
“This definition guarantees neither existence nor uniqueness. On the face of it, a
game may have many stable sets, or it may have none. Most games do, in fact, have
many stable sets, but the problem of existence was open for many years. It was solved by
Lucas (1969), who constructed a ten-person TU coalitional game without any stable set.
Later, Lucas and Rabie (1982) constructed a fourteen-person coalitional game without
any stable set and with an empty core to boot.”
12.2.1 Preferences
Bernoulli [1738], the treatment that best meets our needs starts with Von
Neumann & Morgenstern [1944], which is based on frequency of
occurrence rather than probability.
Their idea is that it is self-evident (in the sense of creating axioms)
that between two alternatives, A and B, from past behaviors a decision
maker can choose the one that has more utility. This knowledge is not
enough however to provide a basis for utility in game theory. A decision
maker needs to know more. Between the alternative A and the alternative
of some combination of B and C, again by past behaviors a decision
maker can choose the alternative that has more utility. When the concept
of mixture is made precise, a numerical utility function can be defined
that expresses a utility for any mixture of choices. They adopt the most
conservative approach, which is that the mixture is based on the
frequency of occurrences of choices B and C in the past. This approach is
consistent with the usual application of the scientific method [Fisher,
1925], as it does not invoke Bayesian or inverse probability (Sec. 12.1).
The issue of how we predict the future becomes more complex when
we deal with more complex decisions than picking preferences. Now we
have issues of uncertainty in our outcomes based on decisions made by
others or factors outside of our control. We hold to the view that despite
these complexities, a process view is possible in which the future flows
from past behaviors of all the agents involved in the decision process.
We may be able to calculate that flow but only if we make assumptions
about those underlying processes. Hence, we make no use of probability
as estimation.
We do make use of probability or statistics in the usual way it is
employed with the scientific method to discredit assumptions that appear
extremely unlikely. Thus in our dynamic theory we base our assumptions
of what is known about decisions in the present and past and use decision
process theory to determine future behaviors. We use the scientific
method to modify our theory based on observations and measurements.
12.2.2 Measurement
times and average the values obtained. There are differences each time
based on the uncertainties of the measurement process that are unrelated
to the concept of length. In decision processes, there are uncertainties in
measuring the utilities that are unrelated to the presumed concept of
utility between A and a given combination of B and C.
Thus decision process theory allows us to use past data on
frequencies to make choices. Based on this approach, we expect the
axiomatic proof of utility [Von Neumann & Morgenstern, 1944] to
follow unchanged. As a consequence, we should be able to compare
frequency distributions of each decision maker and allocate a numerical
utility, the payoff, to each of them. In order to assure that we can carry
out this program, we envision that sufficient repetitions of the decision
process can be observed in order to arrive at reasonably accurate
measures. We thus specify what we mean by a point in time: it is an
interval that is large enough to carry out the above program, though not
so long as to open the discussion of what behaviors might be steady
state1.
We believe there are numerous examples of this process used today in
the commercial world, examples that assign time-dependent numerical
utilities and payoffs to processes that are inherently human and
subjective. Consider as an example the production of code by software
developers. The production of code results from a collection of decisions.
If the decisions were correct, the code works; otherwise the code has
parts that don’t work and has defects that will be identified in the
discovery or testing stage.
The percentage of code that works versus code that doesn’t work
represents a frequency distribution that is knowable in principle at the
time the code is produced but in practice is not known until later. One
can nevertheless model the production of code as a quantity of code with
a certain defect fraction. The defect fraction is not an equilibrium
statement. It depends on the skill and experience of the software
developers. Over time and based on feedback during the discovery stage,
the number of defects will go down.
1
Cf. Sec. 3.9, where we discuss an analogy to the breakdown of classical mechanics or
optics when the time or space intervals become too small.
Decisions and Determinism 441
We assert that it makes sense to consider the cardinal (we know the
actual value) as well as ordinal (we only know the relative order)
properties of utility based on comparing choices that involve
combinations with frequencies [Von Neumann & Morgenstern, 1944,
p. 617]. For choices that involve certainty, only ordinal utility is needed.
The consideration of mixed strategies however brings us back to a
cardinal view (up to linear transformations).
There appears however, a belief that remains that the cardinal view is
too restrictive and only the ordinal view is needed. It is framed as the
view that only the set theoretic view is needed because the calculus view
lacks the capability to produce complex effects [Luce & Raiffa, 1957,
p. 18]:
u A fu B 1 f u C . (12.1)
Before drawing their conclusion, we note that the word investment was
used here because of our starting example.
We don’t however require a concept of money, but a concept that we
are willing to split our effort between B and C by some fraction. In the
context of decision process theory, the simplest way to envision splitting
our effort is by our actions. We thus envision that we prefer putting our
effort in A to splitting our effort between B and C based on the fraction
f .
If this is the case, then the numerical utilities will be related in the
above way, Eq. (12.1). These ideas are made mathematically precise by
Von Neumann & Morgenstern [1944], who show that under reasonable
mathematical assumptions, the utility function is determined up to a
linear transformation. In other words if there is any other utility
444 Geometry, Language and Strategy—Vol. 2
function u that also satisfies Eq. (12.1), then there will be numerical
constants a , b that relate the two utility functions for any choice A:
u A au A b . (12.2)
12.2.5 Payoffs
G G12
G 11 . (12.3)
G21 G22
Decisions and Determinism 447
The most defensive strategy for each player is for them to consider the
worst that can happen and then pick the best of these worst cases.
It is not always possible however to find such a steady state solution
since not all games are strictly determined. For games that are not strictly
determined, it is still possible to find a defensive strategy for two-person
zero-sum non-cooperative games using mixed strategies [Von Neumann
& Morgenstern, 1944]:
U r : U r 1 ,
r
W s : W s 1 .
s
that there will always be at least one solution to this problem. The steady
state behaviors for this class of games will be these defensive solutions.
The steady state behaviors provide a rule or association of strategic
choices to the given payoff matrix elements.
For strictly determined games, we can often determine the dominant
strategies by inspection. It is in general more complicated to determine
the mixed strategy steady state behaviors for games that are not strictly
determined. There is one case however that is particularly easy: again it
is the 2 2 game.
We consider such a game in which neither player has a dominant
strategy. We compute the following differences for each row and column
as follows:
G p2 1 p2 2 odds1
p1 1 G11 G12 G22 G21
. (12.6)
p1 2 G21 G22 G11 G12
odds 2 G22 G12 G11 G21 G22 G11 G12 G21
Since neither player has a dominant strategy, either the row and column
elements labeled odds will all be positive or all negative. If they are all
negative, we multiply the elements by minus one. It can be demonstrated
that the steady state behavior Eq. (12.5) occurs when player 1 picks a
mixed strategy based on the odds1 column and player 2 picks the mixed
strategy based on the odds2 row. To get the normalized mixed strategy,
divide the odds by their sum, given in the odds2/odds1 entry.
The ease with which one can compute the mixed strategies for 2 2
games makes them attractive as examples (Cf. the exercises at the end of
this chapter), but not representative of the richness present in zero-sum
non-cooperative games. By increasing the number of pure strategies for
each player from the 2 2 case, the decision process can be made
significantly more realistic. It also becomes much tougher to find the
steady state solutions by hand or by calculator. In addition to the
Decisions and Determinism 449
U r : U r 1 :
r
U G
r
r
rs v. (12.7)
u G
r
r
rs 1. (12.8)
W s : W s 1 :
s
G W
s
rs
s
v. (12.9)
G
s
rs ws 1 . (12.10)
ua 0 ,
u F
a
a
ab 0. (12.11)
Not all games are limited to two players, not all games are non-
cooperative and not all games are zero sum. These extensions can be
made and we find that the treatment of such games nevertheless follows
the above outline.
We argue as follows. In decision process theory, we ascribe a code of
conduct to cover persistent cooperation, cooperation that survives all
dynamic processes. There is a payoff matrix for each player (including
pseudo-players that owe their existence to the code). As argued in Sec.
12.2, we remove the constraint that the utilities of each player are zero-
sum or transferable. Still, under rather general conditions we expect that
there will be steady state behaviors, Nash equilibrium [Nash, 1951],
which are determined by the payoff directions. Such behaviors have the
attribute that they assume all players behave rationally and assume that
with rational behavior, no player can expect to benefit more than what is
prescribed to the steady state behavior by deviating from that
452 Geometry, Language and Strategy—Vol. 2
Though this provides the law of competition, there are some caveats.
First, the law depends on the cooperation potential, jk , Cf. Sec.
12.4. The interaction is the product of the gradient of the cooperation
potential and the payoff matrix (Cf. Eq. (4.16)).
Decisions and Determinism 453
(1) “All preplay messages formulated by one player are transmitted without distortion
to the other player.
(2) “All agreements are binding, and they are enforceable by the rules of the game.
(3) “A player’s evaluations of the outcomes of the game are not disturbed by these
preplay negotiations.”
The cooperative decision processes are those in which the players make
binding decisions that hold throughout the duration of play, what we
termed a code of conduct in the previous section.
Our starting point is [Luce & Raiffa, 1957, p. 118]:
2 1 1 1
G1 , G2 . (12.13)
1 1 1 2
At the outset, we see that the two payoffs are not zero-sum. In Sec. 12.3
we show how to compute the strategies for competitive games. Using
those techniques, we compute the most defensive strategy, which is that
each player assumes no cooperation with the other player.
We assume that each player plays a competitive game assuming his
own payoff matrix. The ideal strategies for player 1 are then to play his
strategies with the odds 2:3. We would say that player 1 is playing
against an imaginary opponent, his view of how player 2 would play.
This imaginary player 2 would play the odds 2:3. The real player 2
however plays the odds based on his view of the decision process and
plays the odds 3:2. His view of the imaginary player 1 is that that player
would play the odds 3:2 as well. The real player and imaginary player
don’t align. If each player only played this defensive strategy, they
would each receive a payoff of 1 5 . We call these strategies the max-min
or defensive solutions.
The players might however alternate between the first and second
choices in synch with each other, so they would receive the payoffs
2,1 or 1,2 . The average payoff 3 2 to each player for this behavior is
much better. To achieve this improved payoff, the players must
cooperate with each other.
The difficult issue is the nature of the cooperation and how to
compute the outcome of that cooperation while still appropriately taking
into account the competitive nature that lurks behind each player’s sense
of utility (Cf. Sec. 12.3). One solution [Von Neumann & Morgenstern,
1944] is to consider the possibilities for the two players viewed in the
two-dimensional space of their utilities. Players will cooperate if they
can achieve more than what they could do based on the max-min
solutions. This defines the sub-region of the utility diagram in which
cooperation can occur. Cooperation can be persistent, in which case we
invoke the code of conduct argument from the previous section.
Alternatively, cooperation can be dynamic.
456 Geometry, Language and Strategy—Vol. 2
In this case, we find that cooperation does not replace the competitive
nature (Cf. Sec. 12.3) of the decision process but is an additional force.
We propose as fundamental, the law of cooperation, Eq. (4.21), stating
that the cooperation potential jk between player j and player k is
determined by the payoffs of the two players and the inertial stress
between them:
1
2 g ab a jl b lk T j k 1 4 F j ab Fk ab . (12.14)
For a simple but illustrative model of the stresses, Eq. (3.43), the
organizational system stress between distinct players
T k p V Vk p k , is proportional to the product of the player
j j j
dV b
g ab abcV bV c Vk Fabk V b 1 2 V jV k a jk qa . (12.15)
d
This provides the closed loop (Cf. Sec. 13.2) result that the time rate of
change of the flow (the acceleration) depends on inertial (or frame)
effects ( abc term), competitive effects (that depends on the payoff) and
cooperative effects (that depend on
the cooperative potential). We discuss
the source contributions in Sec. 12.5.
Assuming only cooperative
effects, steady state behavior occurs
when at a maximum or minimum
potential of the cooperative potential.
Since the product of the interest flows
and cooperation potential involve the
Figure 12.1. Prisoner's dilemma square of each player’s interest flow,
11 it is reasonable to expect the overall
acceleration or force to be a
maximum at equilibrium. In that case, moving away from the point
would be a restoring (negative) force. This would hold for the
autonomous cooperation potential (the self-cooperation effect, Fig. 12.1)
as well as the cooperation between distinct players.
We might further expect that the
cooperation potential for distinct
players would be a maximum when
the interest flows have the same sign
(both givers and both takers) and a
minimum when they have opposite
signs, Fig. 12.2. Based on the model
from Sec. 8.2, we see from the above
two figures that our expectations are
qualitatively borne out. The model Figure 12.2. Prisoner’s dilemma
also demonstrates that each player 12
may exhibit both buyer and seller
attributes, depending on the strategic value, Fig. 8.15. The theory
provides a quantitative realization that incorporates all the effects.
458 Geometry, Language and Strategy—Vol. 2
“Consensus: ‘The process of abandoning all beliefs, principles, values and policies in
search of something in which no one believes, but to which no one objects’. Margaret
Thatcher, 1993.
“Coalition: The art of enlisting divergent and independent interests to attain a
valuable objective that everyone believes in, if perhaps for somewhat self-interested
reasons.
“Cognition: The process of educating people in preparation for using another
technique to attain unity of purpose.
“Coercion: Invoking higher authority to attain a short-term goal. It is most effective
in preparation for coalition building OR when a decisive victory is possible.”
We have dealt with two aspects of decision making that are covered
extensively in game theory: competition and cooperation. There is a third
aspect however. Though not always recognized, decisions occur in a
context of an organizational system.
Certainly competition and cooperation provide one such context, but
there are others that are like constraints in that they impact the outcome
but we need to know only some aspects of these constraints. If we
included every conceivable player and all conceivable environmental
factors, we would not need constraints. We can’t totally ignore all of the
other players however, any more than we can ignore constraints in
physical processes, Sec. 1.6.
Decisions are made in the context of a family between family
members, in a community, political system or in a society. Decisions
may be made in the context of a business organization. They may be
made in the context of an environmental system. Thus even the physical
environment provides a system of constraints that decisions need to live
within. In each of these cases the system provides an organizing source
that determines the context.
In decision process theory, there are both forces that pull the system
together as well as forces that push the system apart. The inertial
organizing sources generate forces that can do both as well. In physical
systems they are related to the concept that an elastic system that is
deformed returns to the same shape as before, once the deformation
forces are removed. The shape is a type of organization. In a fluid, flow
goes from high pressure to low pressure. The system likes to return to a
state of equal pressure everywhere. This is a specific type of organization
that is characteristic of most organizational systems. They don’t like
changes. They change under stress, but revert to the original form after
the stress is removed. These effects are not dissimilar to cooperative
effects.
In addition, inertial forces may generate vorticity or rotations, such as
Coriolis forces for fluids. In this regard, inertial forces act like payoffs.
When time is an isometry, this is not just an analogy but an exact
statement: there will be a payoff field associated with time as if it were a
Decisions and Determinism 461
code of conduct. We see this in the solutions when there is more than one
active strategy. Thus we believe there are organizational payoffs and
organizational cooperative effects that are equally important in making
decisions as competitive and cooperative payoffs.
If the competitive and cooperative forces are absent, the principle of least
action Eq. (4.17), with Tabpayoff Tabinactive 0 , determines the time and
space bonding, the contextual frame potential gab between active
strategies in terms of the inertial organizing source energy momentum
Tab only:
12.6 Outcomes
From this chapter, the student will have learned the distinctions in game
theory and decision process theory for two important concepts:
uncertainty and determinism. Decision process theory approaches
decisions as based on deterministic mechanisms as opposed to Bayesian
probabilities. Uncertainties exist but are separated between the frequency
with which a choice is made and the uncertainty associated with making
that choice, akin to the separation made in the measurement process in
science and engineering.
The attainment of the outcomes of this chapter is facilitated by doing
the exercises in the following section. Based on this investment, the
student should achieve the more detailed outcomes below based on
section.
From Sec. 12.1, the student will have learned the advantages and
pitfalls of applying the concepts of probability and uncertainty to
decision processes. The student will have learned that in decision
process theory, each player sees an advantage to diversify their
portfolio of choices and choose plays according to some frequency
distribution of pure strategies, called a mixed strategy. It is not
required that this frequency distribution be a prediction of what is
likely to happen in the future.
From Sec. 12.2, the student will understand how to identify a different
utility function for each agent. The theory identifies the resultant
payoff fields with the energy of the system. Energy is convertible and
comparable, though utility is not.
From Sec. 12.3, the student will have learned the relationship between
the max-min rule of game theory and our dynamic law of competition
for obtaining the flow vectors from the payoffs, Eq. (12.12). The
economic equivalence principle is that our view and the max-min
rule are the same for two-person zero-sum non-cooperative games.
From Sec. 12.4, the student will have learned the proposed
fundamental law of cooperation, Eq. (12.14). Players contribute
differently to this law depending on their interest flows, whether they
are givers or takers. The law of cooperation in decision process
theory requires a common ground, an overlap between the players’
464 Geometry, Language and Strategy—Vol. 2
12.7 Exercises
60 100
100 80 .
Determine the mixed strategies for each. The rows and columns are
labeled: Blue 1=bomb carrier in less-favored position; Blue 2=bomb
carrier in favored position; Red 1=attack on less favored position;
Red 2=attack favored position.
(4) We are told the following river tale by Williams, [1966, p. 50]
“Steve is approached by a stranger who suggests they match coins. Steve says that
it’s too hot for violent exercise. The stranger says, “let’s just lie here and speak the
words ‘heads’ or ‘tails’—and to make it interesting I’ll give you $30 when I call
Decisions and Determinism 465
‘tails’ and you call ‘heads’, and $10 when it’s the other way around. And just to
make it fair—you give me $20 when we match.”
The payoff matrix is
20 30
10 20 ;
4 1
3 4 .
0 1
3 0 .
Umbrellas cost him 50 cents and sell for $1; glasses cost 15 cents and sell for 50
cents. He is willing to invest $250 in the project. Everything that isn’t sold is a total
loss (the children play with them).”
The payoff matrix is buying versus selling with positions rain and
shine:
250 150
150 350 .
0 1 1
1 0 1 .
1 1 0
(9) Determine the optimum strategy for playing the game of Morra
[Williams, 1966, p. 163]. There are two players. Each player holds
up either one, two or three fingers and simultaneously guesses what
the other player will display. If just one person guesses correctly,
the payoff is the number of fingers held up. There are 9 pure
strategies for each player that can be labeled with two numbers, the
number of fingers held up and the guess as to what the other player
will do. The pure strategy 13 is to hold up 1 finger and guess 3.
Show that the steady state behavior involves only three of the pure
strategies, 13, 22 and 31. Show that they should be played in the
ratio 5:4:3.
Chapter 13
We have noted that the idea of using physics and differential equations
for modeling economic behaviors is quite old [Edgeworth, 1881].
Though this approach has not generated a following and indeed is dated
based on what we now know about the physical and economic worlds, it
contains some important concepts. Physical laws are often determined by
observing a phenomenon in isolation, from which a behavior is extracted
that is true locally.
For example, the world we live in is observationally flat. The
observation is something we really understand and relate to. When we
make local observations about flatness at different points of our world,
the difficult issue is how to express what we have learned in a consistent
way. Just because the world is locally flat does not imply that the world
is globally flat. The insight we obtain from theoretical views such as
differential geometry is a way to stitch local views together. We are thus
led from a local view to a global view of the system we are studying.
This approach is common in the domain of the physical sciences. It
has also been applied with success in the social sciences. The concept of
using differential equations to model social phenomena was proposed by
Forrester [1961] and used by the Club of Rome [Meadows, Meadows,
Randers, & Behrens, 1972, p. 26] to describe the global changes one
might expect given initial conditions on such things as population, birth
rates and resource consumption:
467
468 Geometry, Language and Strategy—Vol. 2
“We too, have used a model. Ours is a formal, written model of the world1. It
constitutes a preliminary attempt to improve our mental models of long term, global
problems by combining the large amount of information that is already in human minds
and in the written records with the new information processing tools that mankind’s
increasing knowledge has produced—the scientific method, systems analysis and the
modern computer.”
“Although the perspectives of the world’s people vary in space and in time, every
human concern falls somewhere in the space-time graph. The majority of the world’s
people are concerned with the matters that affect only family or friends over a short
period of time. Others look further in time or over a larger area—a city or a nation. Only
a very few people have a global perspective that extends into the future.”
1
“[From the Club of Rome] The prototype model on which we have based our work was designed
by Professor Jay Forrester of the Massachusetts Institute of Technology. A description of that model
has been published in his book, World Dynamics (Cambridge, Mass.: Wright-Allen Press, 1971).”
Global Systems View 469
has been learned about economic behaviors from game theory but to
extend those theories.
Before addressing the large scale structures that are possible in such
theories and the motivation for expecting such behaviors we obtain from
systems dynamics, we note that game theory has not entirely relied on
the current algebraic approach. There have been studies of continuous
and learning behaviors for repetitive games [Clemhout & Wan, 1989].
There have been approaches that use the physics metaphor to focus on
the stochastic nature of economic processes [Murphy, 1965]. Decision
Analysis also focuses on the stochastic nature [Howard, 1964], according
to the Wikipedia article on this subject [Howard, 2010].
We find these approaches interesting and useful, though we have
reservations towards this approach as expressed in Sec. 12.1 based on the
way they deal with uncertainty. In addition, though some of these
approaches are differential in character, we feel that they don’t address
the global connections suggested by the Club of Rome. In this context,
we feel that decision process theory is most closely aligned with the
approach suggested by Forrester, [1961] based on the causal and
continuous nature of time.
There are many insights to be gained. We start with a summary of the
key elements, not the least of which is the availability of improved
computational techniques. An important element of great practical
advantage is that there are multiple software packages that make
application of systems dynamics no more difficult than applying a
spreadsheet model to an accounting problem. In this regard, we have
found the iThink® software by High Performance Systems [Richmond,
2001] to be helpful.
The prey, such as rabbits might also increase because of their food
source (carrots). The bi-directional character of growth is represented by
flows that have arrows that point in both directions. We represent the
continuous nature of time and approximate the discrete nature of creation
or demise of predators and prey by a continuous amount in their
reservoir. The picture we create in terms of flow results in a quantitative
description of the populations Fig. 13.2, which in this case has unusual
structure in no small part due to the additional food source of carrots.
Global Systems View 473
…In short, Operational Thinking is a big deal! It’s a big deal because, like Closed-
Loop Thinking, it has to do with how you structure the relationships between the
elements you include in your mental models. Specifically, Operational Thinking says that
neither “correlation,” nor “impact,” nor “influence” is good enough for describing how
things are related. Only causation will do!”
dynamic process that follows from the principle of least action. The
conservation of energy and momentum is expressed as the dynamic law
of organization Eq. (4.17), which determines the contextual frame g ab
between active strategies in terms of the total energy and momentum of
the system:
dV b
Qa g ab abcV bV c Vk Fabk V b 1 2 V jV k a jk qa . (13.3)
d
This is in contrast to the local source terms on the right that are
specifically related to competition, cooperation and organization.
In a more general sense, the frame effects are the consequences of the
topological connectivity assumptions about utilities. We may think of the
Global Systems View 479
13.6 Opportunity
We see that the stresses play an important role in the theory. The stress
tensor components T have been added to reflect the attributes of the
decision process that are part of the context. There may be millions of
people in the society, but in a specific decision process we are focused on
the few people actually making the decision. We can’t entirely ignore the
million people however. They carry energy and momentum; as sources
they generate the forces for each of the three laws we have just
identified.
Forces in general generate acceleration and for the central frame
active acceleration, Qa in Eqs. (6.78) and (13.3), we associate the idea
of opportunity. The opportunity results from a tension between
alternatives and so depends on directions in space and time. We use the
word tension because we believe it represents the force that moves the
decision makers toward or away from choices based on their current
knowledge of past behaviors. This is the tension that moves the system
towards the future.
Opportunity depends on the competitive force, cooperative force and
the elasticity or connectivity between strategic possibilities, such as the
production and consumption required of a company to produce a product
and make a profit. We can define a distinct opportunity for each: interest
opportunity, cooperative opportunity and organizational opportunity,
respectively.
Decision processes have complex opportunity connections. The
challenge is to capture the essence of such connections to accurately
portray observed effects. We provided a decision process theory model
for the inertial organizing source in Sec. 3.7, a perfect fluid. We have a
different model that seems appropriate in the player fixed frame model,
Sec. 5.4.4. These suggest a range of possibilities.
Using Sec. 3.7 as an example and based on ideas of organizational
dynamics described in Vol. 1, we can choose an ideal elastic situation in
480 Geometry, Language and Strategy—Vol. 2
b p
q a h ab . (13.4)
p
2
See Ex. 4 for a systems dynamics approach.
Global Systems View 481
pronounced then the inertial mass is very large. Future behavior may in
fact remain stable because there are no forces sufficiently strong to move
the system. The inertial mass is not a prediction of the future but a
statement about the reality of the decision process based on present and
past observations. The inertial mass effect can be overcome if other
forces become stronger.
Opportunity also includes our notion of player’s interest flow. In the
central frame, this is the force Vk FakV in Eq. (13.3); there are no payoff
contributions in the central frame. We associated interest flow with the
possibility of a player being a buyer or a seller. Buying and selling are
social transactions [Mill, 1947]:
“Again, trade is a social act. Whoever undertakes to sell any description of goods to
the public, does what affects the interests of other persons, and of society in general; and
thus his conduct, in principle, comes within the jurisdiction of society: accordingly, it was
once held to be the duty of government, in all cases which were considered of
importance, to fix prices, and regulate the processes of manufacture. But it is now
recognized, though not until after a long struggle, that both the cheapness and the good
quality of commodities are most effectually provided for by leaving the producers and
sellers perfectly free, under the sole check of equal freedom to the buyers for supplying
themselves elsewhere. This is the so-called doctrine of Free Trade, which rests on
grounds different from, though equally solid with, the principle of individual liberty
asserted in this Essay.”
Do we keep the car or do we junk it? We have researched the cost of the
repair of the mechanical problems and have determined the cost to be
$2K. If the car breaks down tomorrow, we will be out the $2K as well as
having to get the car towed to the junk yard. If we knew for sure the car
would break down tomorrow, we would be ahead by simply driving it to
the junk yard today. However, we don’t know for sure the car will break
down. In fact we have been driving the car for the past year with this
known problem. If the car drives for another year we save the price of
buying a new car. Is the $2K a real cost? What we can say for sure is that
it reflects the opportunity tradeoff of getting rid of the car versus keeping
the car.
In decision process theory, we analyze the problem as follows. We
have data from the mechanic who has serviced our car and many similar
cars with the same problems. Based on what he has told us we have a
current assessment of the utility of the car and whether the car will make
it to the future. This is not the same as knowledge of that future. We
make no assumption that the mechanic is in fact accurate in his
assessment. Based on our current assessment of the car’s utility we
make our decision. If we could iterate the experience, then we would
change our assessment of future utility as we gained or lost trust in our
mechanic. Our mechanic too might improve. What we know for sure
however is the future assessment based on our past experience.
We propose that there is an opportunity utility that expresses the
opinion of future utility for any given strategy, one that is neither
cooperative nor competitive. So, just as payoffs express the view of how
players will respond in a competitive situation, opportunity utility
expresses the view of how the future might unfold. This view is not a
prediction of the future however; it is simply the current view of that
future, just as the payoff matrix is the player’s view or utility measure of
how all the other players will behave. We can compute the utility over
time to estimate whether the current view will hold; even if it doesn’t, we
get a prediction for what the future utility will be.
Since we take utility to be energy, we consider keeping the car or
junking the car to reflect (potential) energy. If the potential energy is
higher to keep the car we hold on to it. If the potential is lower, we
continue to drive it. The potential reflects a reality about our experiences
Global Systems View 485
to date with respect to the car. These experiences are about the past and
present. To the extent that the process captures the essence of the
ongoing decision processes, we do in the end gain knowledge of the
future behaviors, though not as probability predictions but as systems
dynamic predictions based on causal behavior.
13.10 Outcomes
In this chapter the student will have learned that as part of closing the
loop, there are three laws that govern aspects of the decision making:
competition, cooperation and organization. The decision making process
is a system in which all these laws operate. In general it is not sufficient
to characterize a decision process as competitive, cooperative or
organizational since all three attributes may be present. It is important to
consider the closed loops of the process, which help identify the global
characteristics such as the limits to growth and the limits to choice. In
Part 3, we will examine in much greater detail the consequences of this
system view for three active strategy systems.
The attainment of the outcomes of this chapter is facilitated by doing
the exercises in the following section. Based on this investment, the
student should achieve the more detailed outcomes below.
From Sec. 13.1, the student will have learned an appreciation of the
limits to growth and the limits to choice. They provide a global
connectivity that is not visible at the local level. The student will have
learned the importance of closed loops. Closed loops have the same
significant effect in decision process theory as they would be
expected to have from an Operations Systems viewpoint. For
example, in decision process theory, the flow of decisions creates the
values of the payoff fields. The values of the payoff fields dictate the
behavior of the flows.
From Sec. 13.2, the student will have an exposure to systems
dynamics thinking and their approach to causality using stocks and
flows.
From Sec. 13.3, the student should understand the basis for the
approach taken in decision process theory.
From Sec. 13.4, the student gets a reprise of the three laws.
Global Systems View 489
13.11 Exercises
(3) For Fig. 13.3, assume that DeathRate is the product of the
DeathFraction and the square of the Population divided by the Goal.
Show that the growth is limited.
(4) In Fig. 13.4, we provide an example of a typical application of a
Systems Dynamics model. It models the delivery time for a large
software project. It is a global system that is made up of local
component loops, whose validity can be individually verified. For
example the FeatureBacklog is dependent on the rate
RequestForNewFeatures and the DevelopmentRate. Identify the
other local components.
(5) Building solutions to practical problems using Systems Dynamics
from small component parts as in Ex. 4, is facilitated by having well
understood archetypes. Using the World Wide Web, find and
describe each of the most common ones.
(6) Systems thinking is not a set of tools but a language [Richmond,
2001]. The nouns are the stocks, the verbs are the flows. Sentences
are formed by expressing simple processes, such as archetypes.
Paragraphs are formed by using the sentences to describe part of a
system. Global systems consist of many paragraphs. With this in
mind, look again at Fig. 13.4. What story (global system) does it
tell?
Global Systems View 491
x a , U 0 a V0 a
U a cos V a sin . (14.1)
495
496 Geometry, Language and Strategy—Vol. 2
has two strategies. We use as examples, a war game and a social decision
process: the attack-defense model and the work-wealth-wisdom model
respectively. In Chap. 17, we focus on the steady state geometry aspects
of the model. In Chaps. 18 and 19, we address the structural and social
distinctions, respectively, which arise from the model and suggest how
they play out in the real world.
14.1 Introduction
To orient the discussion, recall that for game theory solutions (Sec. 12.3),
strategic results are not dependent on position in strategy space: the
payoff matrix is constant and the metric is unity. In contrast, except at
equilibrium, we expect players to change their mixed strategies with
position and time. Furthermore, we envision that the strategies are not
normalized. We have some guidance from game theory on the dynamics
of such un-normalized mixed strategies. We also have guidance from
decision process theory, Cf. Sec. 2.1.
There is another distinction that we did not emphasize with our
example of the prisoner’s dilemma. At a point in space, we don’t
distinguish between an active strategy and a code of conduct. We can
always treat an inactive strategy or direction as active and leave it to the
dynamics of the theory to show that that direction is conserved. This is
true of the physical world as well: we can choose to ignore the fact that
motion is on a sphere, leaving it to the equations to establish that fact.
The key is that we have a way to move from one description to the other.
So with the prisoner’s dilemma, we look at the situation first from the
perspective that each prisoner has two strategies. We then look at the
situation from a different perspective in which some of those strategies
are in fact part of a code of conduct. In our analysis of strategy in this
chapter we need to keep these perspectives in mind.
In Sec. 1.4, we proposed that strategies are a measure of action: the
effort one applies to a specific choice or a mixture of choices is
determined according to frequency. We expect such a measure to be non-
negative and have suitable units. Thus any specific strategy would satisfy
0 yk . This differs from game theory because these strategies are
not normalized for each player.
In particular we suggest there are consequences due to the overall
size. The size is measured in terms of a scale set by the unit of
measurement, which we have provisionally identified as the effort. It
makes sense to separate the behaviors due to the size and the behaviors
that are relative to that size. Any ratio of strategies represents the relative
values whereas any strategy or sum of strategies represents the size.
500 Geometry, Language and Strategy—Vol. 2
1
rK xk ln K .
n K kK
1
r
n
xk ln ,
where the sum extends over all strategies.
The player preference scale is determined by the player effort and the
decision preference scale is determined by the decision effort
respectively:
rK ln K xk 0
kK
. (14.2)
1
r ln nK rK xk 0
n K
This formulation is different from that used in game theory but related in
that it achieves the similar effect that the normalized mixed strategies
will depend only on the relative preferences, not on the preference
scales. The sum of the un-normalized preferences for each player will be
determined by its player preference scale or the overall decision
preference scale. For decision process theory we argue that the strategy
flows yk correspond to game theory strategies. The flows transform as
vectors so we can equally use the preferences xk . Ratios of such strategy
flows will be differences of preferences and hence independent of scale.
By separating the relative preferences from the preference scale, we
distinguish between those choices that are due to the efforts put into each
strategy (resources) and those choices that are independent of such
efforts (skill).
To elaborate, we think of strategy as measuring the effort put forth for
a pure strategy or mixed set of strategies. The effort is an amount, which
is non-negative and can increase or decrease. Essentially, this effort is a
quantitative measure of the player’s strategic opinion about a given
strategy. As with populations, we think of steady state situations with
either a constant rate of increase or decrease. If the population rate is
proportional to the population itself, the steady state increase or decrease
will lead to either exponential growth or exponential decay. Preference
thus measures the slope of the exponential increase or decrease.
In this case a constant rate corresponds to a constant rate of change of
the preference. In this context, we relate such constant rates in decision
process theory to inactive behaviors. Because this desired behavior of
the preference flow is a constant, we have a useful alternative to the
normalized strategies of game theory. However decision process theory
and game theory are not the same: we see strategic opinion as being
fundamentally dynamic. The concept of constant effort replaces the
502 Geometry, Language and Strategy—Vol. 2
concept of static choice. Some preference flows will thus be inactive and
others active.
By using weighted sums, game theory is consistent with this choice.
However, it does suggest that each of the player preference scales rK be
inactive or minimally that the decision preference scale r be inactive.
This does make some sense in terms of the typical values used for the
payoff matrix. If both time and the decision preference scale were
inactive, we would expect that the sum of the time components of the
payoff matrix would be zero. This is consistent with zero-sum and
constant-sum games. If all player preference scales were part of the
inactive strategies that are not worldview player strategies and so formed
the code of conduct, we would call such decision processes purely skill
driven processes. These would include ordinary games in which the
effort available to each player is set as part of the rules of the game. It
also includes predator and prey situations where the resources are
constrained.
In the prisoner’s dilemma, we started from game theory but then
generated the converse possibility (Chaps. 7-9). We argued that the
prisoners could choose a code of conduct consisting only of the relative
strategic preferences. Furthermore we argued that the sum of the
preference scales of the two prisoners might also be part of that code of
conduct. Either way, this is close to the extreme case in which every
player relative preference s jk is inactive. We call such behaviors purely
resource driven processes. Here, cooperation is more important than
personal gain. Here, players are competing for resources rather than
optimizing for personal gain, so we expect the difference of the player
preference scales to behave dynamically.
If time is also inactive, the player payoff time components F j mt of
each player would be equal. Moreover, since there are no closed loop
flows that are entirely within the player subspace, there are no sources
that generate payoff fields orthogonal to that subspace, Cf. Eq. (2.13) and
Eq. (2.19). In simple language, there are no self-payoffs. We observed
this behavior in our numerical calculations in Chaps. 7-9.
We shall cover both resource driven behavior and skill driven
behaviors. We allow for mixtures of these two extremes and hence
illustrate all of the above possibilities. We provide illustrative numerical
Vorticity and Active Strategies 503
examples (Cf. Sec. 7.6, where for the prisoner’s dilemma, we adopted the
same goals). We start by specifying the system in a state with known or
assumed properties as in Sec. 13.9.
Let’s “start” the system on a spatial boundary that contains a still point,
defined as zero active acceleration, which facilitates our study of
whether it stays at that point or not. We shall refer to the behaviors at our
“starting” point as the known behaviors. These behaviors provide the
boundary conditions for the time independent differential equations of
decision process theory. The equations include the phasor solutions
characterized by a frequency . We are thus stating the conditions on a
boundary in the space of active strategies.
The general characteristics of a decision process in decision process
theory are contained in Eq. (4.5) and depend on the strategies we identify
as active and inactive in the normal-form coordinate basis:
We transform from one basis to another using Eq. (2.43) and compare
terms.
We can also change basis to one in which we formulate each decision
process identifying inactive strategies as those associated with players as
well as strategies that are inactive based on an agreed code of conduct.
We call this the symmetric normal-form coordinate basis whose form is
identical to Eq. (14.3), but the labels are different. We use the normal-
504 Geometry, Language and Strategy—Vol. 2
dx
V . (14.5)
d
We normalize the flow to unit length, Eq. (7.9) and Eq. (14.4):
dx dx
ˆ 1. (14.6)
d d
1 2 rc u1 u2 u3 .
U 1 U 1E E
E 1 . (14.7)
ˆ U 1 U 1E E U 1 U 1 E E
We go through each vector and use the same algorithm. Each new vector
must be orthogonal to all previous vectors and have unit length relative
to the initial metric. This process can be done in any frame of reference.
The last strategies will be the active strategies uk . In the symmetric
normal-form coordinate frame, these strategies are orthogonal to each
inactive strategy. In picking the associative orthonormal directions k ,
we set E jk 0 E jk 0 for each inactive strategy j .
Since the initial active strategy flows are known, to complete the
specification of the flow vector, we need the magnitude of the time
component of the flow V 0 V t m0 and the player inactive strategy
flows. We start the inactive flows at zero. We start with zero player
engagement (charge). This is not unreasonable if we assume that we start
at an equilibrium position qk 0 .
506 Geometry, Language and Strategy—Vol. 2
E j q E j 2 E j
. (14.8)
e q e 2 e
We also provide the equation for the player engagement e . If there are
non-zero player interests , the player will develop a non-zero
coupling away from equilibrium. Moreover, when there are two or more
active strategies, there can be a non-zero player passion p , whose
equations are part of the summary in the next section.
Given the initial flow directions and the values of the complete
symmetric co-moving orthonormal set at some initial boundary, we
determine the symmetric co-moving orthonormal frame at all other
positions from the equations in Table 5.1:
E j E j . (14.9)
For diagonal solutions to the player bonding, if we start with the inactive
direction j aligned with its co-moving counterpart j , this alignment
remains at all other values even though the scale is not preserved. These
equations provide coupled differential equations in the active strategy
directions k and can be uniquely solved once the orientation potentials
are determined, which we deal with next.
The form of these equations is based on the player fixed frame model.
The pressure components are:
0 h
h 1 h
p
n 1
q 2 1 2
1
With two or more active strategies, we have non-zero player passion and
player payoff fields, which were absent in the single strategy models.
In addition to gradient equations, we also have the “curl” equations
given in Tables 5.4 and 5.5, respectively, which were zero in the single
strategy models:
0,
q 2
q 2
q 2 0. (14.12)
For three or more active strategies, it follows that if the player payoffs
and player interests are initially non-zero, in general the composite
payoff gradients will also be non-zero. For two or less active
strategies, the above equations are identically satisfied and provide no
additional constraints (Sec. 14.10). They become important only for three
or more active strategies.
Thus the payoffs provide the key indication of vortex type behavior.
The result for three or more active strategies is therefore of particular
interest. The composite payoff reflects the decision vorticity
Vorticity and Active Strategies 509
DF Ea
e Ea Ea Ea . (14.13)
For each proper player , the player payoff field describes the amount
of rotation the player sees as their worldview changes.
In general, we see from the first line of Eq. (14.12) that the player
payoff for one player might influence the payoff of another through the
player bonding. However in the player fixed frame model, the coupling
can be diagonalized. This means that for certain “ideal” players,
they each determine their payoffs independently. We shall say more
about this in Sec. 14.9.
must be accounted for. Stresses in general give rise to strains and vice
versa. The bond matrix is an example of strain.
Based on the player fixed frame model, the bond matrix is diagonal.
This is accomplished with diagonal inertial components Eq. (6.20):
0
0 . (14.15)
p
h
h 1 h p
n 1
q 1 2 1 2
1 2 3 3 2 0. (14.16)
We require:
h
h
1 h
0. (14.17)
n 1
The equations summarized in the last section hold for any number of
active or inactive strategies. Since we will spend some time with three
spatial dimensions, it will be convenient to develop the formulas with
that in mind.
Therefore, we focus on three active strategies, na h 3 . For now
we leave the number of inactive strategies ni h variable. Let
n na ni be the total active and inactive strategies. The number of
inactive strategies is then h n 3 . For three active strategies, we use
the notation 1 x, 2 y , 3 z .
For each of the payoff fields we use a notation that is common and
helpful in physics for an antisymmetric tensor with three spatial
dimensions. There are three payoff fields B1 23 for each (proper)
player , with the cyclic choices of the active indices.
Similarly there are three decision composite payoff fields based on
1 23 and its cyclic permutations. We define 3-vectors as follows
for the player payoff vector Bα and composite payoff ω , along with
relevant scalar fields:
B α 23 3 1
, 1 2
ω 23 3 1
,
1 2
I 2
α 1 22 23 ,
P p
α 1 p2 p3 ,
q q1 q2 q3 q,
Θ 1 2
3 ,
σ
αβ 1 ,
2 3
512 Geometry, Language and Strategy—Vol. 2
,
h . (14.18)
ni
14.9.1 Pressure
n 1
ni p q Θ 1 2 1 Θ Θ 1 2 σ αβ σ αβ
n 1 ni
3 4 I α I Bα Bα 3ω ω .
α (14.19)
14.9.2 Acceleration
The acceleration terms are also similar. From Eq. (14.10) we have:
p
q q q q Θ 1 2 I α I α 2ω ω p . (14.20)
n 1
We use the fact that the “curl” of q is zero. These equations differ from
single strategy model calculations because the acceleration is generated
from a potential and the decision composite payoff term has been added,
whose sign is opposite to the pressure and energy density contributions.
This shows that the decision composite payoff reduces the strength of the
inertial term. In the numerical computations, we shall see when this
effect is important.
1
Θ 1 2 σαβ σαβ 1 2 1 Θ Θ 3 4 Iα I Bα B 3ω ω .(14.21)
α α
ni
We use the fact that the “curl” of Θ is zero. The energy density, player
payoff, player shear and player bonding all contribute an attractive force
for the player bonding potential. The player interest I α and composite
payoff ω contributions are repulsive.
We can split the bond player bonding matrix into two terms: the above
player bonding scalar and the bond shear tensor, whose equation is
determined directly from Eq. (6.24):
514 Geometry, Language and Strategy—Vol. 2
σ αβ q Θ σ αβ 1
ni h . (14.22)
We use the fact that the “curl” of σ αβ is zero. The scalar potential for the
shear has as its source the stresses associated with each player.
For steady state solutions we impose the constraint that the vector
components of shear commute as matrices in the inactive spaces. One
way to achieve that is to transform the known values of the shear vectors
σαβ 0 to a frame in which they become diagonal, σ αβ . In that same
frame, set the initial conditions of the transformed stresses to be
diagonal. With these initial conditions set, transform all the equations
back to the original frame. Since in the transformed frame, the shear
solution to Eq. (14.22) will remain diagonal, the resultant shear in the
original frame will commute: there is a constant transformation matrix
that diagonalizes the shear at all points. This will be useful in simplifying
some of the equations below.
Note that we have started with a simplified model Eq. (14.15) in
which the stresses are zero. We can still have a non-zero transformation
for the shear components. We are free to consider more complicated
models in which the stresses are not zero. The strategy for solving these
models is the same.
We see qualitatively new behaviors with the equations for the gradient
and “curl” for the player interest Iα :
n 1
I α 4ω B α σ αβ I β 2q i Θ Iα
ni
. (14.23)
1
I α q Θ I α σ αβ I β
ni
equilibrium is that the player payoffs align with the composite payoff.
This generates strong player interest, passion and engagement, where
player passion is the mixed stress components Pα p .
The player payoff behavior is specified as well from Eq. (14.10) and
Eq. (14.12) for the “curl” and gradient:
n 1
Bα Pα ω I α σ αβ Bβ q i Θ Bα
ni . (14.24)
1
Bα Θ Bα σ αβ B 0
β
ni
The gradient and “curl” equations for the player passion follow from
Eq. (6.18):
ni 1
Pα q Pα Θ Pα σαβ P β
ni
. (14.25)
1
Pα q Θ Pα σ αβ P β
ni
516 Geometry, Language and Strategy—Vol. 2
The gradient and “curl” equations for the composite payoff are:
ω I α Bα q ω
. (14.26)
ω Θ 2q ω
There are six active stress components in Eq. (14.11), which are
determined from the above equations:
1
p q q q
ni
2 B B 2Bα Bα 2 2ω ω
p
1 2 I I p . (14.27)
n 1
The first two contributions are the stresses between players and reflect
the overlap of the player payoffs and the overlap of the player interests.
In other words, when players share common ground there will be
stresses.
E j E j e
1
E j σ α β E j ΘE j . (14.29)
ni
1
e I α qe Θe σ α β e
ni
We see that if the pure inactive strategy j is initially set along the
inactive strategy j , the pure inactive strategy continues to be along that
strategy.
For each player, surfaces of constant player bonding generate the
same frame transformation. We also reconfirm our understanding that the
player interest is directly related to the gradient of the player
engagement for that player. We can use this information to determine the
remaining variables.
A 2e q a1 2e q a2 2e q a3
2a q
. (14.30)
f A 2e q ω e Bα
A 0
1 2 e q 2 A e Pα ω Θ 3q e Iα
2
Bα I α e q 1 e Θ 2σ αβ e . (14.31)
ni
The seasonal potential for the averaged magnetic fields creates a frame
rotation in the central co-moving holonomic frame and provides the basis
for taking co-moving frame derivatives with respect to the proper active
strategies, Eq. (6.58) expressed using normal partial derivatives (that
mutually commute):
. (14.32)
2a q
1 e e
4a a 4a q q q 2 2 x a x a
2 2a q x a q Θ x a (14.33)
2Θ a 6a q q Θ q 2q q x a 0.
We then use the harmonics of Sec. 6.3 to solve these partial differential
equations for the position scalars x a x, y , z , based on the scalar
potentials that are independent of proper-time.
We add to our set of partial differential equations coefficients that
depend only on the proper active strategies. We have demonstrated this
strategy earlier in Chaps. 7-10. This is the logical generalization to the
case of two or more spatial strategies. We call these n 1 dimensional
models, where n is the number of spatial dimensions. These equations
appear to have a dependency on proper time that might be hard to
eliminate. Though we showed this not the case in Chap. 6, a more direct
approach is to go to the central co-moving frame.
520 Geometry, Language and Strategy—Vol. 2
x a 1 e e 4a a e 2 q x a 4e q a x a
q Θ x a 2e q a q Θ x a 0 . (14.34)
U 0 a q Θ U 0 a
2 eq a q Θ V0 a 4e q a V0 a 0,
V0 a q Θ V0 a 0,
U a 2 e 2 q 1 e e 4a a U a
q Θ U a
4 e q a V a 2 e q a q Θ V a 0,
V e
a 2 2q
1 e e
4a a V a
q Θ V a
4 e q a U a 2 e q a q Θ U a 0. (14.35)
The solutions are attenuated frame-waves that propagate from the initial
conditions as a function of central time, Sec. 6.10, Ex. 37. The
attenuation is determined, among other factors, by the acceleration and
player bonding q Θ . The equations can also be written as a single
equation using complex numbers. The content of this equation is then the
magnitude and phase of the solution, called the phasor.
Vorticity and Active Strategies 521
14.9.14 Waves
2 q q q
. (14.36)
x z
We have dropped all of the terms except the quadratic term. From our
single strategy models, we expect that the acceleration flow moves
purely along a single (say the z ) direction. We expect no movement
along the other (say the x ) direction.
If this other direction is active, we can have movement along that
direction. Part of the effect follows directly from the behavior of
Laplace’s equation:
2 q 0 . (14.37)
q x, z q xm , z : m N , N qm
. (14.38)
qz x, z z q xm , z : m N , N z qm
1
x q qm 1 qm1 x qm
2
. (14.39)
1
x x q 2 qm 2 qm2 2qm x x qm
4
B xz
xz
I α 2x 2z
Pα p x p z . (14.40)
q qx qz
Θ x z
0
We see new behaviors in the pressure, Fig. 14.5 and the acceleration
Fig. 14.6 when we change the boundary conditions to allow structure in
x , which we do in Ex. 9. In this example, we assume that the composite
payoff and player payoffs are small.
The structural changes in the acceleration flow, Fig. 14.7 show the
onset of the lateral flow we identified as possible in the Laplace
equation, Fig. 14.1 and the non-linear Laplace equation Fig. 14.2. The
single strategy model assumes no lateral flow. We hope to learn the
consequences of these effects.
The small pulse in the acceleration generates effects in all the other
scalars. For example, the player interest for player 1 is modified,
Fig. 14.8, so that there is more flow along the center x 0 than at the
edges. If the pulse is turned off, these streamlines become uniform flows
independent of x .
The player passion flow acts like a current that generates the player
payoff. The assumption of ownership shows that this flow moves in the
x, z direction producing a payoff for player 1 that moves along the y
axis (Fig. 14.11); the corresponding behavior for player 2 has a payoff
that moves along the x axis. We noted that the common cause of the two
players is a circulation around the z axis.
This suggests that a code of conduct could be added with the same
basic behavior (Ex. 14). In this case we obtain a new player payoff,
Fig. 14.17. This modifies the composite
payoff, resulting in Fig. 14.18.
The player passion for player 3 that
generates this payoff, Fig. 14.19, has
essentially a flow along the z axis at the
center of the x, y plane. To conserve the
flow, we see that there is an inward flow
towards this axis all along the z direction.
The player interest, Fig. 14.20 is different in
that there is a non-zero initial gradient so that
the code of conduct payoff contributes to the Figure 14.19. Passion vector
composite payoff at z 0 . field P3 with code of conduct
532 Geometry, Language and Strategy—Vol. 2
14.15 Outcomes
The student will have learned how to identify and apply known
conditions to the decision process field equations for 2 1 , 3 1 and
higher dimensions. In the process, the student should be able to identify
those scalar fields that are steady state in the steady state models and
those fields that depend on proper-time. At a high level the student
should understand in what sense the flow equations generates “vorticity”
solutions that are harmonic and distinguish between the harmonic
components that are part of the initial conditions and vorticity
components that are intrinsic and set by the composite payoff, player
payoffs, engagements, entitlements, passions and interests of the decision
process.
Vorticity and Active Strategies 533
14.16 Exercises
(1) Show that for the conductivity model, the passion vectors in
Eq. (14.25) can be determined in terms of scalars using Sec. 6.10
Ex. 14, Pα P P :
n 2
2 P i h 2 P . (14.41)
ni
(2) Show that the integrating factor technique can also be used to write
Iα I i , Eq. (14.23) and ω , Eq. (14.26) in terms of
scalar potentials and Bα B A Eq. (14.24) in terms of vector
potentials. Determine the relationship of i to the player
engagement e .
Vorticity and Active Strategies 537
(3) We expect new and interesting features with three or more active
(spatial) strategies. Nevertheless, there are useful things to learn
from models with two active spatial strategies. Show that in the case
of Eq. (14.15) and zero shear 0 , the acceleration, player
bonding and player interest behaviors from Eq. (14.10) can
be expressed in terms of I α 2x 2z , q q1 q2 ,
Θ x z , Pα p x pz , B xz and xz as
follows, showing that the player composite payoff and player payoff
fields are not coupled:
x z ,
h n 2 ni ,
1
e I α q Θα e ,
ni
n 1
I α 4 B 2q i Θ Iα ,
ni
1 1
x I z z I1 qx x I z qz z I x ,
ni ni
p
q q q q Θ 2 2 1 2 I α I α p ,
n 1
n 1
1 2 i Θ Θ 3 4 I α I α 3 2 B B . (14.42)
ni
(4) With the same set of assumptions as Ex. 3, show that the player
passion, player payoffs and composite payoff are determined in the
ownership model by:
Pα P ,
n 1
Pα q i Θ Pα ,
ni
n 1
z B qz i z B I z x P ,
ni
538 Geometry, Language and Strategy—Vol. 2
n 1
x B qx i x B I x z P ,
ni
z z 2 q , x x 2 q ,
w exp 2q . (14.43)
(5) With the same assumptions as Ex. 3, show that the average pressure
is determined by:
ni 1
q Θ 12 Θ Θ 3 4 Iα Iα B B 3 2
ni
p . (14.44)
n
ni
n 1
(6) With the same assumptions as Ex. 3, show that the active and
inactive pressure components are:
n 1
p B B 1 2 I α Iβ . (14.45)
(7) With the same assumptions as Ex. 3, show that the equations reduce
to the following generalization of the single strategy model when
the player payoffs and composite payoff are zero:
n 1
I α 2q i Θ Iα ,
ni
1 1
x I z z I x qx x I z qz z I x ,
ni ni
1
q q q q Θ 1 2 I α I α p 1 ,
n 1
Vorticity and Active Strategies 539
ni 1
p 1 2 Θ Θ 3 4 Iα Iα ,
ni
Pα 0 . (14.46)
(8) Using Ex. 7, generate the pressure Fig. 14.3, and acceleration curves
Fig. 14.4, using n 6 spatial dimensions, the implied ni 4
inactive dimensions, 1 , resiliency 5 , 3 mesh points for x
in the interval 1, 1 , z in this interval and the following initial
values:
q x,0 qz x,0 0
x ,0 z x,0 0
. (14.47)
I x x,0 e x,0 0 0 0 0
I z x,0 4 4 0 0
(9) Building now on Ex. 8, adding back the possibility of vorticity and
player payoffs, provide structure in x with x dependent boundary
conditions and an increase in the number of mesh points to 11 as
given below, and a resultant smaller interval z 0.65, 0.65 ,
generate the pressure Fig. 14.5 and acceleration surfaces Fig. 14.6 as
well as the streamlines for the acceleration flow Fig. 14.7 and player
interest flow Fig. 14.8. Do the results depend sensitively on the
number of mesh points above about 10? Note that the boundary
conditions for the composite payoff (up to a scale), the player
interest and for the ownership model, the (gradient of the) player
passion are determined by the field equations on the boundary.
1
I x x,0 x e x,0 qx x,0 x x,0 e x,0 ,
ni
z P x,0 x B x,0 I x x,0 x,0
n 1
qx x,0 i x x,0 B x,0 ,
ni
540 Geometry, Language and Strategy—Vol. 2
x
q x,0 1 5 cos2 ,
2
qz x,0 0 ,
x,0 z x,0 0 ,
e x,0 0 0 0 0 ,
I z x,0 4 4 0 0 ,
B x,0 P x,0 0 ,
x,0 110 exp x,0 2 q x,0 . (14.48)
(10) Using the same model as in Ex. 9, and making the changes below to
the initial conditions, show that the player interest (Fig. 14.21) and
player passion (Fig. 14.22) are as indicated below using a mesh with
11 points.
q x,0 0
x 1 x
B x,0 110 cos2 , 10 cos2 ,0,0 . (14.49)
2 2
e x,0 0 0 5 5
1 1
Figure 14.21. Interest flow I1 for Figure 14.22. Player passion P1 for
Ex. 10 Ex. 10
Vorticity and Active Strategies 541
(11) Using the same model as in Ex. 9, and making the changes below to
the initial conditions, show that the player interest (Fig. 14.23) and
player passion (Fig. 14.24) are as indicated using a mesh with 11
points.
q x,0 0
. (14.50)
e x,0 1 20 sin x 1 20 sin x 1
20 cos x 1 20 cos x
Figure 14.23. Player interest I1 for Figure 14.24. Player passion P1 for
Ex. 11 Ex. 11
(12) The various factors can lead to complex behaviors. Using the same
model as in Ex. 9, and making the changes below to the initial
conditions, show that the player passion for player 1 (Fig. 14.25)
and player passion (Fig. 14.26) are as indicated below using a mesh
with 11 points. This shows that they can be chosen to be exactly out
of phase using the player payoffs and can be said to provide each
player, ownership of the stresses along his or her flow direction,
which as shown are independent.
q x,0 0
e x,0 0 0 1
10 110 . (14.51)
B x,0 110 cos2 1 2 x, 110 cos2 1 2 x 1 4 ,0,0
542 Geometry, Language and Strategy—Vol. 2
Figure 14.25. Player passion P1 for Figure 14.26. Player passion P2 for
Ex. 12 Ex. 12
payoff, Fig. 14.18. We also see new behaviors in the passion for
player 3, Fig. 14.19 and in the interest, Fig. 14.20.
e x, y ,0 0 0 1
5
. (14.53)
P x, y ,0 1
10 cos 2 1
10 x 110 cos2 110 y 1
10 cos2 1
10 x cos2 110 y
Chapter 15
Game Equivalency
545
546 Geometry, Language and Strategy—Vol. 2
In Sec. 15.6, we use the fact that we can write the payoffs and their
potentials in terms of a symmetric metric to deduce values for metric
components. We use these results to translate the game payoffs into
potential field gradients, Sec. 15.7, and entitlement payoffs, Sec. 15.8.
The player interest or game theory value helps set the player
engagement, Sec. 15.9. New aspects of decision process theory not
covered by game theory are the cooperative payoffs, Sec. 15.10 and the
seasonal vector potentials Sec. 15.11, which lead to natural resonances.
We return to the inactive strategies in Sec. 15.12 with an explicit
accounting of scale invariance for two-person games with three active
strategies. We summarize how to set initial values for such models in
Sec. 15.13.
15.1 Equivalency
of the focus strategies if we wish. We can specify that the payoffs have a
functional dependency on the skill strategies if we wish. For equivalency
as used here, we require that the dependency on the skill strategies be
periodic, which assumes that for each skill strategy, there is a distance
specifying the period. This defines the initial boundary.
With these assumptions and the further assumption that the
dependency on the skill strategies remains periodic, it is plausible that
the partial differential equations have unique solutions. We have
specified the equations on the initial boundary that consists of the closed
focus boundary with arbitrary skill strategies and the periodic boundary
for each of the skill strategies for any value of the focus strategies. We
find in practice that this leads to unique solutions.
Our practical set of cases will be two-person games where each
person has two strategies. There are four game strategies. The focus
subspace is one dimensional and the focus boundary is the point z 0 .
Let’s call the skill strategies x, y and the scale invariance strategy r .
We specify every scalar x, y , z on the initial boundary as follows.
For z 0 we specify an arbitrary function x, y ,0 subject only to the
constraints that the function is periodic in the skill strategies. For every
z 0 we impose the boundary condition that the solutions remain
periodic in the skill strategies. The partial differential equations are then
expected to yield unique results for this scalar function. The same holds
true for vector and tensor quantities.
We now identify those functions that can be specified by game theory
and hence invoke a game equivalency. A necessary ingredient of
equivalency is that in some sense, the competitive forces are the only
forces that matter. Even with just competitive forces, for multi-player
games, we have a form of cooperation through the focus strategies. Other
than that, there should be no other forces. We explore this in the next
section.
Suppose there are only competitive forces. First, we are assuming that the
payoffs are independent of time and that the flows are independent of
550 Geometry, Language and Strategy—Vol. 2
dV b
g ab Vk Fabk V b . (15.1)
d
552 Geometry, Language and Strategy—Vol. 2
15.3 Engagement
V j E j e E j . (15.2)
Given that we align the frames initially in such a way that the frame
transformation is the unit matrix, the initial boundary value of the charge
is the (negative of the) player engagement.
For game equivalency, the active acceleration is Q 0 . In addition,
the co-moving frame is aligned with the normal coordinate frame
(indicated by ). Therefore we obtain constraints based on Eq. (4.74):
f j
2 q E j E j e E j e E j 1 E k
Ek
. (15.3)
f j
2 E j 2 E j
q 2 e e e
Q 0
1 e e
2
These relationships relate the player interest to the player strategy bias
on the initial boundary:
f j
k
2 j e 2 e e e j I j I e e j
1 e e k
j
e j f
2 e e e I . (15.5)
1 e e
2
We rewrite these relations to express the player interest at the still point
in terms of the game value, player bonding and player engagement.
j f j e jVk f k
I 2 j e . (15.6)
1 e e 1 e e 2
We conclude that when the player engagement is zero, the player interest
is set by f j .
We use this result and Eq. (4.69) to determine the player engagement
gradient in terms of the game value and player bonding on the initial
boundary:
e j I j q e j j e
f j
. (15.7)
e j
j e e j e e
1 e e
15.4 Entitlement
“But Lareau means it in the best sense of the term: ‘they acted as though they had a
right to pursue their own individual preferences and to actively manage interactions in
institutional settings.’”
j
f j
2 E j 2 E j 2 2 e j . (15.8)
bonding matrix is a diagonal matrix, in which case we say that the
decision process is neutral: the frequency of choice between two pure
strategies is equal.
For a neutral decision process, player entitlement means that there is
no variation of the player interest or the player engagement. We see that
the physics analogy was helpful. If there is no charge, there will be no
electric field. If there is no player engagement, there is no player
interest. Conversely, if there is non-zero player interest, then we expect
that a non-zero player engagement will develop in the focus subspace. In
general the decision process is not neutral; we allow for non-diagonal
solutions and obtain non-zero player interest.
At the still point, we think it may make sense to take the player
engagement to be zero for worldview players as a way to investigate the
effects of the player interest. This agrees with the common sense notion
that at the still point, the players need to pay attention only to their view
of the world. Also, their choice of player engagement is optimal. Though
this makes sense for the game theory players, it does not make sense for
the code of conduct players. As we shall see with the numerical
examples in the next section, in general we expect the still point player
engagement for the code of conduct players em to be non-zero and
relatively large as the value is set by Nash equilibrium values.
ˆ jk jk
ˆ ja jk Aak . (15.9)
ˆab g ab jk Aaj Abk
j, k 1 , 2 ,
m i1 , i2 ,
a , b t , u1 , u2 , u3 . (15.10)
aˆ jm a jk A jk a A a jk A jk F
k
m
k
m
k
m
k
am
uˆ jm m jk F um
k
tˆ jm m jk F ktm
. (15.11)
tˆ ju uˆ jt m jk F ktu
uˆ ju uˆ ju m jk F uk u
We now have two views. One characterized by Eq. (15.11) and the other
by Eq. (15.13) below. The link between the two views is the left-hand
side of each equation. This allows us to take values in one view and
compute the consequences in the other view. The two equations have
been idealized. There will be small changes as a result of the
orthogonalization process because the energy flow is not identically
along the time direction. For the discussion below, we ignore these
effects but do take them into account in our numerical calculations.
We relate the player payoffs to the gradients of the potential fields.
For the symmetric normal-form coordinate basis, we have the same
normal-form holonomic metric potentials ˆ , which are expressed in
the form Eq. (15.9), with the below noted differences in active and
inactive variables and a different tensor A j a for the payoff potential:
a , b t , u1 , u2 , u3
j, k 1 , 2 ,, i1 , i2 ,
ˆ jk jk . (15.12)
ˆ ja jk Ak a
ˆab g ab jk A j a Ak b
562 Geometry, Language and Strategy—Vol. 2
j, k 1 , 2 ,
m i1 , i2 ,
u, u u1 , u2 , u3
u jm m jk F kum . (15.13)
0 t jm m jk F ktm
tˆ ju uˆ jt m jk F tuk F k ut F utk
uˆ ju uˆ ju m jk F kuu F k uu F uu
k
We take the behaviors that we have determined and translate them to the
symmetric co-moving orthonormal coordinate basis. We start with the
Game Equivalency 563
j
co-moving player strategy bias field f and co-moving player payoff
field f j from Eq. (4.74):
f j
2 q E j E j e E j e E j 1 E k Ek
f j
E a E b F j ab F j tu . (15.14)
f j
E j
E j
E E F
a b j
ab F j
uu
f j
12 B3 E j 3 E j E a1 E b2 F j ab F j u1u2 ,
αj
B 3 e j ω3 F uj1u2 ,
& cyclic permutations. (15.15)
Using the concept of player entitlement above, we see that for worldview
players we have e j 0 and so the assumption here is that on the initial
boundary, worldview players make decisions based only on their
entitlement payoff.
We find that this assumption is not particularly restrictive since in
reality, non-zero player interest generates player engagement in the
neighborhood of the initial boundary. Rather, this suggests how we
isolate the player entitlement payoffs by finding regions where the player
engagement might be zero.
1
For the numerical work, we don’t make this approximation but use the vectors as
determined by the orthogonalization process and take the exact expressions for the scalars
in terms of the payoffs and other known quantities, imposing all necessary constraint
equations. The approximations here are only to provide insight into the choices that are
subsequently made.
564 Geometry, Language and Strategy—Vol. 2
j 1 , 2 ,
1 , 2 , . (15.16)
f j
I
j
2e
j
1 e e F
j
tu
We note (Sec. 15.3) that the player engagement is a distinction that goes
unnoticed in game theory. A multiplicative factor in the payoff or utility
makes no change in the strategic content of game theory. When these
payoffs are not the same, the player engagement factors can make a
difference and in decision process theory, can change the Nash
equilibrium. In game theory, this is viewed as a change in the player
utility scales, which may be considered as not comparable.
It was suggested by Thomas & Kane [2010] that such differences
might distinguish players that are egoists from altruists. Such a
difference might provide a means to distinguish decision process theory
from game theory. We believe that the player engagement is a natural
concept and one that expands on the meaning of the value a decision has
to a player.
To measure the player engagement, we need two steps. First we need
to know the absolute utility function so that we have knowledge of the
unit of measure for the player payoffs F k ab . Then we need to measure
the product by looking at the consequence of the force Eq. (15.1) that
yields oscillatory behavior whose frequency is determined by the product
of the player engagement and the player payoff, Vk F k ab . One measures
the frequency. This approach is superior to simply setting the player
engagement to zero. The oscillations represent natural seasonal or cyclic
behaviors that characterize the decision process (Sec. 15.11).
Though we have the option of setting the player engagement to zero
for worldview players, this is not typically an option for institutional
player. Nash equilibrium determines a flow, which with game
equivalency becomes an inactive charge and hence the player
engagement for the institutional player.
When there are both worldview players and institutional players, we
use the same forms Eq. (15.7) to relate payoffs and player engagement to
the co-moving player bias f j , with the appropriate change in meaning
566 Geometry, Language and Strategy—Vol. 2
for the variables since we now have fewer active strategies and more
inactive strategies. We need information about the cooperative player
bonding components .
We get insight on these player bonding components by looking at the
behavior of jm . This behavior can be derived from Eqs. (15.13) and
Table 5.1 (Cf. Ex. 23 from Sec. 4.11, Eq. (4.85), specialized to the
streamline solution):
jm m jk E u F kum m jk F kum
.(15.17)
jm 2q E j Em 2 E j Em E j Em 2 E j Em
k
At the still point, the player payoff F um for the worldview players
determines the components of the player bonding that lie along the
mixed axes of the worldview player directions and the institutional
player directions:
j , k 1 , 2 ,
m i1 , i2 ,
2
j
F um 2 q e j e m 2 j m h j m I j e m Im e j
ni
e j 0 and u focus direction . (15.18)
j
F um 2 j m I j e m
j f j
I 2 j e
1 e e
j em F tuj
j
F um 2 j m 2 e e m
1 e e
In general, the bond shear tensor does not need to be a diagonal matrix.
It is essentially the co-moving frame player j payoff between an active
strategy u and an inactive code of conduct strategy m . The bond shear
is determined by the player game value and the player payoff between
the scale invariance direction and one of the player skill preferences.
The bond shear generates cooperative forces.
Game Equivalency 567
Let’s take an arbitrary player bonding matrix at the still point and
transform to a frame in which the matrix is diagonal. In this basis, we
have defined a set of players that act as if there is no cooperation
between them. Using this form, we compute the cooperative payoffs jk ,
Eq. (4.70): the initial metric and its gradients are diagonal, which is
consistent with our notion of what it means to be an independent player
or agent.
The diagonal nature of the player bond components maintains that
notion at all other strategic values away from the still point. If we
transform back to the original frame, these effects will not be as
apparent. Even though the player fixed frame model requires a fairly
strong constraint on the form of the bonding matrix, it still allows for
some non-trivial effects based on such coordinate transformations. We
suggest this effect becomes more complex if we generalize from the
player fixed frame model to one in which we impose only the condition
that time is an isometry: a general central frame model.
The possibility of gradients of the cooperative payoffs raises new
distinctions. We found in the previous section as well as in other
numerical calculations that the cooperative payoffs, Eq. (14.21) generate
locked behavior or strategic clustering that appears to us similar to
galaxy formation in the physical sciences. Because of this strategic
clustering, at the still point we have a higher energy density than
elsewhere. Because the energy density is assumed to be proportional to
the average pressure, we also see clustering of the average pressure. We
see a second cooperative shear effect in Eq. (14.22). This may associate
the strategic clustering to a specific player. We expect to see limits to
choice (Sec. 10.9) and limits to growth (Sec. 10.12) as we did in one-
dimensional models.
The effect of these steady state scalar fields is to determine the
decision flow V , along with the acceleration, player bonding and
vorticity tensors along that flow. We recall that the flow of decision
processes can be envisioned as the motion of a point in the space of
preference frequencies of players or agents of the decision process,
making decisions observable and measurable. When we have
568 Geometry, Language and Strategy—Vol. 2
j , k 1 , 2 ,
e j ek 0 . (15.19)
2
u jk h 2 u j k
ni u j k
DF Ea
Ea 1 2 I e Ea . (15.20)
We have inserted the player fixed frame model values, Eq. (4.49).
We learn something very interesting. The frame will maintain its
orientation along the streamline if we have small or negligible player
engagement and if we have zero composite payoffs. We propose that we
add the requirement of zero or small composite payoff to our set of
requirements for game equivalency. Even if the player engagement is
570 Geometry, Language and Strategy—Vol. 2
not zero, we see that the frame rotation is mainly along the flow
direction, which suggests that there is more flow but no change in
direction.
With this assumption, we get the following result from Ex. 2:
Az z Az 0
A 2e q a . (15.22)
f A 2e ω e B
q α
This sets the seasonal vector potential to be zero at z 0 . Since we have
a complete set of equations that determine the composite payoff ω and
player payoffs Bα , the seasonal payoff f is determined. We show in
Sec. 6.10, Ex. 13 that the “curl” of the seasonal payoff is zero, f 0 ,
as a consequence of the equations for the other fields. Hence the payoff
can be expressed in terms of a vector potential. With these constraints,
we only have to write the equation for the remaining two components of
the vector potential.
The first equation in the list below is imposed on the initial boundary:
x Ay x, y ,0 y Ax x, y ,0 f z x, y ,0 ,
z Ay x, y ,z f x x, y ,z ,
z Ax x, y ,z f y x, y ,z . (15.23)
The other two equations are differential equations that hold everywhere.
There is no ambiguity in setting the focus component Az to zero
everywhere since there are no conflicting equations. It follows from the
572 Geometry, Language and Strategy—Vol. 2
1 1
y1 r2 u2 , y3 r1 u1 ,
2 2
1 1
y2 r2 u2 , y4 r1 u1 . (15.24)
2 2
2
This assumption can be relaxed however. The assumption is however a convenient one
for gaining a preliminary understanding the results.
Game Equivalency 573
T j S j
vj :
2S j3
u1 u2 u3 r t
u 0 S j3 S j2 T j1 0
1
u S j3 0 S j1 T j2 0
F abj 2 j
. (15.25)
u3 S 2 j
S 1 j
0 T 3 j
2 v j m
r T j1 T j 2 T j 3 0 0
t 0 0 2vj mj 0 0
F j ab y3 y4 y1 y2 t
2
y3 0 1 T j 2 S j1 a j c j vj mj
y4 1 2 T j
2 S j1 0 b j d j v j m j
(15.26)
y1 aj bj 0 1
2 T j
1 S j2 v j mj
y2 cj dj 1 2 T j
1 S j2 0 v j mj
t v j mj vj mj vj mj vj mj 0
aj 2
4 T S T S T S
j
1
j
2
j
2
j
1
1
2
j
3
j
3
b
j 2
4 T S T S T S
j
1
j
2
j
2
j
1
1
2
j
3
j
3
. (15.27)
c
j 2
4 T S T S T S
j
1
j
2
j
2
j
1
1
2
j
3
j
3
dj 2
T S T S T S
4
j
1
j
2
j
2
j
1
1
2
j
3
j
3
The remaining variable is the scale m j , which sets the relative size of
the null vector for the payoff:
1 T j1 1 T j1 1 T j2 1 T j2
Nj , , , , m j . (15.28)
2 2S j 3 2 2 2
j j j
2S 3 2S 3 2S 3
We assume the still point is Nash equilibrium, so the scale parameter will
be the same for each player as well as the other components of the null
vector. Away from the still point, all these parameters may vary and no
single null point may exist for all payoff matrices.
The assumption of scale invariance is that the effort scale r is
inactive. The conservation of effort is analogous to the idea in game
theory of a zero-sum or constant-sum game. In this case, we are making
a dynamic assumption that no scalar or tensor in theory depends on the
effort scale. This means in particular that the payoff elements
F j ar a A j r r A j a are independent of that scale as well as the payoff
potential A j r . We see the constraint: if chose another direction a that is
also inactive, then the payoff is identically zero. With a single inactive
strategy, the player payoff F j ar a A j r can have any arbitrary value on
the initial boundary.
The payoff between two active strategies, F j ab a A j b b A j a can
take on any value on the initial boundary. The real distinction between
these payoffs and those that have one inactive direction is the spatial
dependencies. From differential geometry we determine that when there
is an isometry, the flow along that direction will be conserved, which is
not the general case. Of course on the initial boundary, we are fixed at
one point in time and so can’t observe whether a charge is conserved or
not.
Game Equivalency 575
j 1 , 2
j e r F tuj
F 2 j r 2
j
e e r
1 e e
ur
1 e
j
f j
z
I z j 2er zr
j
e
r r F 2mv
j
tu3 j
x, y : f j
I j 2e r 3 j 1 e e 0
r r
2 z j r I z j e r F j
u3 r T j
3
. (15.29)
2 x j r 1 e r e r T j1
2 y j r 1 e r re T j
2
e E j f
j
f j
j
xy B z F uj1u2 S j 3
j
B x F uj2u3 S j1
j
B y F uj3u1 S j 2
We have simplified the expressions by assuming that the players (not the
institutional player) are entitled.
After setting values for the co-moving scalar fields at the still point,
these equations determine the seven parameters for each player, which
we connect to game theory values if available (Cf. Exs. 11-17).
The equilibrium null vector N a associated with the transformed
payoff Eq. (15.25) is T j 2 S j 3 T j1 S j 3 0 1 m j . In particular the
flow V r along the scale direction N r and the player engagement at the
still point, e r e E r E r V r , are not zero either. Although this
charge is non-zero, we must inquire about the value of the player
worldview charges e . They determine the player engagement at the still
point. In our example for the Prisoner’s Dilemma, we assumed that
initially the player charges were zero (Sec. 7.9). We justified this by
suggesting at the still point that the players will behave as if they are
neutral or uncharged.
As discussed in Sec. 15.4, our justification for worldview players is
that these flows are zero based on player entitlement at the still point. It
is not a strong argument however. It may be more relevant to note that
Game Equivalency 577
the player engagement changes rapidly away from the initial boundary in
our solutions. Since we don’t have a good argument at this time for an
initial value, zero is perhaps the best start point since in any case we will
be looking at the behaviors in a neighborhood.
In decision process theory, we specify a great deal more information
that bears on the dynamic behaviors than we would specify in game
theory. In particular our co-moving frame values may extend the usual
discussion if for example there are factions or if the components
x , y : f j are not zero. Such changes leave the core game matrix
alone but may change the null vector. Our perspective is that the null
vector plays much less a role in decision process theory than in game
theory since the actual behavior is determined by the collective effects of
many fields, such as the seasonal payoff f .
Of course, it is plausible that under appropriate conditions the null
vector of the seasonal payoff and the null vectors of each player payoff
will be proportional. In the more general case, we still expect there to be
still points, steady state behaviors and non-steady state behaviors. Note
that Exs. 18 and 19 as forms for honoring or breaking contracts (Sec.
11.7, Ex. 1) are examples of this. All of these are consequences of
decision process theory.
15.14 Outcomes
In this chapter, the student will have learned the distinction between
strategies and preferences as well as between skill preferences and effort
preferences. The dynamic behaviors of decision process theory rest on a
set of initial boundary conditions and on time evolution equations.
The student will have learned that the initial boundary and the initial
boundary conditions are set in part by game equivalence, a principle that
relates many of the concepts and attributes of game theory to the
dynamic theory. This includes the payoff matrices from game theory,
which set the values of both competitive and cooperative fields.
There are some new terms not set by this principle: the player
engagement is new and is like “charge” in physics. The product of the
578 Geometry, Language and Strategy—Vol. 2
15.15 Exercises
(1) Develop the equations below and explain why one would expect
helical structures based on them and how they relate to helical
structures in [Vol. 1, p. 124]. Note that the structures assume that
the payoff and most but not all metric components are constant
along the streamline as seen in Eq. (2.49), which can be re-
expressed using Eq. (4.32) showing the possibility of helical
behavior because of feedback loops both between the time
components and space components as well as between the space
components:
E a q 1 2 jk E j E k Ek f k E a a bc E b E c ,
E a 1 e e E
1
2
k
f k q e E a
(2) If you ignore the bond forces abc , show that the time dependence
of the frame equations Eq. (4.32), which we simplified in Ex. 1, Eq.
(15.30), can be put into the form (Exs. 2-5 and Cf. Chap. 6, Exs. 19-
29):
X a E a x Ea y Eaz E a ,
1
1 E k Ek ,
1 e e
q 2 e e e
Q , Q Qx Qy Qz ,
1 e e
2
2 Ek f
1 k
1 2 e q f e 1 2 e q f ,
1 2 Ek f k 1 2 e q f , Ω yz zx xy ,
Game Equivalency 581
0 z y 1Qx
dX a
z 0 x 1Q y a
X . (15.31)
d y x 0 1Qz
Qx Q y Qz 0
X a Aa X 0 e
0 z y 1Qx
z 0 x 1Q y
X 0 X 0 . (15.32)
y x 0 1Qz
Qx Q y Qz 0
Ω Ω QQ 4 1 QΩ
2 2
ΩΩ 1QQ 1
2
X 0 1
2 X X ,
0 0 X 0 1
2 X X X ,
0 0
0
X 0 i X 0 i X i Re X 0 i X 0 i i Im X 0 i ,
*
0
A i A i A 1 2 Aa Aa ,
a a * a
(5) In Ex. 4 we covered the case that the eigenvalues of Eq. (15.31) are
not degenerate and there are no tidal forces. We assumed that the
active acceleration Q 0 was zero: the rate of change of the flow
(in the normal-form coordinate basis) vanishes,
E a a bc E b E c 0 . We call this the still point. We define free
fall behavior as being at the still point with no active geometry tidal
forces a bc 0 . Show that the eigenvector equations now simplify
since we have one solution with E a an arbitrary constant in
proper-time. Is the converse true: if this flow is constant and if the
active geometry acceleration is zero, will a bc 0 ? There will also
be one solution with E a constant, though constrained to be in the
null space of the rotation Ω . Show that the remaining
eigenfrequency is Ω . Thus show that the general solution for
the degenerate case is:
(6) Assume that a vector function A is given that satisfies Eq. (6.65) of
Ex. 13 of Sec. 6.10, A f , where A 2 e q a . We
impose boundary conditions on a surface z 0 that includes the
gauge condition A 0 . Assume that all the boundary conditions
are known on this surface except the vector potential Az , and the
gradient z Az . Assume that the direction z can be used to specify
Cauchy conditions for the partial differential equations. The partial
differential equations will then have a solution for each choice of
the vector potential Az . Using these solutions show that the gauge
function A satisfies the following partial differential
equation for any initial function Az :
J f
A J . (15.36)
z x, y , z x z Ax y z Ay J z x x Az y y Az
2 f x g y h k ,
M M
m 2 n 2 mn kmn i mf m m,n n mn
m M n M
M M
i
m M n M
ng m m,n n g mn
M M
m M n M
hm m,n n mn ,
(9) Show that the coordinate transformation that goes from Eq. (15.26)
to Eq. (15.25) is orthogonal U T U I and is given below:
F j U T F jU,
0 1
2
12 1
2 0
0 1 12 1 0
2 2
. (15.38)
U 1 2 0 1
2
1
2 0
1
2 0 0
1 1
2 2
0 0 0 0 1
Game Equivalency 585
(10) If there are no factions, there are no self-payoffs. In this case show
that the usual rules of game theory give the same game value and
strategy associated with Eq. (15.28). Furthermore, show the
following relationships, where the first two follow because there are
no self-payoffs. Prove that in game theory, for mixed strategies it is
necessary that S j 3 0 . Also show that in game theory, one can add
to the game matrix components a common constant so one can
always choose values so that T j 3 0 . Finally show that the mixed
strategy solutions are explicitly as shown:
2T j1 2 S j 2 1
2 ( a j b j c j d j ),
2T j 2 2 S j1 1
2 ( a j b j c j d j ),
S j3 1
2 ( a j c j d j b j ),
T j3 1
2 (a j b j c j d j ),
S j1S j 2
vj 1
2 T j3 ,
S j3
2S j 2 2S j2 2 S j1 2 S j1 j
N j 1 2 1 , 1 S j , 1 S j , 1 S j , m . (15.39)
1 1 1
S j 3
2 2 2
3 3 3
(11) Using the same transformation Eq. (15.38), show that the prisoner’s
dilemma model (Chap. 7) for player 1, Eq. (7.3), has non-zero time
components in the new basis and S 13 0 :
0 0 1
10 2
1
10 2
1
10 2m
0 0 9 10 2 9 9
10 2 10 2m
F 1
ab 110 2 9
10 2
0 1 7
5m . (15.40)
1
10 2
9
10 2
1 0 1
2m
1 0
10 2m 10 2m 5m 1 2m
9 7
0 9
10
1
10 0 9 10 m
9 0 1 9 9 10 m
10 10
F 1ab 10 1
1 0 1
10
9
10 m ,
0 10 10
9 1 0 9
10 m
9 9 10 m 10 m
9 9 0
10m 10 m
0 0 1
5 2
0 0
0 0 5 2 09 0
F 1ab 5 2
1 9
5 2
0 1 9
5m . (15.41)
0 0 1 0 0
0 0 5m 0
9 0
(13) Transform the game matrix from Ex. 3 in Sec. 12.7 using the
transformation Eq. (15.38) and show that it is given below. Also
show that the null vector before transformation is stated correctly
below:
0 30 5 2 5 2 0
30 0 5 2 5 2 0
520
F abj 5 2 5 2 0 170
3m ,
5 2 5 2 170 0 0
520
0 0 0 0
3m
1 2 1 2
N , , , ,m . (15.42)
3 3 3 3
(14) Transform the game matrix from Ex. 4 in Sec. 12.7 using the
transformation Eq. (15.38) and show that it is given below. Also
show that the null vector before transformation is stated correctly
below.
Game Equivalency 587
0 40 5 2 5 2 0
40 0 5 2 5 2 0
5
F abj 5 2 5 2 0 0
2m ,
5 2 5 2 0 0 0
5
0 0 0 0
2m
5 3 3 5
N , , , , m . (15.43)
8 8 8 8
(15) Transform the game matrix from Ex. 5 in Sec. 12.7 using the
transformation Eq. (15.38) and show that it is given below. Also
show that the null vector before transformation is stated correctly
below:
1 1
0 2 0
2 2
2 1 1
0 0
2 2
1 1 13
F abj 0 6 ,
2 2 2m
1 1
6 0 0
2 2
0 13
0 0 0
2m
3 1 1 3
N , , , , m . (15.44)
4 4 4 4
(16) Transform the game matrix from Ex. 6 in Sec. 12.7 using the
transformation Eq. (15.38) and show that it is given below. Also
show that the null vector before transformation is stated correctly
below:
588 Geometry, Language and Strategy—Vol. 2
1 1
0 2 0
2 2
2 1 1
0 0
2 2
1 1 3
F abj 0 2 ,
2 2 2m
1 1 2 0 0
2 2
0 3
0 0 0
2m
1 3 3 1
N , , , , m . (15.45)
4 4 4 4
(17) Transform the game matrix from Ex. 7 in Sec. 12.7 using the
transformation Eq. (15.38) and show that it is given below. Also
show that the null vector before transformation is stated correctly
below:
0 450 25 2 25 2 0
450 0 25 2 25 2 0
1300
F abj 25 2 25 2 0 150
9m ,
25 2 25 2 150 0 0
1300
0 0 0 0
9m
5 4 5 4
N , , , ,m . (15.46)
9 9 9 9
(18) Transform the game matrix for player 1 from Ex. 1 in Sec. 11.7
using the transformation Eq. (15.38) and show that it has the form
given below. Also show that the null vector before transformation is
stated correctly below. Are there other null vectors?
Game Equivalency 589
0 0 2z 0 0
0 0 2y 0 0
2 w
F 1ab 2 z 2y 0 x
m ,
0 0 x 0 0
0 2w
0 0 0
m
w w (15.47)
N 1 , ,0,0, m .
z z
(19) Transform the game matrix for player 2 from Ex. 1 in Sec. 11.7
using the transformation Eq. (15.38) and show that it has the form
given below. Also show that the null vector before transformation is
stated correctly below. Are there other null vectors?
0 0 2y 0 0
0 0 2z 0 0
2 w
F 2ab 2 y 2 z 0 x
m ,
0 0 x 0 0
0 2w
0 0 0
m
w w
N 2 , ,0,0, m . (15.48)
z z
Chapter 16
591
592 Geometry, Language and Strategy—Vol. 2
Eq. (15.25) can be used as illustrated in Ex. 15, Sec. 15.15, which also
provides the null vector (Nash equilibrium) Eq. (15.44). For this model,
we focus on game equivalence including an investigation of the free fall
behaviors. We finish our discussion of the attack-defense model in Sec.
16.5 with a discussion of gauge invariance for the seasonal vector
potential.
In Sec. 16.6 we define a social process using a work-wealth-wisdom
model that illustrates differences in our approach from game theory.
We use this model in Sec. 16.7 to illustrate streamlines when there are
only natural resonances and Sec. 16.8 when there are added forced
harmonics. This leads to a discussion of travelling waves in Sec. 16.9,
standing waves in Sec. 16.10 and striking behaviors in recurrence plots
in Sec. 16.11.
War games epitomize the short term concept of winning versus losing
whereas social decision processes are more about the effects of decisions
over time. This concept is not yet precise and is clearly an over
generalization. From decision process theory, there is some substance to
this idea. A necessary construct of the theory is the payoff associated
with each player, including institutional players associated with codes of
conduct. The payoff generates the concept of winning or losing and is
associated with Nash equilibrium. From this standpoint there is an
optimal strategy for behavior.
The payoff generates the concept of game equivalence in which there
is a natural or resonant frequency associated with the same payoff. We
make that concept more precise by identifying the payoff as the seasonal
payoff f for the 3 1 dimensional models we will be concerned with
in this and the next few chapters. The idea of a resonant harmonic
(seasonal harmonic) cycle implies that there is a time scale, the seasonal
timeframe, associated with the repetition of the event; the larger the
payoff, the shorter the time scale. These are natural harmonics as
opposed to the forced harmonics that are associated with the general
expansion Eq. (14.1).
We believe that these different possibilities generate different types of
questions. We have previously illustrated one such question associated
Two-Person Decision Processes 593
We recall that the attack-defense (AD) game posits that “Blue has two
installations. In normal form, he is capable of successfully defending
either of them, but not both; and Red is capable of attacking either but
not both. Further one of the installations is three times as valuable as the
other.” The attack or defense of the lesser installation is labeled “one,”
the other is “two.” The payoff matrix for blue is:
4 1
GBlue 1
10 3 4 . (16.1)
4 1
GRed 1
10 3 4 . (16.2)
We argue that if “Red” attacks the lesser of the two targets and it is
defended, “Red” loses 4 ; similarly if “Red” attacks the other target and
it is defended. The loss is 1 unit if the attack is made on the greater
value target and it is undefended and 3 for the other case. The payoff
matrix here is from “Red’s” point of view. For our numerical work we
also scale “Red” payoffs by 110 .
The scale factor changes the seasonal timeframe (Sec. 15.11) based
on the resonant harmonic frequency. The competitive aspects of the
attack-defense model (AD model) however, are independent of this scale
factor. In this case we have a zero-sum game. We would expect to see a
difference in the numerical results if we changed this into a constant sum
game.
One expected difference is that the limits to choice (Sec. 10.9)
narrows as we increase the scale factor and hence reduce the seasonal
timeframe. These results are different from the expectation in game
theory where nothing depends on a linear scale transformation. We
expect that the scale changes the “size” of the strategy space; the smaller
the scale the larger the effective space. We have scaled down by a factor
and get a space that has approximately unit size.
We lay out these behaviors in the following sections for the f0AD
model and our choice of parameters (see Ex. 1), which reflects purely
free fall behavior (Sec. 15.1). In the next few sections we elaborate on
this model, after which we consider a social model in Sec. 16.6. Of
Two-Person Decision Processes 595
particular interest in the social model will be a variant of the free fall
model where we add a forced harmonic behavior by means of an added
frequency component (f1WWW model). Some of the lessons learned
there also apply to the f0AD model here.
that have positive pressure and energy density, solutions with locked
behaviors. We find that we get positive values if we increase the
decision mass to md 40 . We are not forced to get a maximum
however. This gives us a minimum pressure at z 0 . To get a maximum
near the origin, we argue as follows.
Up to now we have assumed that the game for each player is simply
related to its player strategy bias field f j as in Eq. (15.29), allowing
for slight discrepancies due to the orthogonalization process. We have
assumed that the null space of the payoff vector is always related to the
Nash equilibrium.
This assumption need not always hold, though in our numerical
examples we maintain the assumption. We obtain locked behavior
consistent with this assumption because we have no information about
the game properties for the institutional player. However, we could have
changed the player strategy field for “Blue” and “Red”. This would
break our assumption about their relationship to game values. Thus, we
have reason to question that assumption.
We can choose the player strategy bias field independently from the
equilibrium flow, which leaves us free to set the form of the flow at what
we have designated as the still point. We are also free to choose the
player strategy bias field to conform to our expectations for the growth
of the player interest along the various strategic directions.
The order of indices has “Red” first and “Blue” second. The
proportionality constant is adjusted so that the flow is a unit vector.
distinct look for the player bonding (Fig. 16.3) and player engagement
(Fig. 16.4). In particular, “Blue” and “Red” exhibit relatively flat
engagement away from the still point, whereas the institutional
engagement becomes large, even though it is close to being “entitled” at
the still point.
In this section we provide suitable initial starting points for free fall
solutions (Sec. 15.15, Ex. 5). Conceptually, we start with the steady
state solutions Eqs. (6.81) and (6.82). These are from Exs. 34-35
associated with linear growth and resonant behavior respectively. Such
solutions are free fall solutions (Exs. 2-6), leading to the known-
condition-based behaviors, Fig. 16.5 and Fig. 16.6. The corresponding
streamlines are linear in the normal coordinate frame active variables
u1 u2 u t .
Figure 16.13. Initial phase space for characteristic potential for f0AD model
In Sec. 16.1, we suggested that war games epitomize the short-term idea
of winning versus losing whereas social decision processes are often
about the effects of decisions over longer timeframes. We understand
what it means to win or lose; the consequences are immediate. It is less
clear for social processes. Consider how issues are framed in the media
as an example. Winning and losing are very clear. What do we listen for
relative to social decisions?
For example, it may be relevant when we hear the word propaganda.
It might be in the context of promotion and sales. What kind of forces do
they represent? Propaganda is the ability to set the narrative; it is the
ability to change a person’s mind. Success in the arena occurs over long
timeframes and so fits with our suggestion above, that social decisions
are related to long timeframes. Let’s take this a step further and inquire
about the forces that are involved.
Two-Person Decision Processes 607
But, you may raise the valid objection that social phenomena are
inherently uncertain and intrinsically about decisions which generate
really discrete yes/no type outcomes. How can such events possibly be
described as part of a continuum? A choice at this instant does not
continuously flow from choices made in the past. It is a gamble and we
might in fact, if given the chance, make exactly the opposite choice if
provided the opportunity.
Yet even in physics we have phenomena at the quantum level that
from a certain point of view are uncertain and when viewed in a certain
way, appear to be discontinuous. That fact has not prevented us from
looking at them from a continuous point of view described by a
differential geometry in both time and space (Sec. 3.9). Indeed, we
suggest that game theory has provided us the perspective that social
interactions in fact can be viewed geometrically if we focus on the mixed
strategies and how they evolve in time as opposed to the pure strategies
that focus on the uncertainties [Von Neumann & Morgenstern, 1944].
We don’t focus on the actual decision, but on the mixture of decisions
that are possible at any point in time.
As an example, consider that over the last decade, wealth has
redistributed itself dramatically [Piketty, 2014 and Stiglitz, 2012]. It has
not happened discontinuously. It has evolved over time and appears to
change continuously across social strata. Some members of the middle
class have become wealthy whereas others have become poor. The
changes reflect a process, not a capricious set of changes. The process is
even more in evidence over long time scales (centuries). This example
might help focus our attention.
Suppose we analyze the flow of wealth and assume for simplicity it is
distributed between two distinct populations. If the populations are
valued similarly and if the interaction is zero-sum, we expect each to
receive the same payoffs in the sense that what one wins, the other loses.
But what if one population believes their payoff, if they win, is much
higher than the other population? We still achieve a zero-sum game if
we balance the product of the population and value for each side.
If it is agreed that one side is 10 times more valuable, then this works
if the other side has 10 times the population. What matters here is the
interplay between competition and propaganda, since there may in fact
Two-Person Decision Processes 609
Let us put these ideas into the format of decision process theory. We
imagine a model in which there are two players, the workers and the
wealthy: player 1 and player 2 respectively. The players have entered
into a contract for increasing the general welfare by innovation and work.
Possible strategies for the wealthy population are:
Honor the contract by increasing the wealth through reinvestment to
create innovation and economic growth: HONOR
610 Geometry, Language and Strategy—Vol. 2
0 GT vWealth mg
F G 0 vWork mg . (16.8)
vWealth m vWork m 0
We interpret vwork as the game value for the worker population and
vWealth as the game value for the wealthy population.
The payoff matrix has a null vector, which we call Nash equilibrium:
f Wealth Y
V Nash
f Work X . (16.9)
mg
Two-Person Decision Processes 611
GX v g X
. (16.10)
G T Y vg Y
The condition that V Nash is the null vector then leads to the following
conditions:
vWealth f Work v g ,
vWork f Wealth v g ,
vWealth v
vg Work vWealth f Wealth vWork f Work 0 . (16.11)
f Work f Wealth
We have a new population value rule that the product of the population
and the value is zero-sum.
For numerical work, we assume vWealth : vWork 9 :1 , so with these
assumptions, we have f Work : f Wealth 9 :1 . The number of workers
outnumbers the number of wealthy by a factor 9, the same factor by
which the game value of the wealthy outnumbers the worker game value.
As a result, we will have the payoff matrix for the worker (and an
identical one for the wealthy):
9 vg
10
0 0 0 9
100
mg
9 v
10 g
0 0 1
100 0
mg
1 v
0
10 g
F 1,2 ab 1100 0 0 . (16.12)
mg
1 v
9 10 g
100 0 0 0
mg
910 vg 9
10 vg 1
10 vg 1
10 vg
0
m mg mg mg
g
612 Geometry, Language and Strategy—Vol. 2
Based on our assumptions, the sum of the preferences for the worker
(third and fourth components) is a factor 9 above the sum of preferences
for the wealthy (first two components). This is illustrated in Fig. 16.15.
During the first cycle we note that the inequality flow exceeds the
conserved flow, which implies that the rate of growth of the wealth
population starts to decrease. As time progresses, this changes and the
rate of growth of the two populations become almost equal: the growth
of the inequality flow becomes zero. Neither condition is stable; rather
we alternate between these two extremes. We do see a trend however:
the inequality continues despite the fluctuations and tends to deepen if
you look at the effect on the overall preferences.
The figure also exhibits competitive information based on the natural
resonance structure. The initial normalized strategic flows in this basis
are:
V u1
,V u2 ,V u ,V r 0.1394..., 0.01549...,0.1095...,0.1369... . (16.14)
With the time component, the flows are normalized to unit length,
gabV aV b jkV jVk 1 . The net work-strategy
V u1 1
2 V HONOR
work Vwork
BREAK
,
represents the difference of the normalized worker strategies to HONOR
or BREAK the contract. It is positive and proportional to the Nash
strategy difference for workers, 72 2 200 from Eq. (16.13). The net wealth-
strategy
V u2 1
2 V HONOR
wealth Vwealth
BREAK
,
represents the difference of the normalized wealth strategies to HONOR
or BREAK the contract. It is negative and proportional to the Nash
strategy difference for wealthy, 8 2 200 from Eq. (16.13).
We see the interesting dynamic in Fig. 16.15 that during the phase in
which the workers more strongly honor the contract, the wealthy more
strongly focus on rents. Conversely when the workers adhere less to their
contract, the wealth strategy is to invest and rent about equally.
614 Geometry, Language and Strategy—Vol. 2
Based on our initial parameters (Ex. 7), the wealth inequality generates a
seasonal natural resonance with a large seasonal timeframe. The frames
rotate based on this same constant: see for example Fig. 16.16.
It is not just our choice of parameters. It has been shown that the type
of wealth inequality effects we have been discussing oscillate over
periods of time spanning multiple decades [Piketty, 2014]. In addition,
there may be short term oscillations not directly related to the
competitive effect under discussion. In decision process theory, such
effects show up as forced harmonic effects and are set by the initial
boundary conditions, Sec. 14.9.14. We describe such effects in this
section.
Two-Person Decision Processes 615
x a x, y , z , Re x a ei
. (16.15)
x a x, y , z , Re Ha x, y ei
a
x, y
ei z
This solution, valid up to second order in a power series expansion
around z 0 , allows us to interpret the general solutions as either
travelling or standing waves. The waves are characterized by a transfer
function Ha x , y e i x , y that is analogous to that in electrical
a
a a i a ,
2 q 4i e q a
2 e 2 q 1 e e 4a a 2i e q a q 0 . (16.16)
a 1 2 z q z
. (16.17)
a 2 e 2 q 1 e e 1 4 z q z
2
. (16.18)
Ua x, y , z Ha e z cos a z a
a
circuits. The reduced phase shift and the transfer modulus are the
outcomes of the theory.
This analysis starts with parameters identified at the origin and
determines transfer functions that characterize the solution around the
origin. We can reverse the process and start with a general solution, Eq.
(14.35) and compute the modulus of the solution and its phase:
Ua x, y, z Ha cos a
Va x, y , z Ha sin a
. (16.19)
a a a z
a
Ha Ha e z
In Fig. 16.20, we see the phase for both the seasonal harmonic
component and the forced harmonic component in the f1WWW model.
The seasonal harmonic component is a standing wave whereas by
construction, the forced harmonic component is a travelling wave
moving to the right with a a 0.92.. based on the input parameters and
Eq. (16.17). We compute the velocity in any direction for a frame wave
as a , so we expect the gradient of the transfer phase to be zero for
the standing wave and 0.92 for the travelling wave, which is what we
see. Note also that the reduced transfer phase shift is not zero for the
forced harmonic.
and so the gradient of the phase would also be zero. We have constructed
such a superposition at z 0 .
The resultant wave phase gradient, Fig. 16.23 shows that both the
seasonal and forced harmonic components are standing waves as
expected. This is further confirmed in Fig. 16.24. We have imposed the
This does not necessarily imply that the gradient of the phase is zero.
Indeed, the phase gradient, Figure 16.25, indicates that although the
travelling wave components along z are small, they are not necessarily
small along the other two directions.
When we turn to the large time scale behavior, we see less structure,
Fig. 16.26 than in Fig. 16.22. We expect standing waves to exhibit the
structure of travelling waves, including reflections, after all transient
behaviors have died out. Once the transients have died out, we may no
longer see what is travelling and what is being reflected. We think it
makes sense that there is correspondingly less structure exhibited. It is
Two-Person Decision Processes 625
16.12 Outcomes
From Sec. 16.5, the student will learn how to set the gauge for the AD
model. The student will have learned the consequence of picking the
covariant gauge is the necessity of solving a set of differential
equations on the initial surface. This is accomplished using the
harmonic power series approximation.
From Sec. 16.6, the student will have learned that game theory ideas
can be extended in the context of decision process theory to a
problem of two players who get different payoff values. This leads to
the work-wealth-wisdom (WWW) model.
From Sec. 16.7, the student will have learned about the streamlines
that result from the seasonal harmonics in the WWW model.
From Sec. 16.8, the student will have learned how to add forced
harmonic effects to the WWW model and how such a f1WWW model
with a single forced harmonic behaves.
From Sec. 16.9, the student will have learned how to see the travelling
waves in the f1WWW model. They can be seen by looking at the
transfer function for each harmonic component. The magnitude and
phase of each component characterizes the behaviors. The phase
contains the information about the travelling wave velocity.
From Sec. 16.10, the student will have learned how to create standing
waves in a f1WWW model. In this case the wave does not move,
being a superposition of waves moving in opposite direction with the
same amplitude.
From Sec. 16.11, the student will learn the value of using recurrence
plots to see different harmonic contributions. The student should be
able to see how such plots could also indicate chaotic behaviors, such
as suggested in Sec. 9.9.
16.13 Exercises
(1) To set the parameters of the attack-defense (f0AD) model, show that
we need to set the payoffs as in Eqs. (16.1) and (16.2). We take the
initial flow to be the Nash equilibrium Eq. (16.4), which assumes a
decision mass of md 40 . For the inactive flow or charge for each
player we take V 1 V 2 1 . Do we have enough information to
Two-Person Decision Processes 629
Assume that we look for solutions that are periodic in the transverse
directions Lx x Lx and Ly y Ly , but not in the z
direction. For this direction, there will be limits to choice,
zlow z zhigh that will come out of the initial variable values. To
specify the steady state scalars, show that we must at least provide
the following initial values, where the values have been chosen to
match those of the figures in this chapter and we use the notation
from Chap. 14 (here x, y, z represents any tensor field):
5, 1,
Lx 5, Ly 5,
Lx , y , z Lx , y , z x, Ly , z x, Ly , z ,
ω B 3 0 0 0,
630 Geometry, Language and Strategy—Vol. 2
f 3 x 0, f 3 y 0, f 3 z 5,
z11 310 , z 22 4 10 , z 33 510 , z12 0 . (16.20)
Note that the player inequality appears to oscillate with the seasonal
harmonic. As a function of time, describe the forces that these
contours represent.
(5) There is a similar behavior with Ex. 4 in Fig. 16.34 for the
acceleration potential. Here the potential oscillates with the seasonal
harmonic frequency. Why is there a maximum? What does this say
about the forces? As we switch from normal coordinate frame to
orthonormal frame, it is helpful to see the behavior of the streamline
contours u t compared to z , Fig. 16.35. What happens
when the streamlines become bunched?
634 Geometry, Language and Strategy—Vol. 2
there will be limits to choice, zlow z zhigh that will come out of
the initial variable values. To specify the steady state scalars, show
that we must at least provide the following initial values, where the
values have been chosen to match those of the figures in this chapter
and we use the notation from Chap. 14 (here x, y, z represents
any tensor field):
5 1
Lx 5 Ly 5
Lx , y , z Lx , y , z x, Ly , z x, Ly , z
. (16.22)
ω B 3 0 0 0
f 3 x 0 f 3 y 0 f 3 z 10
z11 310 z 22 4 10 z 33 510 z12 0
(8) In the numerical work (e.g. Ex. 1), the assumption has been that we
can set the active acceleration to zero at a point. Take this idea one
step further. In a consistent way, set the active acceleration to zero
on the initial surface. Show that this implies that the acceleration
and its gradient along the focused direction are determined on the
surface:
q pulse 1 2 ln 1 e e
. (16.23)
z q pulse e I z e z e
pulse
(10) For the numerical solutions in Ex. 9, investigate the stability of the
topology of the solutions for different mesh sizes. Compare with the
results shown in Figs. 16.39 and 16.40. Based on these results, we
(12) We have seen the seasonal variations of the pressure, Fig. 16.33 and
the acceleration potential, Fig. 16.34. When we include an
additional forced harmonic, the pressure develops corresponding
structure, Fig. 16.41 as does the acceleration potential, Fig. 16.42.
The observed effects are quite small here. What would you have to
do to make the structures significant?
(13) The free fall harmonic solutions in Sec. 15.15 Ex. 5 and in this
section, display two interesting attributes: a linear solution and a
helical solution that circulates around it. This is an inherent property
of the free fall solution.
642 Geometry, Language and Strategy—Vol. 2
1 12 1 e
1 1 2 2 1 2
U a 1 0,0,0 U a 2 0,0,0 1 2 1
3
1
10
1
5 . (16.25)
V a 1 0,0,0 V a 2 0,0,0 0 0 0 0
a u1 u2 u t
Two-Person Decision Processes 645
(16) The streamline “tube” in the f2WWW model, Fig. 16.47, has more
structure than the previous f1WWW model, Fig. 16.43. Discuss the
differences. Take into account the vastly different time differences.
For two forced harmonics, the proper time is in the interval
0 10 .
(18) In what way does the frame transformation, as seen in the normal
frame for the f2WWW model Fig. 16.50, show the effect of a
stronger coupling than the corresponding curve for the f1WWW
model, Fig. 16.44?
648 Geometry, Language and Strategy—Vol. 2
Figure 16.51. Travelling wave recurrence plot for the inequality flow in
the f2WWW model
(19) We see new structures at shorter time frames in the recurrence plots
for the f2WWW model. Explain the structures in Fig. 16.51 for the
inequality flow in this model.
Chapter 17
649
650 Geometry, Language and Strategy—Vol. 2
. (17.1)
Va x, y, z Ha e z sin a z a
a
The steady state fields from Sec. 14.9, and the transfer functions,
depend on the coordinates that are transverse to the energy flow as seen
in the co-moving orthonormal coordinate frame. Adding the transfer
function to the steady state fields expands their role in providing the
geometric picture of the fields. The steady state view provides one guide.
This view however is not complete. As we move along a streamline
solution, the motion may slow down or stop as we approach a region that
does not make sense. It would not make sense for example to have an
event from the future impacting the present. We need a geometric picture
to help us anticipate that possibility. That is a second category that may
help guide us.
We focus in this chapter on each of three geometries indicated above
as well as on the geometry of the regions that make sense for decision
processes. For the models under discussion, these geometries are in 3D:
three dimensions. For illustration purposes in this chapter, we use the
attack-defense model defined in Sec. 16.2 with two forced harmonics
(f2AD).
Decision process theory provides an integrated view of all of the
effects. We pointed out the importance of treating causal behaviors
globally in Sec. 13.1 as a generic goal of modeling the social and
economic behaviors. Our approach here is the same as the one described
in that section. We look for stable behaviors and see how the system
responds to stimuli. The theory closes all the loops so that all the
interactions are captured. We can easily change our assumptions about
initial values, which may change the relative importance of each of the
effects.
We start with a geometric picture of the valid regions in Sec. 17.1: we
specify the regions of validity based on the causal behaviors. In Sec. 17.2
we examine the transfer functions for the f2AD model, which describes
the active geometry. Next, we build on game theory by looking at the
payoffs as fields that describe the active-internal geometry. The internal
geometry relies on the concept of payoffs that in 3D have a vector field
structure articulated in Sec. 17.3.
One ingredient of this structure is an external current, the player
passion field, Sec. 17.4. After considering the geometry of this structure,
we address the coupling of these fields to actions through the player
652 Geometry, Language and Strategy—Vol. 2
engagement, Sec. 17.5. This opens up the discussion about the gradient
geometry.
The player engagement is generated by the player interest, Sec. 17.6,
which in turn generates the player strategy bias, Sec. 17.7, which
generalizes the concept of game value in game theory. Cooperation
between players is set by the player bonding and player shear, Sec. 17.8,
which provide a description of the internal geometry.
We return to the composite payoff, Sec. 17.9, which we take initially
to be zero on the initial surface. Finally we deal with the seasonal
payoff, Sec. 17.10, which can be viewed as an attribute of the active
geometry or of the inactive geometry. We show that the stationary
geometry helps us understand the significant dynamic features of the
theory and simplifies the computations. Note that there is significant
numerical work required to go from the steady state view to the dynamic
time dependent view, just as there is in electrical engineering.
t
E t 0 . (17.2)
This assumption depends on more than just the steady state geometry;
it must be satisfied along every streamline: as the proper time
increases along the streamline, normal time also increases.
For n a active strategies, the volume assumption is that the invariant
volume element g du1 du na dt that is used to define the
action, Eq. (3.29) is non-singular, so the determinant 0 g 1 .
The consequence is that the determinant of the frame transformation
must be non-singular:
654 Geometry, Language and Strategy—Vol. 2
m m 2
E u n Eu
1 e e
na
g ab
0. (17.3)
E t n E t
g ab 1 e e E t E u n
2 2
na m
0
1 e e 0
E u n 0
m
. (17.4)
um
E E t n
E u n E u n
m m
E t
du a du b
g ab 0,
dt dt
du m du n du m
m, n t g mn 2 gtm g tt 0 . (17.5)
dt dt dt
In general, Eq. (17.5) describes a point (if all eigenvalues are the same
and the metric is not Lorentz), an ellipsoid shape or a hyperboloid shape
at each point along the streamline.
We set the value g tt 0 at the initial point along the streamline as
well as the remaining eigenvalues, which are set to be negative. The
shape of the ellipsoid or hyperboloid is determined by the eigenvalues of
du m du m du m
g
mm
m dt dt
2 g tm
dt
gtt 0,
2 2
du m gtm g tm
g mm g tt g mm . (17.6)
m dt g mm m g mm
656 Geometry, Language and Strategy—Vol. 2
The right-hand side is positive initially, since the spatial eigenvalues are
all negative. When an eigenvalue goes negative there is either no solution
or the solutions are an ellipsoid or a hyperboloid.
As examples of the ellipsoid along the streamline, we show two cases
for the f2AD model: one with g tt 0, g tt 0 and one with
g tt 0, g tt 0 . In both cases E t 0, Et 0 .
We see that when g tt 0, g tt 0 , the shape (Fig. 17.1) is a
communication ellipsoid, which implies that at each point, there is an
open set around the streamline and there is a null flow in any direction in
that open set.
this means that for an open set around that point on the streamline, there
will be parts of space that will never be reached. We impose the
communication assumption that you can communicate to any part of
space: g tt 0 . This ensures that the point where all eigen-velocities
vanish occurs strictly inside the ellipsoid.
origin is inside the ellipsoid. If this is not the case, the origin is no longer
inside the ellipsoid as in Fig. 17.4. This argument extends the
communication assumption to require both conditions: g tt 0, g tt 0 .
These constraints in turn impose constraints on the behavior of the
flows and streamlines, Exs. 4-5. We see that the validity regions are
specified by the βn and its dual * βn, Ex. 8, along with E t and the
reduced frame transformation E u n .
m
The dynamics depend not only on the active geometry but on the active-
internal geometry, which is described by the player payoff fields. We
describe them in the co-moving orthonormal frame. For the two-person
games under consideration, these fields are defined in terms of the three-
dimensional field equations in Sec. 14.9. Complete solutions are obtained
from these equations by providing appropriate boundary conditions,
which we do by providing the initial values for the fields and their
gradients along z on the two-dimensional surface z 0 .
662 Geometry, Language and Strategy—Vol. 2
or “blue” team and player 2 is “red” or attack team. The model helps
identify the similarities with game theory and electromagnetic theory as
well as the differences.
For example, the payoff field for “blue” is primarily along the z
direction, Fig. 17.7; similarly for “red”, Fig. 17.8. This direction is
orthogonal to the attack-defense strategies, so it is not a surprise based on
the game theory input. The behavior is present in the initial conditions at
z 0 : the largest initial vector components are shown in Figs. 17.9 and
17.10.
The player payoff can also be derived from a vector potential,
Bα B Aβ , Sec. 14.16, Ex. 2. Though not computed, we infer
from the electromagnetic field analogy that the vector potential fields
that generate these payoff fields will consist of circular orbits in the x, y
plane.
664 Geometry, Language and Strategy—Vol. 2
The gradient equations prevent payoff field lines from being created
or destroyed. We expect field lines that are approximately straight, or
closed loops. For “blue” and “red” payoffs we see such lines.
We might see quite different behaviors if we change the boundary
conditions on z 0 , Figs. 17.9 and 17.10 to oscillate for example or if
we impose conditions on the outer boundaries as opposed to z 0 . Since
the equations are elliptic in nature, there exist solutions; they are likely to
be similar to those found in electromagnetic theory, where there is a rich
set of solutions based on different possible boundary conditions.
The “curl” field equation B α , as the name suggests, determines
the rotation of the field lines. For our solutions, the rotations are small
compared to the strength of the field along the z axis.
We observe a small effect however, which we isolate by looking only
at the transverse components. For “blue” and “red” players we see the
effect by looking at representative slices. Now we see evidence that
the field lines can form closed loops. For “blue” player Fig. 17.11, we
see that there is a circle centered near x 5, y 0 . For “red” player
Fig. 17.12, there is a circle centered near x 0, y 2 .
666 Geometry, Language and Strategy—Vol. 2
It is worth noting that the relative size of the effects has everything to
do with the initial boundary conditions. What is important is that the
“curl” and “divergence” mechanisms are always in play and depending
on the initial conditions, may totally change the sizes of these effects. For
example, the effects that we see are quite different for the code of
conduct (player 3), Fig. 17.13. Here we see circulations around
x 0, y 0 , even on the initial boundary z 0 .
It also illustrates an attribute of equations in differential geometry that
some equations are constraint equations. Indeed the curl component
x B y y B x of Eq. (14.24) is just such a constraint, which must be
satisfied on the initial boundary. Although by construction, we start with
an initial boundary condition close to zero, Fig. 17.14 for B3 z , we can’t
consistently set the other components to zero.
Steady State Geometries in 3D 667
The total payoff field has lines that now flow in both directions,
Fig. 17.15. The field equations are satisfied: that lines disappear or
appear is due to the divergence terms in Eq. (14.24).
Thus in computing model payoffs, we have only scratched the surface
of what is possible based on the boundary conditions and the field
equations. If we move away from
periodic boundary conditions and
from the focus on Nash equilibrium
like solutions, then we will see a
much richer structure, such as what
is seen for electromagnetic static
fields.
The analogy is even closer since
it is possible to write the decision
Figure 17.14. Code of conduct initial
process theory equations for the
boundary condition
payoff in terms of a vector potential,
Sec. 6.10, Exs. 43-45 and Sec. 14.16,
668 Geometry, Language and Strategy—Vol. 2
Ex. 2: we can use the same intuition for the geometry of the
electromagnetic field lines in this theory. This approach generalizes to
higher dimensions.
The geometry for the player payoff field lines is not complete without a
discussion of external effects that generate currents and hence provide
sources for the payoffs.
The mixed stress components of the energy momentum tensor p in
the co-moving orthonormal frame define a source, the player passions
Pα , for the player payoffs. We identify such stresses as arising from
dynamic contributions that we have not explicitly included with the
active players and their strategies. From the field equation for the player
payoffs, Eq. (14.24), we see that the player passions do indeed provide a
source current for the player payoffs.
Steady State Geometries in 3D 669
Figure 17.16. “Blue” player passion density plot for f2AD model
The form of the player passion potentials on the initial boundary has
been chosen arbitrarily. The model is periodic on the transverse
boundary, whose behaviors are given in Fig. 17.16 for “blue” and Fig.
670 Geometry, Language and Strategy—Vol. 2
17.17 for “red”. A similar density plot can be obtained for the code of
conduct player corresponding to Eq. (17.7). See Ex. 11 and Fig. 17.39.
Figure 17.17. “Red” player passion density plot for f2AD model
The density plot for “blue” generates a vector field that will be
vertical, along the positive z direction for z 0 and opposite direction
for z 0 . There will be no substantial transverse flows to this direction.
This qualitative behavior is a consequence of the boundary conditions
chosen. We can imagine different boundary conditions. Indeed, we see
different behaviors for “red”, Fig. 17.19. We find the gradients for the
vector field for this case to be similar.
As a third example, consider the player engagement for the code of
conduct player, Fig. 17.20. This shape is quite different, though here the
field lines are vertical, along the z direction. We get vertical vector
Steady State Geometries in 3D 673
For player engagement, the most interesting case might be one of two
extremes: region in which the player engagement is zero, in which case
the player payoffs are entitled and not constrained. The other case would
be when the player payoff is extremely large. In this case the players are
highly competitive or highly cooperative: they are definitely engaged.
Closely related to the player engagement scalar fields are the player
interest vector fields, Eq. (14.29):
1
Iα e qe Θe σα βe . (17.8)
ni
If we ignore the other fields temporarily, we see that the gradient of the
player engagement scalar field determines the player interest vector
Steady State Geometries in 3D 675
field. There are other fields and we indeed have equations of motion for
the player interest field in Sec. 14.9.5. These fields satisfy their own
boundary conditions.
From these fields we can in fact determine the player engagement
fields: the fields are thus strongly related. Conversely, we can use the
equations to derive the scalar engagement fields from which we compute
the vector player interest fields.
q 0
I x3 I y 3 0 . (17.9)
z11 310 z 22 4 10 z 33 510 z12 0
Based on our analysis of the player engagement, we expect that the
largest components of the player interest will be along the vertical or
longitudinal z direction as opposed to the transverse x y
directions.
The transverse vector field for “blue”, Fig. 17.21, reflects the
boundary choices as can be seen in the z 0 slice. The behavior
propagates to the other slices without much change. We note that the
longitudinal contributions are much larger, as seen in Ex. 13.
which depends on the overlap of the composite payoff and the player
payoff, ω B α . Since we start with game equivalency, this starts at zero.
However, this may not be the most interesting boundary on which to set
the initial conditions. Based on analogy with electric fields, a region of
high density overlap might generate quite different player interest vector
fields and player engagements.
The player interest is closely associated with the player strategy bias,
f j , Eq. (4.74), the analog of the electric field. See also the discussion
in Sec. 15.4. We can rewrite the player strategy bias in terms of a
reduced player strategy bias, fˆ
, which extracts some of the frame
dependent behavior that might become singular:
678 Geometry, Language and Strategy—Vol. 2
E j fˆ
f j
1 e e . (17.10)
fˆ
2q e 2 I e e 2 I 2e
For “blue”, the reduced player strategy bias field, Fig. 17.23, is
primarily vertical. This is a force primarily along the z direction, which
pushes towards “blue” based on the definitions, Eq. (15.24). The origin
of the direction of force is that for the boundary conditions under
consideration, the engagement terms are small and the player interest
term dominates. We can certainly envision situations where this might
not be the case.
Steady State Geometries in 3D 679
We see a similar behavior for “red”, Fig. 17.24, though in this case
the field goes in the other direction. We suggest that the strategy bias
fields are indications of the same concept as value in game theory. For
this reason we suggest the name “bias” as a force that pushes a player in
one direction or another. It suggests the same concept as game value in
game theory.
The player strategy bias field depends not only on the player interest, but
on the acceleration, player bonding and player shear. Bonding and shear
define geometric properties of how a unit square changes along a path.
The player bonding and shear are consequences of the stresses in the
system, which are represented by the decision process pressure field,
Fig. 17.25.
We believe it is noteworthy that the decision process pressure density
has distinctive high pressure areas and low pressure areas. We find that
pressure plays a similar role to that played in weather: fluid flows from
high pressure areas to low pressure areas. We see that there will be
gradient forces here that do precisely that: the forces will move the
decision process from high pressure areas to low pressure areas. The
forces in the theory generate acceleration. The geometry that results will
be the geometry of the decision process flow.
The decision process pressure is proportional, by assumption, to the
energy density of the decision process. The high pressure areas indicate
areas where the process has accumulated, whereas low pressure areas
will be less populated. We indicate the steady state behavior after all
transient effects have died out; even though this is steady state, we
nevertheless allow for standing and travelling waves. It is steady state in
the same sense that AC circuits are steady state, yet describe time
dependent behavior.
680 Geometry, Language and Strategy—Vol. 2
we find it converges, giving us one of many possible solutions that fit the
known conditions and satisfies all of the field equations.
We obtained closure with these known behavior values for the shear
components:
y x x y
11 x, y ,0 0.0469857cos sin 0.064015cos sin
5 5 5 5
y x x y
12 x, y ,0 0.20336cos sin 0.343935cos sin
5 5 5 5
y x x y
13 x, y ,0 0.0563027cos sin 0.151236cos sin
5 5 5 5
. (17.11)
y x x y
22 x, y ,0 0.0782177cos sin 0.132287cos sin
5 5 5 5
y x x y
23 x, y ,0 0.000179326cos sin 0.0562865cos sin
5 5 5 5
y x x y
33 x, y ,0 0.125203cos sin 0.196302cos sin
5 5 5 5
0.3 0 0.362542
x 2 y
z x, y ,0 0 0.4
0.362345 cos2 cos
0.362542 0.362345 10 10
0.5
x y
x, y ,0 0.37561sin 0.588905sin . (17.12)
5 5
6 x 2 y
z x, y,0 cos 2 cos
5 10 10
have the opportunity to compete. This again shows the importance of the
boundary conditions on the qualitative shape of the geometry of the
scalar and vector fields.
As a side note, the equations that generate the seasonal payoff have a
great many terms so it is helpful to identify ways that insure the
calculations are correct; using the fact that the gauge condition should
remain satisfied at all values of z is one helpful check. Our
computations meet this minimal test.
17.11 Outcomes
We have explored the geometry of the theory using the f2AD model as a
guide, along with multiple appeals to properties of partial differential
equations. The geometry of the theory consists not just of the space in
which actions occur, it consists of the fields. There are both scalar and
tensor fields, each with their own special set of equations. However, the
equations separate out into divergence equations for the scalar fields and
for the vector fields, divergence equations and “curl” equations.
In this chapter, we have observed the effects of several mechanisms
such as player interest, engagement, entitlement, passion, and strategy
690 Geometry, Language and Strategy—Vol. 2
bias. We observe that they play as big a role in decision process theory
as do the concepts of game values, payoffs and Nash equilibrium that we
have taken from game theory. These insights contribute to our
understanding of decision processes.
An important key to understand how to apply decision process theory
to realistic situations is not only to understand the causal equations, but
also to understand the boundary conditions that are part of the problem
statement of any realistic problem.
With some caveats, the approach taken here can be modified and
adapted to incorporate any two-person game with two-strategies per
player. Even though relatively simple, such an extension significantly
advances our understanding. We now deal with a very large number of
games that have already been dealt with in the game theory literature.
With somewhat more effort, we can extend the discussion to games
of three or more players as well as to games in which each player has
more than two strategies. Though decision process theory deals with any
number of players who have any number of strategies, the relevant
question is one of practicality. We are greatly helped by the state of
current technology and its ability to solve partial differential equations.
We have used Mathematica’s numerical method of lines to provide
practical solutions and suggest it can handle even more players and more
strategies. There are currently limitations however, though we
understand work is underway to remove many of them. For example, it
will be of great help to have a general technique to numerically solve the
non-linear elliptic partial differential equations for more general
boundary conditions than purely periodic ones. One promising approach
is the use of finite element method, which has been implemented in
Mathematica for linear partial differential equations.
If we extend our numerical efforts to differing numbers of players or
strategies, what might we expect? As in game theory, the possibilities
become more interesting as we increase the number of players. For
example, it is possible with current computer technology to extend our
analysis to three-person processes as well as four player processes with
active strategies u1 r1 r2 u2 r2 r3 u3 r3 r4 . We expect a
more extensive discussion of which strategies are active, which are
inactive and which are parts of a code of conduct. We also anticipate an
Steady State Geometries in 3D 691
In Sec. 17.3, the student will build on game theory by looking at the
payoffs as fields that describe the active-internal geometry. The
student should be able to understand what boundary conditions are
needed to determine the payoff fields and how they might be set in
any particular model.
In Sec. 17.5, the student will learn about a component of decision
process theory that acts like an external current, the player passion
field. The student should understand how the vector field can be
obtained from a scalar field and under what assumptions this makes
sense.
In Sec. 17.5, after considering the geometry of player payoffs, the
student should understand how payoff fields couple to other fields and
create forces. This coupling, the player engagement, has a geometry
that is determined by a scalar field, whose density distribution
determines the surfaces of constant engagement and thus determines
the gradient vector field that results.
The gradient of the player engagement field is a key contributor to the
player interest. In Sec. 17.6, the student will learn about the shape of
this vector field, both from the model considerations as well as from
more general considerations based on possible and different boundary
conditions.
In Sec. 17.7, the student will have learned about the player strategy
bias vector field, which generalizes the concept of game value in
game theory. The geometry of this field is determined in large part,
but not entirely, by the player interest in the f2AD model. The danger
of looking at a specific model is that it depends on relative strengths
of competing mechanisms, which with different boundary conditions
will yield totally different results.
In Sec. 17.8, the student will have learned the internal geometry
behind the decision pressure process, player bonding and player
shear. These are the stresses and strains of the decision process
system. The scalars are described by their surfaces of constant
potential. This leads to an understanding of the associated vector
fields.
Steady State Geometries in 3D 693
In Sec. 17.9, the student will have learned that the composite payoff,
though small in the model under consideration, is nevertheless not
zero.
In Sec. 17.10, the student will have learned that the seasonal payoff,
which is an attribute of the geometry considering time as active, is not
necessarily small. Not only is it the source of the harmonic behaviors,
it can also be thought of as being made up of the weighted
contributions of the player payoffs and their engagements.
17.12 Exercises
provides the determinant of the metric tensor proportional to
m 2
E u n , as shown in Eq. (17.4). The reduced frame transformation
E u n E u m ml ,
l n
E ul E u
m l
nm
E
E n E t
m m t
u
E
u . (17.13)
E n E t E t l E n E E l E ui E u
l l i
t t
u u
E t E t E t
(2) From Sec. 6.10, Ex. 23, the time component of the metric in the
central frame is
694 Geometry, Language and Strategy—Vol. 2
e2 q
g . (17.14)
1 e e
This is the velocity of the coordinate time surface, which defines the
E t
* . (17.17)
Et 1 e e
equal at one point, are not in general equal. Show that the n-form
X Y , which is the wedge product of the two forms, does not
vanish. Show that the “dot” product of the components is
X Y 1 . Show that the “dot” product of the two velocities is:
E t E t 1
* t 1. (17.18)
Et E 1 e e Et E 1 e e
t
um Eu
flow
, (17.19)
E t Et 1 e e
um Eu
Et 1 e e flow t
E
um
E
Et E u n t Et 1 e e u flow .
n m m
(17.20)
E
* E u n
m n m
u flow
2
(7) Show that the flow-β is bounded: βflow . Use Eq. (17.4) and
2
the fact that the dual *βn is bounded by unity. Can you
determine ?
(8) Based on Sec. 17.1 and the Exs. 1-7, show that the frame
transformation is characterized by the βn, the dual *βn, the time flow
E t and the reduced frame transformation matrix E u n , Eq. (17.4).
m
Steady State Geometries in 3D 697
(9) As shown in these figures, show that βn (Fig. 17.36) and its dual *βn
(Fig. 17.37), may be equal at one point, but are not in general equal
everywhere. Why are the shapes so different?
Figure 17.39. Code of conduct passion density plot for the f2AD
model
ω I α B α q ω
. (17.21)
ω Θ 2q ω
700 Geometry, Language and Strategy—Vol. 2
ln Θ 2q
. (17.22)
2 Θ 3q Iα Bα
Based on the numerical work done for the f2AD model, in what
region will the source I α B α in Eq. (17.22) be large? Cf.
Fig. 17.46.
702 Geometry, Language and Strategy—Vol. 2
(16) The form of the equation for the composite payoff in Eq. (17.22) is
valid only for the specialized set of dimensions dealt with in the last
few chapters. The more general form of the equations has been
provided in Chap. 5. One of the equations is the Codacci Eq. (5.36):
2 q
2 q .
2 q 0. (17.24)
d * ω *j
. (17.25)
dω Iα Fα q ω 0
ω *dG
d Θ 2q . (17.26)
d * dG Θ 3q *dG Iα Fα
705
706 Geometry, Language and Strategy—Vol. 2
We assert that decision processes are not stochastic, but connected both
in space and in time: strategic choices made by one group impact choices
another group makes. Despite the obvious uncertainties about what will
happen in the future, there are aspects of the decision process that are
deterministic.
We have pointed out that the resultant equations, which are
hyperbolic partial differential equations, exhibit a mathematical stability
that small changes in the initial conditions result in small changes in the
solutions [Wald, 1984]. We expect that such solutions are characterized
by a structural stability or locality property that the structural character
of a solution does not change dramatically as long as we stay in the near
vicinity of that solution.
The expectations of structural stability are local, not global attributes
of nature or of decision process theory. In nature we have phenomena on
a lake or an ocean that on a calm day, exhibit no wave motion. The
components of the lake are connected. With a small breeze, the lake is
changed only slightly with the addition of small ripples. This is an
example of the expected structural stability.
As the breeze picks up however, the ripples turn into waves that at
some point exhibit breakers, which are not simple extrapolations of the
small ripples. This morphogenesis follows from rather general
arguments, is observed in nature and is expected whenever we have a
theory based on topology and geometry of the type we have been
considering [Thom, 1975].
The observed phenomena of ripples can be used to derive the fluid
equations of fluid dynamics. The equations can be used to extrapolate
behaviors beyond their local origins. The structural anomalies we
observe in hurricanes and tornadoes in meteorology can be simulated
based on these considerations [Friedman, 1989].
We have shown in Secs. 16.9-16.10 that such structural changes also
occur in decision process theory. For example, with only seasonal
harmonics, we get slow rolling standing waves, Fig. 16.22 as seen for the
frame transformation in the f1WWW model. The ripples that we see are
part of an added harmonic that is part of the external force or initial
708 Geometry, Language and Strategy—Vol. 2
In Sec. 12.1 we argued that decision processes are causal not stochastic,
despite the fact that decisions involve uncertainty. We argued against the
Bayesian approach [Bayes, 1764] and in favor of an approach looking at
possible decision choices as a frequency distribution that evolves in time
in a causal and deterministic manner. In Secs. 9.9 and 16.11 we
identified recurrence plots as a direct way to identify whether processes
are causal and deterministic. We found examples of what we expect to
see in time sequence plots: for the prisoner’s dilemma, Figs. 9.23 and
9.24; and for the f1WWW model, Fig. 16.27 and the f2WWW model, Fig.
16.51.
We find examples of this
behavior in a variety of
places. First, in the literature,
such recurrence plots have
been used by a number of
authors to investigate chaotic
behaviors. For example
Sabelli [2005] has used such
plots to study heart beats,
whose time sequences might
Figure 18.1. Time sequence for a typical stock be thought stochastic. He
Jan-Dec 2015 showed that the behaviors are
causal and argued that the behaviors indicated chaotic (or what he termed
biotic) behaviors. Second, in stock market data, we find that stock prices
710 Geometry, Language and Strategy—Vol. 2
fluctuate daily. You might argue that the prices are random and unrelated
to each other from day to day. To test this idea consider a stock, where
we have used Wolfram|Alpha® to obtain stock market data [Stock
Market, 2015], Fig. 18.1.
The behavior of the stock price appears to be random elements.
However, that is not necessarily the same as actually being random. We
see causal structure in the recurrence plot for this data, Fig. 18.2, which
has much in common with our model plots. There is a strong correlation
in time: if the data were random, then we would not see any structure in
the recurrence plot. The plots are not however periodic with a simple
period. It is for this reason we see the choppy behavior in the time
sequence plot. We have not attempted to model the stock prices using
decision process theory, but we believe these preliminary inquiries are
encouraging evidence that time is causally connected.
market, whereas the stock market reflects many markets and many
products. We imagine that the stock market reflects processes that go
beyond our simple attack-defense example.
In the real world, a market is not isolated but carries on in a larger
environment. One way to model that environment is to ignore its effects
as reflecting longer time scale phenomena. That is, we replace the real
strategic variables by ones in which the environment effects are absent.
For the stock market we would look at specific stocks and over limited
periods of time.
Only then might we see the free fall behaviors we are interested in.
We would proceed somewhat like scientists investigating weather
patterns by ignoring the spin of the earth (Coriolis forces) as a first
approximation. At some point however we would return to the problem
and add back such effects.
We examine causality with a second example, this time from the
business domain. A common problem in software development is
delivering a complex project on time. We fabricate a scenario below,
though from our experience, it is a scenario that we believe does in fact
happen quite often. We give a self-contained description here of a more
detailed description framed in systems dynamics language, Fig. 13.4.
The delivery of a complex project, one that involves perhaps 500
developers, reflects a large number of decision processes. It is attractive
for management to think of such processes as occurring stochastically
and to believe that the delivery of a complex project will follow the same
pattern of previous projects of the same scale and complexity. We
imagine that company X has gone through 4 or 5 delivery cycles, each
taking 2 years and each delivered (more or less) on time. The projection
of the next delivery is thus expected by management to take the same
time, 2 years. But an unexpected variation occurs; after development has
started, the customer asks for substantial new functionality, increasing
the size of the project.
Let’s say both the customer and company X agree that the increased
size is 20% and that the company agrees to hire 100 more people and the
customer is willing to pay for the increased cost. Management predicts
that the original schedule will thus be maintained. The underlying
thinking is that the delivery process is stochastic in nature.
712 Geometry, Language and Strategy—Vol. 2
The people are hired but to the dismay of the customer and
management, the project ends up costing significantly more than the 100
people and does not come in on time, even though the additional staff
were hired with sufficient time, based on past data, to get the job done.
Both sides looked at the problem from the same worldview, that of a
complex stochastic process. Neither side anticipated failure.
How would we look at this process and would the expected results be
predicted to be different? A causal approach (such as decision process
theory or systems dynamics) would look for key dynamic mechanisms
that govern the delivery of the software product. Using such an approach,
we would agree that hiring the additional people is a necessary
ingredient. However, we would find on analysis and by interviewing
members of the technical staff that not all employees will be equally
productive on the new project without sufficient training.
The previous releases were staffed by essentially the same people
with only a small turnover. Over time, they built up a high skill level
based on the known requirements. The skill level of that staff was
constant: new hires that were needed as replacements were always added
to the project only after adequate training.
To satisfy the new customer request however, staff were added after
the overall project was well underway. The additional 100 people were
employees new to the project and were not yet trained and so had a much
lower skill level. Though many of the employees and middle level
managers complained that they would not be able to deliver based on the
addition of untrained staff, their voices were overruled since the
management model had no way to use the skill level to modify their
schedule prediction. They interpreted the employee complaints as normal
grousing about being asked to go for a stretch goal.
The consequence of the lower skill level was in fact not felt
immediately. The lower skill level introduced a higher level of defects
into the product, but these were not detected early on. When the product
was moved along to the testing phase, the higher number of defects per
unit of code was detected. The defect insertion rate could have been
predicted based on existing data on new hires from past releases.
Because of the worldview of management, however, this was not done.
Structural Distinctions for Decision Processes 713
The company of course was now in the testing phase and had to hire
even more staff to find and fix the defects. They were also not able to roll
off existing staff as planned to other releases. They hired an additional
unplanned 50 people, though in fact that was enough only to fix the first
round of defects. They spent 6 months additional because of this testing.
The testing time could not be shortened since it depended on going
through a fixed number of tests that each took a fixed amount to execute.
Of course the additional people hired were again unfamiliar with the
product and many of their fixes also broke the release. The failure rate of
their fixes was about the same as the other new hires.
The consequence of this problem was the fix-break time cycle. Under
normal circumstances a release was expected to pass all but 5% of its
tests by the time it entered the final test cycle. The test cycle of 6 months
was part of the expected 2 year delivery. That final cycle would reduce
the number of escaped defects to less than 5%, an acceptable escaped
defect rate. The reduction of defects was in fact a direct measure of the
skill level and training of the employees. The new circumstance dropped
that defect reduction percentage dramatically. They saw only a 50%
reduction in defect rate.
For a 50% reduction, as they went into the final cycle, 50% of the
tests would not pass. If they did a second cycle, they would have 25%
that would not pass; a third cycle and 12% and a fourth cycle would be
6%. To get to where they should have been in past releases they would
need a fifth cycle, which is four extra cycles. Each extra cycle cost them
6 months and 50 people. This extended the project by two years and
added an unplanned 50 extra people on the staff, as well as keeping on
the project staff that had been expected to shift to other releases. The
total project cost should have been 500 people for 2 years: 1000 staff
years. The budget office showed that the overall project cost was 2000
staff years and a 2-year delay in schedule.
Now the example above is fictitious but illustrates how even a very
simple exercise of adding a single causal mechanism (staff expertise) has
a profound impact on the outcome. Feedback loops can be very
expensive in time and money. They are also relatively invisible to the
participants of the decision process. The decision about what features to
deliver to the customer followed standard form, albeit late in time. The
714 Geometry, Language and Strategy—Vol. 2
decision about when the project was complete did not change. From a
stochastic view the late project and the normal projects are similar.
Indeed, it is easy to make the assumption that all behaviors are
stochastic. The alternative approach of modeling the behavior as a causal
processes seems harder as it requires identifying the causal mechanisms.
Nevertheless, understanding the causal mechanisms is beneficial to the
company.
Despite being complicated, it has been observed that such causal
mechanisms are well understood by organizations (Cf. Sec. 13.2). In this
example, employees and managers knew that a lower skill level would
lead to a higher defect rate, even in the highly subjective world of
software development. In a stochastic model, there is no place to put that
information. The real difficulty is that the global consequence of such
mechanisms is not easily deduced without the causal framework. We
believe that decision process theory provides just such a framework, not
only for computations but for providing an appropriate worldview of the
problem.
We believe the approach is also superior to a purely systems dynamics
perspective because the approach takes into account not just the effects
of causality and determinism, but also relates these effects to the
continuous network connection aspects of strategic behaviors. We will
return to this concept in Sec. 18.5, but first turn to an example based on
the above example.
For the software problem above, such effects might also be evidenced
by critical exponents analogous to fluids, which indicate internal scaling
symmetries. For example, there is the well-known mythical staff-month
that says if a team of N people can build a small project; then 10N can
build a project 10 times larger. The fact is that it takes more than 10
times the number of staff: you need N 1 . The number is the critical
exponent, which goes up because in projects, the more people you have
the more managers you need to facilitate communication. Such managers
however don’t build code so the productivity goes down. A theory that
includes the strategic aspects of decision making will include that
[Thomas G. H., 2006]. Such critical exponents reflect the network
connectivity character of fluids.
Structural Distinctions for Decision Processes 715
Morphogenesis need not just involve causality, but may reflect changes
in shapes as functions of strategic positions. A typical system behavior is
a stable behavior until an external event jolts it. The jolt can be
represented by a characteristic superposition of harmonic steady state
waves. We provided an example of this in our software project example
(Sec. 18.2). The jolt in this case was the addition of new software
requirements. Another example of a jolt might be the meltdown of the
market due to toxic securities.
It is convenient to know the base line behavior of the system, such as
it is in free fall. This does not reflect all aspects of the system however.
Realistic systems are usually more complicated because they have been
jolted. We must include the ability to jolt the system or set up steady
state waves in the system to reflect realistic behaviors. Harmonic steady
state waves provide the dynamic mechanism for studying such behavior.
As the software project example illustrates, a jolt to the system may
not generate the result expected by the players at the time. We must
include other dynamic mechanisms to accurately assess the outcome. In
decision process theory, we take into account these other dynamic
mechanisms. We assign a particular harmonic behavior along an initial
streamline and use the theory to compute the behaviors along all other
streamlines. In this way we consider the network connectivity effects; in
the case of the software project, such other effects were the learning
curve of new hires. This corresponds to a streamline that was not
associated with free fall behavior.
Technically, we are moving away from a description in space and
time to a description in space and frequency. Consider the following. At
each point in space, the flows (and more generally the frames) can be
716 Geometry, Language and Strategy—Vol. 2
The initial model represents the system prior to adding a “jolt”. In the
f0AD model, we define the still point behavior as that which occurs when
there is no active acceleration Q 0 , Eq. (15.31): collectively the
absolute forces, competitive forces and cooperative forces cancel. We
now add the jolt, so that the coordinate flows, Fig. 18.3, show the effect
of the added harmonics. The seasonal harmonic is not visible on the scale
of this figure as the frequency is small compared to the added harmonics.
The coordinate behavior along a streamline, Figs. 17.38 and 18.4,
reflects a helix-like behavior.
We convert this time picture of the dynamics to the frequency picture.
For each frequency, we have a phase shift, Fig. 17.5 and modulus, Fig.
17.6. These pictures contain the same information as the time domain
picture. The advantage of looking at the frequency domain is that the
entire problem appears steady state; of course this is deceptive because
the actual behavior is obtained by adding the frequency components,
Structural Distinctions for Decision Processes 717
with their phases, moduli and sinusoidal wave factors, to obtain the time
dependence.
The “jolt” in time is translated into a set of frequencies that must be
included, which in turn generate for each frequency the amplitudes and
phases. However, the time picture is also deceptive since it is not always
easy to compute or envision what happens after a given “jolt”. The
frequency picture provides a complementary approach to the problem.
elections. The ads need not address the issues so in that sense their
frequency is much larger than the seasonal payoff. Other dynamic
mechanisms will determine how these effects propagate. Our view is that
advertising changes the narrative.
in Sec. 9.5 where high stake and low velocity behaviors generated
trapped behaviors, Fig. 9.15. These network distinctions are new and
don’t arise in game theory. They are specific to theories such as decision
process theory. They are also expected to be observed in the market
place.
As a business example, when Japanese practices dominated
manufacturing because of adoption of Deming’s management method
[Walton, 1986], manufacturing in the United States was not unaffected.
Indeed those practices returned to the United States and caused an
effective revolution in quality for US made products. Adopting
techniques currently being used by someone else is a network
connection. We are not waiting for another decision to be made by
ourselves; we are mimicking what someone else is doing because they
are in a similar space at a neighboring point in time.
These are gradient effects over the network, which propagate
causally (Cf. Exs. 9, 10, 16, 17 and Figs. 18.31-18.33). An analogy is the
pressure gradients in weather. The flow of air reflects the causal nature
of the weather. We effectively describe the weather however by
considering not the flow of air but rather the areas of high pressure and
low pressure. The pressure gradients (as seen over distance) from high
pressure to low pressure generate the forces, which help us understand
the causal motion of the air. These gradients reflect the elastic network
connectivity of the atmosphere.
In a similar fashion we look for pressure gradients in decision
processes (e.g. Ex 15). They measure the network connectivity forces.
We look for evidence of network player bonding of strategic areas.
Areas that compress should be under higher pressure than those that
don’t. The ability of a strategic area to resist player bonding, its bounce,
is a measure of the elasticity of the decision process in that area.
Decision processes reflect a complex scenario of network interactions.
The measure of network connectivity will be based on the change
between players executing a particular strategy and other players
executing a neighboring strategy.
Again as a business example, when a company comes out with a
revolutionary and successful product, we might well believe that it is the
result of causal activities that happen solely within that company. That
724 Geometry, Language and Strategy—Vol. 2
Based on the presence of strategy bias fields and player payoff fields
in decision process theory, energy storage is related to the ideas in
electrical engineering circuits. In that domain, engineers have accepted
the idea that electric and magnetic fields are real and can be used to build
useful circuits. We can’t take these ideas at face value however. We must
modify these ideas somewhat since our decision systems don’t have all
the attributes of a circuit.
Fortunately, not all systems that are studied in electrical engineering
are circuits either. Engineers do study transmission lines and view them
as distributed element systems [Wikipedia, Distributed Element Model,
2015]. Maxwell’s partial differential equations define such systems.
They lead to the types of travelling and standing waves we have been
discussing in decision process theory. These equations are based on the
existence of electric and magnetic fields, whose behaviors are not always
easy to deduce.
Circuits, by contrast are lumped element systems. For such systems
[Wikipedia, Lumped Element Model, 2015], it is assumed that the
energy and momentum as well as all of the electrical properties are part
of a lumped component such as a resistor, capacitor or inductor. These
components are connected by idealized wires with no property other than
they carry current. More complicated components can be added without
changing this basic assumption. There are two advantages: the
approximation accurately reflects what engineers want to build and the
equations are ordinary differential equations that are significantly more
tractable. These advantages provide a means for understanding the
distributed element systems.
In some sense, the distinction between lumped elements systems and
distributed element systems goes back to Newton and Huygens on the
nature of light: is it particle or wave. The current version of this
argument is whether matter is particle or wave. Each description has
advantages giving insight into the nature of things. We have taken one
point of view that the field (wave) is the more basic.
This discussion is relevant to decision process theory in that we too
are solving complicated partial differential equations. Moreover, the
mathematics of the equations generates Maxwell like equations. We have
electric and magnetic fields. We have in addition, metric components
Structural Distinctions for Decision Processes 729
that generate gravitational fields. Though we are not proposing to use the
lumped element approximation, the concepts may give useful insights.
The lumped element approximation provides a physical picture of the
objects that store energy. For example, resistors dissipate energy and so
relate to heat and loss; capacitors store electrical energy; and inductors
store magnetic energy. Each of these has an analogy in decision process
theory. In a dynamic system, energy can flow from the strategy bias
fields to the player payoff fields, which corresponds to storage of energy
in a capacitor like object to an inductor like object. Energy can be lost in
the stress fields (pressure fields), corresponding to resistance.
The additional possibilities however are not part of the electrical
engineering analogy: there can be energy stored in the gravitational and
metric fields. This includes a type of magnetism (Coriolis force) as well
as a scalar gravitational field. These are related to the way in which
utility is ascribed to different strategy possibilities.
This approach has been used by others to understand the behaviors of
fluids [Wikipedia, Mobility Analogy, 2015]. The lumped element system
effectively becomes a systems dynamics problem. Not surprisingly, the
stocks that store things in systems dynamics are ideal distinctions for how
energy is stored. In addition to finding analogs of the electrical
engineering concepts of heat, electrical and magnetic energy in fluids,
these studies find properties that are specific to fluids: one such concept
is inertance: it stores the inertia of the fluid based on the fluid’s
compressibility.
Based on such general considerations, we expect that energy in
decision process theory will be stored in specific subsystems as follows.
Energy will be stored as electric energy and there will be identifiable
capacitive subsystems that accomplish this storage. We expect energy
in this subsystem to be identified with the player strategy bias fields,
or game value.
Energy will be stored as magnetic energy and there will be inductive
subsystems that accomplish this storage. We expect energy in this
subsystem to be related to the player payoff fields.
Energy can be stored as metric field energy and there will be gravitas
subsystems that accomplish this storage.
730 Geometry, Language and Strategy—Vol. 2
behavior, Sec. 16.3, we increased the player strategy bias for honoring
the contract to f 3 5 , which has the effect of separating the “blue” and
“red” player engagements. When “blue” acts more aggressively,
Fig. 18.8, we see a charge structure showing morphogenetic changes.
determined by the way in which one sets the initial conditions. Those
conditions could be to require certain equilibrium values or could be to
require certain initial player values (Chap. 14).
What we can say is that for free fall, there will be a mixed strategy
choice that sets the rate at which each strategy changes. There will also
be a seasonal payoff matrix that will determine the resonant frequencies.
Additional dynamic mechanisms from the theory determine the detailed
relationship of these free fall values and initial boundary condition values
for the players.
For the player fixed frame models, using the definitions and Eq.
(15.30), Ex. 1 from Sec. 15.15, we can write the frame equations Eq.
(4.32) as Eq. (15.31). From these equations, we glean that for behavior at
the still point, we have zero active geometry acceleration (Fig. 18.11):
the still point behavior is then set by :
q 2 e e e
Q
1 e e
2
1
1 E k Ek
1 e e
1 2 Ek f k
1 2 e q f e (18.1)
e 1 2 e q f
E a 0
E a E a
We don’t expect this to hold away from the initial boundary. For
example it might not be true along an initial streamline, even though
along that streamline we can take the metric to be Minkowski and make
simple model choices for the orientation potentials, so that they are small
or correspond to a static gravity field. Indeed, these choices describe the
basic assumptions or models used in Vol. 1. They provide a baseline
structure on which we can look for new behaviors.
For example, in Vol. 1 we ignored the fact that the inactive scalar
fields and the payoff fields may vary along the streamline (Cf. Ex. 30,
Sec. 6.10). We ignored the second order effects. If we include these
Structural Distinctions for Decision Processes 737
having no active geometry acceleration forces: these are the sum of the
absolute force, competitive or game theory force and the cooperative
force. Game theory identifies such behavior as Nash equilibrium.
Though there are differences in how game theory and decision
process computes the individual player behavior, both theories provide
essentially the same distinction. We use the insights learned from game
theory to set the initial conditions in decision process theory. We do this
while remaining consistent with the ideas of causality, without resorting
to Bayesian probability as a predictor for future behavior, which is also
consistent with at least one view of game theory [Von Neumann &
Morgenstern, 1944]. The important difference between game theory and
decision process theory is that the latter has second order effects, which
are expected to generate morphogenic changes. These changes result
from the field equations and the specific initial conditions. Different
initial conditions in general lead to structures that are not the same; they
are not diffeomorphic.
Now let us deal with the second behavior, which is provided by the
initial conditions (which in the time domain would be at t 0 ). These
conditions provide the applied force, which generates the acceleration
effects that change the geometric structure. The acceleration effects may
manifest themselves as flows that change in time or as Coriolis effects in
which the flows appear fixed, along with fictitious forces that appear,
reflecting the fact that the description is in a frame that is rotating.
To take into account both behaviors, our strategy has been to choose a
frame of reference so that at the still point there is no acceleration. Away
from that point there may be acceleration. The numerical results and the
mechanisms away from the still point depend on the details of the
acceleration.
The details of the acceleration are generated by the forced harmonics.
We look at the harmonics (Sec. 18.4) in the central holonomic frame,
Sec. 4.6, which is related by (a non-diffeomorphic) linear transformation
to the normal-form coordinate basis. We examine the model in this
frame and then transform the results back to the normal-form coordinate
basis.
In the central frame, the active strategy flows are independent of time
(indeed, in the central holonomic frame for the fixed frame model, the
Structural Distinctions for Decision Processes 739
space components are zero and the time component is unity) while by
assumption, the covariant rate of change of the flow with time is not
constant. This is consistent with the perception that such effects appear to
be absent.
To summarize, there are two mechanisms, each leading to distinct
harmonic behaviors: the free fall harmonics and the forced harmonics.
The latter are harmonic steady state wave behaviors (forced behaviors),
with arbitrary frequencies, and replace a time domain picture of the force
with a frequency domain picture (Sec. 18.3). The former free fall
harmonic structure generates resonant harmonics based on the seasonal
payoff. We may get morphogenic changes when these behaviors are
combined. Therefore, it is useful to separate out the equilibrium
behaviors as an important structural distinction.
18.10 Acceleration
decision. We could ask in terms of frequency, how they would rate their
choices. The survey is not predictive in the probability sense but provides
information about the initial state; decision process theory makes
assertions about how this state evolves in time: how a later survey might
be answered.
In this way the theory generalizes the static flow considerations of
Game theory to steady state flow and dynamic flow. Since the flows
need not be constant, there will be acceleration defined as the rate of
change of each strategy flow. We expect the flow of decisions to change
in direction and magnitude. These changes provide a direct means to
measure the new distinctions in decision process theory.
We distinguish between the player’s worldview of payoffs that they
use to make their decisions and the seasonal worldview that reflects the
decisions that actually have been made. In free fall when there are no
forces other than this seasonal worldview, the strategy choices reflect
rotations. Indeed we argue more generally that the concept of player
payoffs and seasonal payoffs reflect some type of rotation.
Qualitatively, we generalize Nash equilibrium. If players act on the
basis of their individual payoffs, then Nash argues that there will be an
equilibrium choice that is in some sense best in that no player can do
better because of what other players would do in response. The Nash
equilibrium is thus an optimal strategy for the seasonal payoff.
Conversely, if we know the Nash equilibrium we imagine a variety of
rotations that leave this direction unchanged, any one of which is a
possible seasonal payoff. We say that the collective behavior will always
be a rotation, so there will always be a seasonal payoff that measures that
rotation. For any rotation, there will always be directions that are left
fixed. Decision choices along such a direction correspond to Nash
equilibriums.
Acceleration is what takes us away from this equilibrium.
Acceleration behavior is converted from the time domain to the
frequency domain with the steady state harmonics (Sec. 18.3). These
harmonics will exhibit either travelling wave or standing wave behavior,
subject to limits to choice. We expect the acceleration effects to be
captured by the transfer functions. The transfer function behaviors obey
Eq. (14.36) and so depend on only a few of the co-moving frame tensors:
Structural Distinctions for Decision Processes 741
the time metric, the acceleration vector q , the player bonding Θ , the
time metric e 2 q 1 e e and the seasonal vector potential a .
1
The time metric factor, Fig. 18.13, vanishes near the boundary. The
coefficient of the linear term is proportional to 2 e 2 q 1 e e 4aa .
Since the zero frequency solutions are unbounded, we expect that we
may get unbounded solutions near the boundary, which is whenever this
factor changes sign. This is what we find from the numerical
calculations. A large negative acceleration potential accomplishes the
same thing.
We have computed the acceleration potential previously, Fig. 17.26,
where we see that there are no areas where the potential is large and
negative. The other possibility would be singular points due to the player
bonding, whose bonding potential is Fig. 17.27, which is well-behaved
for this example. Finally, the seasonal vector potential is very small,
corresponding to a small seasonal payoff, so is also expected to play a
small role. The main variations that remain are due to the frequency
variations. We expect that very high harmonics lead to high frequency
spatial behaviors.
742 Geometry, Language and Strategy—Vol. 2
from those areas. See the player bonding z 0,0, z , Fig. 16.3 as an
example.
The behavior of those choices would appear to accelerate from the
high pressure areas to the low pressure areas. Such effects generate
changes in momentum, a term often used in sports and in the stock
market. We are familiar with team behavior improving suddenly as if
they pick up outside support. We see their fortunes accelerate.
The opposite of acceleration is inertia. In the f0AD model, we have
assumed a fairly high decision mass, md 40 which corresponds to
significant inertia. Being slow to change reflects large inertial effects. In
fact there might be confusion over behavior that has settled into
equilibrium versus behavior that is not changing because of inertia. We
think of companies that are not making a profit but are set in their ways.
There may in fact be winning strategies that they could adopt if it were
not for the inertia and lethargy of the organization.
18.11 Outcomes
will have found support for the three W’s: work, wisdom and wealth.
These could be the basis for successful decision structures.
From Sec. 18.6, the student will have learned that symmetries
underlie a key set of structures. In decision process theory they
determine what it means to be an individual, what it means to honor
a contract, and what sustainability might involve. In general,
symmetries are not derived from a theory but may exist in the theory
as aspects that are part of the empirical framework. Once observed, it
is the job of the theory to represent them.
From Sec. 18.7, the student will have learned that the fields we have
in mind are real and carry energy that can be stored and can be part of
replicable components of any process. In a lumped element
approximation in analogy to electrical engineering, one can imagine
capacitors that store electrical energy (capacitance, which in decision
process theory would correspond to value), inductors that can store
magnetic energy (inductance, corresponding to utility through
payoffs), gravitational energy (gravitas, cooperation and player
bonding, as resonant energy) and fluid mechanical energy (intertance,
stresses due to players that are not explicitly included).
From Sec. 18.8, the student will have learned that there are not only
limits to growth, but limits to choice. This distinction places a
structural constraint on the size of the spatial geometry. It helps
distinguish how a system moves from a “flat” behavior to a curved or
chaotic behavior. The student will have learned some of the
mechanisms that cause this to happen.
In Sec. 18.9, the student will have learned the importance of
identifying the areas that are flat, in which the behavior is free fall
and most closely aligned with game theory. Decision process theory
will have solutions in which behavior along a path will have a still
point (Sec. 15.1) at which the active geometry acceleration is zero.
This generalizes the notion of Nash equilibrium from game theory
and allows the use of game theory as a baseline for characteristic
decision behavior. A special case of still point behavior is one in
which the strategies appear to be in a flat space. Free fall behavior is
determined by the seasonal harmonic behaviors with frequencies set
746 Geometry, Language and Strategy—Vol. 2
18.12 Exercises
(1) Modify the attack-defense model f2AD (Sec. 16.13) to the f2AʹD
model in which the payoffs will be changed for “red” to Eq. (16.3)
as suggested in Sec. 16.3:
0 0 2 5 310 813160
0 0 110 2 5 813160
Fab 2 5 813160 ,
1 blue 1
10 0 0
3
10
2
5 0 0 813160
813 813160 813160 813160 0
160
0 0 0 1
10 301 60
0 0 1 0 301 60
5
Fab 0 60 .
2 red
15 0 0 301 (18.2)
1
10 0 0 0 301
60
301 0
60 301 60 301 60 301 60
md 4
B 3 110 0 1
10
f 3
z 10
. (18.3)
e1 e2 0
1 2 2 e
H a k 0,0,0 2 10 3
10
4
10
5
10
Figure 18.19. Recurrence plot for the f2AʹD model relative strategy
component
750 Geometry, Language and Strategy—Vol. 2
(4) Based on the initial conditions, Eq. (18.3), the player aggression
components f2AʹD model have a common factor of 4 10 and two
additional frequency components. We see in Fig. 18.17, the
frequency dependence of the transfer function for both the seasonal
resonance frequency and the forced frequencies. Plot the modulus
as a function of frequency. Add more points to this plot to obtain a
feeling of the overall frequency dependence. What does it tell you
about the time dependence?
(5) For the f2AʹD model, the forced harmonics are travelling waves, Fig.
18.18. Because of the limits to choice, the attenuation factor
p 1.801 , Eq. (16.17) is close to the forced harmonic frequencies.
What happens if we choose this as the first harmonic 1 1 2 as in
the f2AD model?
(6) For the f2AʹD model, the recurrence plot, Fig. 18.19 demonstrates
causality and determinism. Discuss the differences of this plot to
that of the f2WWW model, Fig. 16.51.
(7) For the f2AʹD model, think of the co-moving frame transverse active
strategy z as a measure of player aggression. The gradient term is
seen to decrease the flow V u as z increases, Fig. 18.20. What
actions do you expect “blue” to take as a consequence? Compare
your answer to the model results, Fig. 18.21.
(8) For the f2AʹD model, we have computed the contour flows,
Fig. 18.22. Because of network connections, argue that the
recurrence plots for the player aggression will change as you
increase z , the relative player aggression. Based on the contour plot
provided, how will the recurrence plots change?
Figure 18.28. Initial active acceleration vector for the f2AʹD model
(12) The f2AʹD model is an example of high bias that generates limits to
choice. An example of the impact of high bias is provided in Ex. 3.
Another example is provided by the density distribution of the time
metric factor 1 e e , Fig. 18.25. Compare the structure to
Fig. 18.13.
756 Geometry, Language and Strategy—Vol. 2
(13) Compare the initial values of the acceleration gradient for the f2AʹD
model, Fig. 18.26 to that of the f0AD model, Fig. 18.14. Why is the
former so much larger? As a second comparison, look at the
acceleration potential for the f2AʹD model, Fig. 18.27. Why is the
active acceleration so much smaller for this model, Fig. 18.28?
(14) For the f2AʹD model, the active acceleration becomes large near the
boundaries, Fig. 18.29. Explain why this comes about.
Figure 18.31. Recurrence plot for stock price ratio (dB) between
two similar companies
(15) There are strong player biases for the f2AʹD model, leading to a
change in the pressure, Fig. 18.30 compared to the f2AD model from
Sec. 17.8, Fig. 17.25. Explain the quantitative increase in pressure
with the new model and the change in structure.
(16) In Sec. 18.5, the argument was made that not only should we find
causal relationships for fields, but also for neighboring values of
fields, such as Fig. 18.24 and Exs. 9-10. Consider two similar
companies such as Nokia and Alcatel. In our view, the difference of
their stock prices reflects a price gradient over nearby strategies. We
758 Geometry, Language and Strategy—Vol. 2
Figure 18.32. Mean price of two similar Figure 18.33. Price ratio of two similar
companies companies
Chapter 19
759
760 Geometry, Language and Strategy—Vol. 2
but steady-state. Time is then the symmetry variable. For such behaviors,
we expect to find as evidence, field behaviors that are constant when
viewed from the appropriate point of view: namely when viewed in a co-
moving frame in which the active strategy flows are zero.
f 2 2 e . (19.1)
Our interpretation is that each player has two views of the decision
process corresponding to the two terms in the equation.
In one view the player makes choices depending on the collective
behavior represented by the composite payoff, . In the other view, the
player makes choices depending purely on their own worldview, ;
they are empowered without regard to other player behaviors.
In a positive sense of the word, they act with entitlement (Sec. 15.4).
They believe they have a right to make their own decisions and manage
their own interactions. They are focused and forcefully state their
position in making their decision. They are in the “zone”. Entitlement
might be viewed as a self-centered and egocentric component, though
without the intent to necessarily harm others. Entitlement just doesn’t
take into account the other point of view; they only see the other through
their private payoff map.
In our models of a societal framework, the entitlement effects are
steady-state, though as with the charge, Figure 19.1 and other steady
state fields, the entitlement effects will be seen as time and space
dependent in the normal form coordinate basis. The private payoff map
is the entitlement payoff in Eq. (19.1), 2 , which in the 3D notation
is 2Bα .
766 Geometry, Language and Strategy—Vol. 2
Away from the still point, we don’t expect the composite payoff
to vanish. We close the loop and express the composite payoff in terms of
the seasonal payoff and the entitlement payoffs of all of the players,
Eq. (14.30):
each of the other players will do: it is that player’s view of the payoffs.
As an example, for the f2WWW model, the player payoffs for “work”
Fig. 19.2, and “wealth” Fig. 19.3, represent and extend the game
equivalency behaviors.
may favor one, the other or neither of these two terms. If there is no
player engagement then the player payoff is all entitlement. If there is no
entitlement then the player payoff is determined by the player
engagement payoff, e ω . The theory produces this model with two
mechanisms and allows us to study their mutual interaction. The
seasonal payoff and the entitlement payoffs of all the players determine
the composite payoff, Eq. (19.2), such as Fig. 19.5 for the f2WWW
model. We hope to find empirical evidence for this decomposition.
loses and what payoffs occur in each case. The strength of compliance is
set by the player engagement. This is the opposite case of the player
paying no attention to what others appear to be accepting as the rules of
the game and acting based on their own sense of entitlement.
don’t know, but whatever it is it must be great because the line is really
long.”
The idea that players may interact with each other outside the confines of
the competitive decision process was envisioned by Von Neumann &
Morgenstern [1944]. However, parts of that proposal were flawed (Cf.
the discussion in Sec. 12.4). We believe we have a replacement of that
concept without the flaws that were identified. We replace their proposal
with the idea of cooperation potentials, Sec. 12.4, which we refine here
as mutual player support potentials. These are social constructs within
decision process theory.
Though the mutual player support potentials are specific to decision
process theory, we see some similarity between them and the
imputations and side payments introduced by Von Neumann &
Morgenstern [1944]. They are payments outside the framework of the
competitive valuations of the forces as measured by the payoffs. The
difference is that in decision process theory, these side payments are not
outside of the theory, which as constructed is a self-consistent
Social Distinctions for Decision Processes 779
19.7 Hidden-in-Plain-Sight
19.8 Outcomes
Based on this chapter, the student should have learned a number of social
distinctions that might arise as we move from local behaviors to global
Social Distinctions for Decision Processes 783
19.9 Exercises
(1) Discuss the differences between the code of conduct charge for the
f2WWW model, Fig. 19.1 and the corresponding plot for the f2AʹD
model, Fig. 19.8 in terms of the social aspects of each model. Note
the former is designed as a social model and the latter an attack-
defense model.
(2) Discuss the differences between the code of conduct charge for the
f2WWW model, Fig. 19.9 and the corresponding plot for the f2AʹD
model, Fig. 19.8 in terms of the social aspects of each model. In this
case the f2WWW model plot is for an extended time scale. How is
this different from Ex. 1?
786 Geometry, Language and Strategy—Vol. 2
(3) For the f2AʹD model, for the entitled payoffs for “blue” and “red”,
we have Figs. 19.10 and 19.11. Compare this with the f2WWW
model, Figs. 19.2 and 19.3. Note that in both cases, the behaviors
follow the transverse boundaries for points sufficiently near.
Discuss the implications of these behaviors for the two models.
(4) Compare the f2AʹD model for the entitled code of conduct payoff,
Fig. 19.12 with the corresponding plot Fig. 19.4 for the f2WWW
model. In this case, how do the boundary conditions influence the
behaviors? Include in the definition of the boundary, the behavior
on the surface z 0 . Similarly compare the composite payoff
Figs. 19.13 with 19.5.
(5) Charge contours for the f2WWW model are shown in Figs. 19.14
and 19.15. Compare and contrast this with the corresponding plot
for the f2AD model, Figs. 18.9 and 18.10.
(6) The player engagement payoffs, e ω , for the f2WWW model are
plotted in Figs. 19.16-19.18. Compare these payoffs with the player
entitlement payoffs, Figs. 19.2-19.4. Based on the sizes as provided
in the legends, describe the resultant player payoff in the central
co-moving frame, Eq. (15.8).
790 Geometry, Language and Strategy—Vol. 2
Afterword
This is phase II of a project that grew out of a desire to create a new way
of talking about decision processes. In phase I, Vol. 1, I provided an
introduction to the ideas and qualitative reasons and simplified
calculations for seeing how one might address decision processes using
the language of physics and differential geometry. The idea was to use a
language that was created to describe and understand processes that
evolve in time according to specified dynamic mechanisms. In the first
phase, no attempt was made to provide a comprehensive solution to the
equations, nor was there an attempt to educate the readers to the level
where they could solve the equations themselves.
This is phase II of the project. I have identified the types of solutions
found in phase I as the analog of AC resonant solutions; game theory is
analogous to DC circuits. In addition to extending the coverage to AC
steady state solutions, in this volume, I attempt to make this
identification mathematically sound. The goal has been to provide the
necessary background to bring anyone with basic mathematics and
engineering foundations up to the level of a new way of talking about
decision processes so that they can apply the theory to real world
problems. The goal is that the reader should gain sufficient expertise to
understand how the ideas apply. That has entailed providing essential
background for the mathematics, the physics and the economic theory.
The next phase is the interesting challenge to apply the tools from
phase II and the ideas from phase I to practical decision processes. It is
also the opportunity to broaden the models considered and investigate
transient solutions more thoroughly. I envision that the way of looking at
decisions that has been elaborated here will involve looking at empirical
795
796 Geometry, Language and Strategy—Vol. 2
data in ways that will generate significant insights. It may also require
new ways to look at data. I intend to chart progress for this phase on the
following website: http://decisionprocesstheory.com. This site will also
have helpful information on the first two phases of work, including
Mathematica notebooks that are not provided in this volume.
It is of course difficult to predict how future studies may progress.
Our current view is as follows. So far, we have used the player fixed
frame model as our guide. We believe our next guide will be the player
ownership model, which extends what we have learned with an energy
momentum tensor that reflects the holonomic constraints.
(1) Pick a problem for analysis and start with a game theoretic view.
(2) Assume there is a stationary holonomic frame in which time is
inactive and mutually commutes with the player inactive strategies,
including any code of conduct strategies. Identify the players,
including the codes of conduct players.
(3) Choose the independent metric components to be the gauge
invariant quantities jk g
ab
F j ab , (which in a centrally
co-moving frame determine player engagements V j and seasonal
flows V ). Choose the gauge independent holonomic variables to
be time and the active strategies.
(4) Transform to the frame in which the flows V ,V j and charges
V ,V j are zero.
(5) Set the values of the energy momentum tensor using the player
ownership structures in this frame (Sec. 6.10, Ex. 40):
j j, k o J , o J
J jk k , J ,
0 otherwise 0 otherwise
J jk 0 0
QJ 0 0 0 . (20.1)
0 0 J
J i l j V V j J jl ,
J a V V J ,
J
V j V V j J j kV k V V J V ,
J i j J
j
J a
j
k
QJ J i k J J a .
(20.2)
J i k J
J a
p J Q J g Q J J Q J Q J . (20.3)
J J
(8) Set the gauge; e.g. use the seasonal gauge, Sec. 6, Eq. (6.98).
(9) Solve the resultant elliptic partial differential field equations in the
central stationary frame. The equations are now all elliptic since
they are independent of time. Use a general purpose solver to focus
on solutions rather than techniques.
(10) Match the boundary condition of the problem under investigation
using the harmonic wave equation.
(11) Compare the results of the simulation against observed behaviors to
establish the relevance of the model.
799
800 Geometry, Language and Strategy—Vol. 2
807
808 Geometry, Language and Strategy—Vol. 2
seasonal payoff, 241, 246, 331, 569, steady state surface geometry, 262
688, 737, 739 still point, 503, 545, 550, 582, 737
seasonal player, 570 stocks, 471
seasonal potential, 216, 246, 518, 530 Stokes’ Law, 42
seasonal timeframe, 592, 594, 612 strain, 196, 303, 309
seasonal vector potential, 241, 517, strategic boundary, 40, 49
559 strategic clustering, 567
self, 375, 422 strategic decomposition, 371
self-interest, 302, 413, 421 strategic form, 430
self-love, 416 strategic occupation, 75
sellers, 456 strategic opinion, 501
shear potential scalar, 681 strategic surface, 40, 45
signal, 52 strategic viscosity, 221
single active strategy, 234 strategic volume, 40
sinusoid, 742 strategy contours, 790
skill boundary, 556 streamline proper-time scalar field,
skill preference subspace, 556 214
social distinctions, 705 streamline solutions, 214, 234
solipsistic, 568 streamlines, 345
space dimension, 130, 147 stress, 309
spacelike, 141 stress tensor, 198, 201
spaceline, 364 stress tensor components, 101
speed, 89 strictly determined, 432, 447
spin, 196 strong ownership, 252
staff-seconds, 39 structural coupling, 200
staff-years, 39 structural distinctions, 705
standard of behavior, 277, 421, 436 structural stability, 264, 303, 706, 707,
static games, 335 708, 744
stationary, 161 structure constants, 59
stationary flow, 12 subsidy-seeking, 418
stationary holonomic frame, 796 successful decision structures, 724,
stationary line gauge, 330 745
steady state, 209, 210 sustainability, 706, 745
steady state behaviors, 403, 448 sustainable, 764
steady state flow, 12 symmetric decision process, 281, 372
steady state geometry, 299, 301, 324 symmetric normal-form coordinate
steady state harmonics, 228 basis, 284, 503
steady state hypothesis, 211, 217, 221, symmetric prisoner’s dilemma, 340
525 symmetry, 105
steady state orthonormal frame, 211 system response harmonics, 497, 591
steady state scenarios, 261, 264, 272 systems dynamics, 262
steady state solution, 447
816 Geometry, Language and Strategy—Vol. 2
The Series on Knots and Everything: is a book series polarized around the theory of
knots. Volume 1 in the series is Louis H Kauffman’s Knots and Physics.
One purpose of this series is to continue the exploration of many of the themes
indicated in Volume 1. These themes reach out beyond knot theory into physics,
mathematics, logic, linguistics, philosophy, biology and practical experience. All
of these outreaches have relations with knot theory when knot theory is regarded as
a pivot or meeting place for apparently separate ideas. Knots act as such a pivotal
place. We do not fully understand why this is so. The series represents stages in the
exploration of this nexus.
Details of the titles in this series to date give a picture of the enterprise.
Published*:
*The complete list of the published volumes in the series can also be found at
http://www.worldscientific.com/series/skae