A First Course in “In Silico Medicine”

Volume 2
Series Editor
Masao Tanaka
Professor of Osaka University
1-3 Machikaneyama, Toyonaka
Osaka 560-8531, Japan
[email protected]
Shinji Doi • Junko Inoue • Zhenxing Pan
Kunichika Tsumoto

Computational
Electrophysiology
Dynamical Systems and Bifurcations

123
Shinji Doi Zhenxing Pan
Professor Doctoral candidate
Graduate School of Engineering Graduate School of Engineering
Kyoto University Osaka University
Kyoto-Daigaku Katsura, Nishikyo-ku 2-1 Yamadaoka, Suita
Kyoto 615-8510, Japan Osaka 565-0871, Japan
[email protected] [email protected]

Junko Inoue Kunichika Tsumoto

Associate Professor Specially Appointed Assistant Professor
Faculty of Human Science The Center for Advanced Medical
Kyoto Koka Women’s University Engineering and Informatics
38 Kadono-cho, Nishikyogoku, Ukyo-ku Osaka University
Kyoto 615-0882, Japan 2-2 Yamadaoka, Suita
[email protected] Osaka 565-0871, Japan
[email protected]

ISBN 978-4-431-53861-5 e-ISBN 978-4-431-53862-2

DOI 10.1007/978-4-431-53862-2
Springer Tokyo Dordrecht Heidelberg London New York

Library of Congress Control Number: 2009943556

MATLAB R
is a registered trademark of The MathWorks, Inc., 3 Apple Hill Drive, Natick, MA 01760-
2098, USA. http://www.mathworks.com

c Springer 2010
This work is subject to copyright. All rights are reserved, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting,
reproduction on microfilm or in any other way, and storage in data banks.
The use of general descriptive names, registered names, trademarks, etc. in this publication does not
imply, even in the absence of a specific statement, that such names are exempt from the relevant protective
laws and regulations and therefore free for general use.

Cover Illustration: 
c MEIcenter

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

Preface

Life is a dynamic, ambiguous, and fleeting system. Computational biology and

systems biology use mathematical or computational models, usually denoted as
dynamical systems such as difference equations or differential equations, for the
analysis and understanding of such biological systems. In general, these models in-
clude many parameters, and these parameters inherently possess much ambiguity
or uncertainty; thus it is important to treat the models with the consideration of
such parameter uncertainty or changes. Bifurcation theory in nonlinear dynamical
systems provides us with a powerful tool for the analysis of the effect of parame-
ter change on a system and detects a critical parameter value when the qualitative
nature of the system changes, whereas numerical or computer simulations give us
only one solution under a fixed set of parameter values and initial values.
Our aim is to provide an introduction to computational electrophysiology, par-
ticularly to the nonlinear dynamics of Hodgkin–Huxley (HH)-type models, together
with a brief introduction to the dynamical system and to bifurcation analysis. This
textbook includes many examples of numerical computations of bifurcation anal-
ysis on various models, ranging from the classical HH model of a squid giant
axon to the Luo–Rudy (LRd) dynamic model of a cardiac cell, using the famous
computer software XPPAUT [see B. Ermentrout (2002) Simulating, Analyzing, and
Animating Dynamical Systems: A Guide to XPPAUT for Researchers and Students.
SIAM, Philadelphia], including AUTO software (http://cmvl.cs.concordia.ca/auto/).
We hope that this work will be useful as a practical, quick guide to the numerical
bifurcation analysis of other models that readers have to analyze.
This textbook, Volume 2 in the three-volume series A First Course in “In Silico
Medicine,” is organized as follows:
Chapter 1 provides the basics of dynamical system theory necessary for understa-
nding the dynamics and bifurcations arising in the models encountered in the
succeeding chapters. Chapter 1 also provides very brief instructions in XPPAUT,
so that readers can perform bifurcation analysis similar to that presented in this
textbook.
Chapter 2 explores the nonlinear dynamics of the famous HH model of a neuronal
excitable membrane in detail and explains what makes up its neuronal features,
such as the excitability threshold and refractoriness. Part of the explanation of
HH dynamics is indebted to the very old but impressive paper [FitzHugh (1960)

v
vi Preface

Thresholds and plateaus in the Hodgkin-Huxley nerve equations, J. Gen. Physiol.

43, pp. 867–896]. This seminal work provides a good example of the direction that
contemporary computational biology should take. The geometric methods in the
dynamical system rather than brute-force computer simulations are very useful for
understanding the essential features of neuronal and nonlinear dynamics.
Using the Bonhoeffer–van der Pol (BVP) neuronal model [or the
FitzHugh–Nagumo (FHN) model], which is a simplification of the HH model,
Chapter 3 explains neuronal dynamics more geometrically and intuitively. In the
rest of Chapter 3, we take a slightly lengthy, roundabout route via more simplified
or abstract neuronal models. In particular, we show that many neuronal models aris-
ing in different contexts are systematically explained by a simple one-dimensional
mapping called the phase transition curve (PTC). Through these excursions, we
would like to emphasize the importance of not only the physiological or physical
models but also the abstract models.
Chapter 4 returns to the HH model but explores it from a different viewpoint:
robustness and sensitivity. The bifurcation structure of the HH model on various bi-
furcation parameters is thoroughly analyzed, and thus the robustness and sensitivity
of the HH model on various parameters are clarified.
Chapter 5 analyzes the bifurcation structure of other HH-type models: the
Yanagihara–Noma–Irisawa (YNI) model of a cardiac pacemaker cell and the LRd
model of a cardiac ventricular cell. These models are HH-type models but possess
greater complexity than the classical HH model of a squid giant axon. Thus the
robustness and sensitivity of cardiac cells on various parameters (ion channels) are
also explored.
We would like to express our sincere gratitude to many people, especially to
Professors Shunsuke Sato, Hiroshi Kawakami, and Jose Pedro (Pepe) Segundo for
their stimulating discussion, teaching, and encouragement. We are very grateful to
Professors Taishin Nomura and Yoshihisa Kurachi, and also for the support of the
Global COE Program “In Silico Medicine” at Osaka University.

Masao Tanaka
Series Editor
Shinji Doi
Junko Inoue
Zhenxing Pan
Kunichika Tsumoto
Authors
Contents

1 A Very Short Trip on Dynamical Systems. . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . 1

1.1 Difference Equations, Maps, and Linear Algebra . . . . . . . . . . . .. . . . . . . . . . . 1
1.2 Differential Equations, Vector Fields, and Phase Planes . . . . .. . . . . . . . . . . 6
1.3 Linearization, Stabilities, Coordinate Transformation . . . . . . .. . . . . . . . . . . 11
1.3.1 Linearization and Stabilities. . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . 12
1.3.2 Coordinate Transformations.. . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . 12
1.4 Nonlinear Dynamical Systems and Bifurcations . . . . . . . . . . . . .. . . . . . . . . . . 13
1.4.1 Bifurcations of Discrete Dynamical Systems . . . . . . . .. . . . . . . . . . . 13
1.4.2 Bifurcations of Continuous Dynamical Systems . . . . .. . . . . . . . . . . 18
1.4.3 Forced Dynamical Systems and Poincaré Maps . . . . .. . . . . . . . . . . 24
1.4.4 Attractors and Basins of Attraction . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . 25
1.5 Computational Bifurcation Analysis . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . 27

2 The Hodgkin–Huxley Theory of Neuronal Excitation . . . . . . . . . .. . . . . . . . . . . 37

2.1 What is a Neuron? Neuron is a Signal Converter .. . . . . . . . . . . .. . . . . . . . . . . 37
2.2 The Hodgkin–Huxley Formulation of Excitable Cell Membranes . . . . . 38
2.3 Nonlinear Dynamical Analysis of the Original HH Equations . . . . . . . . . 40
2.3.1 Simple Dynamics of Gating Variables .. . . . . . . . . . . . . . .. . . . . . . . . . . 42
2.3.2 FitzHugh’s Subsystem Analysis of the HH Equations .. . . . . . . . . 44
2.3.3 Dynamic I –V Relation of the Squid Giant Axon Membrane . . 47
2.3.4 Dimension Reduction of the HH Equations . . . . . . . . . .. . . . . . . . . . . 49
2.3.5 Bifurcation Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . 52

3 Computational and Mathematical Models of Neurons . . . . . . . . .. . . . . . . . . . . 55

3.1 Phase Plane Dynamics in the Context of Spiking Neuron . . .. . . . . . . . . . . 55
3.1.1 Excitability and Flows of the BVP Neuron Model .. .. . . . . . . . . . . 56
3.1.2 Bifurcations in the BVP Neuron Model . . . . . . . . . . . . . .. . . . . . . . . . . 59
3.2 Simple Models of Neurons and Neuronal Oscillators . . . . . . . .. . . . . . . . . . . 69
3.2.1 Integrate-and-Fire Neuron .. . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . 69
3.2.2 Leaky Integrate-and-Fire Neuron . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . 72
3.2.3 Radial Isochron Clock as a Model of Pacemaker Cells . . . . . . . . . 74
3.2.4 Caianiello’s Neuron Model .. . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . 76
3.2.5 Chaotic Neuron Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . 78

vii
viii Contents

3.3 A Variant of the BVP Neuron Model

and Its Response to Periodic Input . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . 79
3.3.1 Piecewise-Linearized BVP Model . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . 79
3.3.2 Singular Version of the Model and Its Response
to Sinusoidal Input .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . 82
3.3.3 Phase Lockings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . 84
3.3.4 Bifurcation Diagrams .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . 86
3.4 Stochastic Neuron Models .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . 88
3.4.1 Stochastic Integrate-and-Fire Neuron .. . . . . . . . . . . . . . . .. . . . . . . . . . . 89
3.4.2 Stochastic Leaky Integrate-and-Fire Neurons . . . . . . . .. . . . . . . . . . . 90
3.5 Stochastic Phase-Lockings and Bifurcations
in Noisy Neuron Model .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . 93
3.5.1 Stochastic Phase Lockings . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . 94
3.5.2 Stochastic Bifurcation Diagrams .. . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . 97

4 Whole System Analysis of Hodgkin–Huxley Systems . . . . . . . . . . .. . . . . . . . . . . 99

4.1 Changing the Parameters: Sensitivity and Robustness . . . . . . .. . . . . . . . . . . 99
4.2 Bifurcations of the Hodgkin–Huxley Neurons . . . . . . . . . . . . . . .. . . . . . . . . . .105
4.3 Two-Parameter Bifurcation Analysis of the HH Equations ... . . . . . . . . . .111
4.4 Numerical Bifurcation Analysis by XPPAUT . . . . . . . . . . . . . . . .. . . . . . . . . . .114

5 Hodgkin–Huxley-Type Models of Cardiac Muscle Cells . . . . . . .. . . . . . . . . . .119

5.1 Action Potentials in a Heart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . .119
5.2 Pacemaker Cell Model .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . .121
5.2.1 The YNI Model of Cardiac Pacemaker Cell. . . . . . . . . .. . . . . . . . . . .121
5.2.2 Bifurcation Analysis of the YNI Model . . . . . . . . . . . . . .. . . . . . . . . . .124
5.2.3 Two-Parameter Bifurcation Analysis of the YNI Model.. . . . . . .127
5.2.4 Bifurcation and Parameter Sensitivity . . . . . . . . . . . . . . . .. . . . . . . . . . .130
5.3 Ventricular Cell Model.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . .134
5.3.1 The LRd Model of Ventricular Cell . . . . . . . . . . . . . . . . . . .. . . . . . . . . . .134
5.3.2 Bifurcation Structure of the LRd Model .. . . . . . . . . . . . .. . . . . . . . . . .136
5.4 Other HH-Type Models of Cardiac Cells . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . .138

References .. . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . .143

Index . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . .151
Chapter 1
A Very Short Trip on Dynamical Systems

The theory of dynamical systems has grown up to a very big discipline in

mathematics now, and deals mainly with measurable (or ergodic), topological,
and differentiable (or smooth) dynamics of systems. The dynamical system theory
is related to other areas of mathematics also, for example, to the (analytic or geomet-
ric) singular perturbation theory. The singular perturbation theory is very important
to analyze and understand the mathematical models in the electrophysiology (see
Coombes and Bressloff 2005; Fenichel 1979; Guckenheimer 1996; Izhikevich 2006;
Jones 1996; Jones and Kopell 1994, for example), but is not treated in this book.
This chapter briefly introduces only the minimal materials in the broad dynamical
system theory, necessary for the understanding of the following mathematical and
computational models in the electrophysiology. This chapter also introduces the
practical method of numerical bifurcation analysis by using the numerical analysis
softwares XPPAUT and AUTO.

1.1 Difference Equations, Maps, and Linear Algebra

Natural, social and artificial systems change hour by hour. Dynamical system is
a mathematics for the modeling and the analysis of such systems’ behavior. Dy-
namical systems incorporate the state and its time change in a system. Consider a
difference equation or discrete-time dynamical system:

x.n C 1/ D f .x.n//; x.n/ 2 RN ; n D 0; 1; 2; : : : : (1.1)

x.n/ is an N -dimensional real vector (i.e. x.n/ D (x1 .n/, x2 .n/, . . . , xN .n//T
where ./T denotes a transpose of a vector or a matrix ./) and represents a state
(called a state point or simply a state) of a system at a time n. The N -dimensional
map f transfers the state x.n/ into another state x.n C 1/ D f .x.n//. The state
x.0/ is referred as the initial state. We can denote the n-th state x.n/ using the
initial state:
n times
‚ …„ ƒ
n
x.n/ D f .x.0// D f ı f ı    ı f .x.0//: (1.2)

S. Doi et al., Computational Electrophysiology, 1

DOI 10.1007/978-4-431-53862-2 1,  c Springer 2010
2 1 A Very Short Trip on Dynamical Systems

Iterations of the map f generate a sequence of states (or a set of states):

fx.n/g D x.0/; x.1/; x.2/; : : : (1.3)

which is called an orbit (or trajectory or solution) of the system (1.1). A set of all
states (in our case, RN ) is called a state space (or phase space). If a state x  2 RN
satisfies the relation f .x  / D x  , then the orbit started from x  stays at the point
x  forever:
x.n/ D x.n  1/ D    D x.1/ D x.0/ D x  : (1.4)
The special state x  is called a fixed point or an equilibrium point.
The simplest example of the discrete dynamical system (1.1) is the case that the
map f is a linear matrix A:

x.n C 1/ D Ax.n/; x.n/ 2 RN ; n D 0; 1; 2; : : : ; (1.5)

where A is an invertible N  N matrix. Note that the origin x D .0; : : : ; 0/T is

always a unique fixed point since this system is linear and A is invertible. Then, the
orbit or general solution of this system is obtained as

x.n/ D An x.0/: (1.6)

First, consider a simple case that N D 2 and A is a 2  2 matrix:

   
a b x1 .n/
x.n C 1/ D x.n/; x.n/ D ; n D 0; 1; 2; : : : : (1.7)
c d x2 .n/

We also suppose that b D c D 0. In this case, we can solve (obtain its orbit) easily:
   
n an 0 an x1 .0/
x.n/ D A x.0/ D x.0/ D : (1.8)
0 dn d n x2 .0/

Figure 1.1 shows various orbits for various initial values .x1 .0/; x2 .0//T .
The panel (a) corresponds to the case 0 < a < d < 1 (and b D c D 0). All
orbit move toward the origin .0; 0/T as n ! 1. This result is apparent since
jaj; jd j < 1. (Note that the orbits are discrete points [dots]. The lines with arrow
between dots are not orbits, which are shown for the sake of presentation clarity
in the figure.) Orbits with special initial values on the x or y axes move straightly
toward the origin. General orbits do not move straightly but approach the origin
along the y-axis. This is because jaj < jd j and thus the x-component vanishes
more quickly than the y-component; the y-component is dominant when the time
n is large enough. In this case, the origin is the fixed point of this system and also
is stable since all orbits nearby the origin approach it as time n increases. Another
example is illustrated in panel (b) in which 0 < a < 1 < d and b D c D 0. In this
case, all orbits on the x-axis converge on the origin since jaj < 1. On the other
hand, all orbits on the y-axis diverge since jd j > 1. Other orbits except for these
special orbits diverge upward or downward although they approach the y-axis (the
x-component vanishes). In this case, the origin is an unstable fixed point.
1.1 Difference Equations, Maps, and Linear Algebra 3

a b
5 5
4 4
3 3
2 2
1 1
x(2)

x(2)
0 0
−1 −1
−2 −2
−3 −3
−4 −4
−5 −5
−5 0 5 −5 0 5
x(1) x(1)
c d
5 5
4 4
3 3
2 2
1 1
x(2)

x(2)

0 0
−1 −1
−2 −2
−3 −3
−4 −4
−5 −5
−5 0 5 −5 0 5
x(1) x(1)
e f
5 5
4 4
3 3
2 2
1 1
x(2)
x(2)

0 0
−1 −1
−2 −2
−3 −3
−4 −4
−5 −5
−5 0 5 −5 0 5
x(1) x(1)

Fig. 1.1 Phase plane and orbits of the linear dynamical system (1.7). (a) a D 0:5, b D 0, c D 0,
d D 0:8. (b) a D 0:7, b D 0, c D 0, d D 1:5. (c) a D 0:65, b D 0:15, c D 0:15,
d D 0:65. (d) a D 0:7 cos.=6/, b D 0:7 sin.=6/, c D b, d D a. (e) a D 1 cos.=6/,
b D 1 sin.=6/, c D b, d D a. (f) a D 1 cos.0:51/, b D 1 sin.0:51/, c D b, d D a. The
origin .0; 0/ is the fixed point and specifically called (a) node, (b) saddle, (c) node, (d) focus, (e, f)
center

Next, consider more general case of the matrix A:

 
0:65 0:15
AD : (1.9)
0:15 0:65
4 1 A Very Short Trip on Dynamical Systems

Figure 1.1c shows various orbits for various initial values when A is given by this
matrix. Differently from the panel (a), only orbits with special initial values on the
line y D x or y D x move straightly and other general orbits curvedly approach
the origin along the line y D x. This reason is explained recalling the basic of
linear algebra. A vector x ¤ 0 is called an eigenvector of a matrix A if the following
equation holds
Ax D x; (1.10)
where the number  is called an eigenvalue of A. We can immediately verify that
the eigenvalues of the matrix A of (1.9) are 0.5 and 0.8, and that the corresponding
eigenvectors are .˙1; 1/T . Thus, the orbits with initial values on the lines y D x or
y D x become as follows:

   n    
˛ 0:65 0:15 1 1
x.n/ D An D˛ D ˛  0:5n ; (1.11)
˛ 0:15 0:65 1 1
   n    
˛ 0:65 0:15 1 1
x.n/ D An D˛ D ˛  0:8n : (1.12)
˛ 0:15 0:65 1 1

Therefore, all orbits with initial values on lines y D x or y D x remain on the

lines and approach the origin as n increases.
Let the eigenvalue and the eigenvector of a matrix A be  and v. Then, in general,
the orbit with an initial value which is multiple of v becomes

x.n/ D An .˛v/ D ˛n v: (1.13)

Thus, all orbits whose initial values (vectors) are scalar-multiples of eigenvector
move straightly whereas other orbits move curvedly. (Note that both x- and y-axes
are eigenvectors in panel (a).)
Next, consider another example of a matrix A:
 
r cos  r sin 
AD (1.14)
r sin  r cos 

which has complex eigenvalues r cos  ˙ i r sin  (i is the imaginary unit). In this
case, the orbit is obtained as follows:
 n
r cos  r sin 
x.n/ D A x.0/ D
n
x.0/
r sin  r cos 
 
cos n  sin n
D rn x.0/: (1.15)
sin n cos n

Thus, the orbit moves counter-clockwise at the speed proportional to  and ap-
proaches the origin if 0  r < 1. Figure 1.1d shows various orbits when r D 0:7
and  D =6. All orbits wind counter-clockwise and approach the origin since
1.1 Difference Equations, Maps, and Linear Algebra 5

r < 1. If r > 1 the directions of all orbits will be reversed and all orbits spiral
outward the phase plane. Panels (e) and (f) explain a special case between r < 1
and r > 1: r D 1. In panel (e), we set  D =6, and all orbits started from the x-axis
(the x > 0 range) return to their starting points after twelve iterations of the map:
x.12/ D x.0/. This is apparent because n D 12.=6/ D 2 in (1.15). In general,
an orbit fx.n/g such that x.p/ D x.0/ and x.i / ¤ x.0/ for 1  i < p is called a
periodic orbit with period p. All orbits in panel (e) are periodic orbits with period
12. In panel (f), the value of  is set as  D 0:51 which is slightly smaller than
 D =6  0:53 of panel (e). Although, the orbits in panels (e) and (f) look similar
each other, the orbits of (f) can not reach the x-axis after twelve iterations of the map
since  D 0:51 < =6. Note that those orbits in (f) cannot become periodic since
any integer-multiples of 0.51 (rational number) cannot become multiple of 2 (irra-
tional number). Such orbits in panel (f) are called quasi-periodic orbits. (Compare
with the similar phase plane of Fig. 1.3 in the following continuous-time case.)
Exercise 1.1. The following is an example of MATLAB commands to make the
phase plane of Fig. 1.1d:
%This is comment. Phase plane of 2D difference eqs.
clf; mm=5; axis([-mm, mm, -mm, mm]); axis(’square’);
xlabel(’x(1)’);ylabel(’x(2)’);
hold on
plot([-mm mm], [0 0], ’k’); plot([0 0], [-mm mm], ’k’);
a=0.7*cos(pi/6); b=-0.7*sin(pi/6); c=-b; d=a;
A=[a b; c d];
x0=[3.5;3.5];
for i=1:15; plot(x0(1),x0(2),’.k’); x=A*x0;
quiver(x0(1),x0(2),x(1)-x0(1),x(2)-x0(2),0,’k’); x0=x;
end
x0=[-3.5;3.5];
for i=1:15; plot(x0(1),x0(2),’.k’); x=A*x0;
quiver(x0(1),x0(2),x(1)-x0(1),x(2)-x0(2),0,’k’); x0=x;
end
x0=[-3.5;-3.5];
for i=1:15; plot(x0(1),x0(2),’.k’); x=A*x0;
quiver(x0(1),x0(2),x(1)-x0(1),x(2)-x0(2),0,’k’); x0=x;
end
x0=[3.5;-3.5];
for i=1:15; plot(x0(1),x0(2),’.k’); x=A*x0;
quiver(x0(1),x0(2),x(1)-x0(1),x(2)-x0(2),0,’k’); x0=x;
end
hold off
Execute these commands on MATLAB and confirm that Fig. 1.1d can be obtained.
Modify these commands to make other phase planes of Fig. 1.1.
(The MATLAB command quiver(x,y,u,v) plots vector(s) as arrow(s) with
components (u,v) at the point(s) (x,y). For more explanation, we can use MATLAB
help as: help quiver. The option ’.k’ of plot(x0(1),x0(2),’.k’) means that
it plots with a dot mark and with black color.) 
6 1 A Very Short Trip on Dynamical Systems

1.2 Differential Equations, Vector Fields, and Phase Planes

In this section, we consider the case that time is continuous. (Please compare the fol-
lowing description with that of the previous section.) In this case, dynamical systems
are usually described by differential equations rather than difference equations. The
dynamical system which corresponds to the discrete-time dynamical system (1.1)
becomes
d
x.t/ D f .x.t//; x.t/ 2 RN ; t 2 R: (1.16)
dt
The solution x.t/ which satisfies this differential equation is called an orbit or tra-
jectory of the system (1.16) similarly to the discrete-time system. The state x.0/ is
referred as the initial state again. The special state point x  such that f .x  / D 0 is
called a fixed point, an equilibrium point or a steady state similarly to the discrete-
time case (but note that the difference of the definition between discrete- and
continuous-time cases: f .x  / D x  vs. f .x  / D 0). The right-hand side (r.h.s.)
f .x.t// 2 RN of the differential equation (1.16) is a vector and is called a vector
field. The vector field assigns the vector f .x/ to each point x of the state space RN .
The simplest example of the continuous dynamical system (1.16) is also the case
that the map f is a linear matrix A:

d
x.t/ D Ax.t/; x.t/ 2 RN ; t 2 R; (1.17)
dt
where A is an invertible N  N matrix. Note that only the origin x D .0; : : : ; 0/T is
the fixed point or an equilibrium point since this system is linear and A is invertible.
The general solutions can be obtained by

x.t/ D exp.At/x.0/; (1.18)

where the exponential of a matrix A is formally defined as follows:

1
X
1 1 1 1 k
exp.A/  I C A C A2 C A3 C    D A : (1.19)
1Š 2Š 3Š kŠ
kD0

Note that the actual computation of this exponential for a general matrix A is
not so easy.
First, consider a finite difference approximation of the differential d x.t/=dt in
(1.17):
d x.t C ıt/  x.t/
x.t/  ; (1.20)
dt ıt
where ıt is a small time interval. Then (1.17) becomes

x.t C ıt/  x.t/

D f .x.t// D Ax.t/ (1.21)
ıt
1.2 Differential Equations, Vector Fields, and Phase Planes 7

and thus we obtain

x.t C ıt/ D x.t/ C ıtf .x.t// D x.t/ C ıtAx.t/ D .I C ıtA/x.t/: (1.22)

Namely, the discrete dynamical system (1.5) has appeared again, although the time
step is not unity but ıt here: this is not important. Thus, the orbit of (1.22) is obtained
by (if t=ıt is an integer)

x.t/ D .I C ıtA/t =ıt x.0/: (1.23)

Let us consider some examples. We suppose that N D 2 and A is a 2  2 matrix:

   
d x.t/ D a b x1 .t/
x.t/; x.t/ D ; t 2 R: (1.24)
dt c d x2 .t/

Figure 1.2 shows such examples where we set as a D  0:5, b D 0, c D 0 and

d D  0:2. The curve with dots shows the orbit where ıt D 1 and the broken curve
without dot corresponds to ıt D 0:5. The solid curve without dot shows the theoret-
ical solution:
 
0:5t 0
x.t/ D exp.At/x.0/ D exp x.0/
0 0:2t
 0:5t 
e 0
D x.0/: (1.25)
0 e 0:2t

1
x(2)

Fig. 1.2 Orbits of a finite 0

difference approximation
(1.22) of linear differential −1
equations. a D 0:5, b D 0, −2
c D 0, d D 0:2. The curve
with dots shows the orbit −3
where ıt D 1 and the broken
curve without dot −4
corresponds to ıt D 0:5. The −5
solid curve without dot shows −5 0 5
the theoretical solution x(1)
8 1 A Very Short Trip on Dynamical Systems

Orbits are shown for two initial values: x.0/T D .4:5; 4:5/ and .4:5; 0/. For the
former initial value, we can see that the discretely approximated orbit approaches
the theoretical one as the value of ıt decreases. Note that the approximate orbit with
ıt D 1 is not so bad, although the value of ıt is extremely large and thus the orbit
jumps at big step (see the dot marks). For the latter initial value, two approximate
orbits and the theoretical one overlap completely. We also note that for ıt D 1, the
discrete approximation (1.23) becomes
 t =ıt
1  0:5ıt 0
x.t/ D .I C ıtA/ x.0/ D
t =ıt
x.0/
0 1  0:2ıt
 t
0:5 0
D x.0/: (1.26)
0 0:8

In the original dynamical system (differential equations), since the eigenvalues of

the matrix A are 0:5 and 0:2, and are less than zero, thus the fixed point or equi-
librium point of the differential equation (1.24) is stable (please try to take a limit
of the r.h.s. of (1.25) when t ! 1). In the approximated dynamical system (1.26),
since the eigenvalues of the matrix of the right-hand side (r.h.s.) are 0.5 and 0.8, and
their absolute values are less than unity, thus the fixed point at the origin of the dis-
cretely approximated system is also stable when ıt D 1. However, if the value of ıt
is increased over 4, the fixed point becomes unstable in the discrete approximation
since the absolute value of one eigenvalue .1  0:5ıt/ exceeds unity.
Let us return to the linear differential equations (1.24) apart from the dis-
crete approximations. Figure 1.3 shows various orbits for various initial values
.x1 .0/; x2 .0//T . Panel (a) corresponds to the case a < d < 0 (and b D c D 0).
The (red) heavy curves are the orbits with various initial values and short lines with
arrow show the vector field of the system (1.24); the vectors f .x/ D Ax are shown
by the arrows for each point x 2 R2 . All orbits are tangent to the vector field at
each point of the orbits. This is apparent because d x.t/=dt means a velocity vector.
Also, from the discrete approximation (1.22), we can see that the orbit x.t/ moves
in the direction of the vector field f .x.t// with an infinitesimally small step size ıt.
The integral x.t/ of d x.t/=dt is just the infinite summation of such infinitesimally
small steps.
Note that the eigenvalues of the matrix A in Fig. 1.3a are a D 0:5 and d D 0:2.
All orbits converge on the origin .0; 0/T as t ! 1. This result is apparent since
both eigenvalues are negative. Orbits with special initial values on the x or y axes
(these are corresponding eigenvectors) move straightly toward the origin. General
orbits do not move straightly but curvedly approach the origin along the y-axis.
This is because jaj > jd j (please check that the inequality is opposite in the discrete
case) and thus the x-component vanishes more quickly than the y-component; the
y-component is dominant when the time t is large enough. In this case, the origin
is the fixed point of this system and also is stable. Another example is illustrated
1.2 Differential Equations, Vector Fields, and Phase Planes 9

a b
5 5
4 4
3 3
2 2
1 1
x(2)

x(2)
0 0
−1 −1
−2 −2
−3 −3
−4 −4
−5 −5
−5 0 5 −5 0 5
x(1) x(1)
c d
5 5
4 4
3 3
2 2
1 1
x(2)

x(2)

0 0
−1 −1
−2 −2
−3 −3
−4 −4
−5 −5
−5 0 5 −5 0 5
x(1) x(1)
e
5
4
3
2
1
x(2)

0
−1
−2
−3
−4
−5
−5 0 5
x(1)

Fig. 1.3 Vector fields and orbits of the linear dynamical system (1.24) in a two-dimensional phase
plane. (a) a D 0:5, b D 0, c D 0, d D 0:2. (b) a D 0:3, b D 0, c D 0, d D 0:5.
(c) a D 0:35, b D 0:15, c D 0:15, d D 0:35. (d) a D 0:3, b D 0:5, c D b, d D a.
(e) a D 0:0, b D 2, c D b, d D a. The origin .0; 0/ is the fixed point and specifically called
(a) node, (b) saddle, (c) node, (d) focus, (e) center

in panel (b) in which a < 0 < d and b D c D 0. In this case, all orbits on the
x-axis converge on the origin since a < 0. On the other hand, all orbits on the y-axis
diverge since d > 0. Other orbits except for these special orbits diverge upward or
10 1 A Very Short Trip on Dynamical Systems

downward although they approach the y-axis (x-components vanish). In this case,
the origin is an unstable fixed point.
Next, consider the more general example of the matrix A:
 
0:35 0:15
AD : (1.27)
0:15 0:35

Figure 1.3c shows various orbits for various initial values when A is given by this
matrix. Differently from the panel (a), only orbits with special initial values on the
line y D x or y D x move straightly and other general orbits curvedly approach
the origin along the line y D x. We can easily explain the reason based on lin-
ear algebra. First, note that the eigenvalues of the matrix are 0:5 and 0:2, and
corresponding eigenvectors are .1; 1/T and .1; 1/T , respectively:
    
0:35 0:15 1 1
D 0:5 ;
0:15 0:35 1 1
    
0:35 0:15 1 1
D 0:2 : (1.28)
0:15 0:35 1 1

The orbit from an initial value .1; 1/T , which is the eigenvector, is obtained by

X1 k  k  
t 0:35 0:15 1
x.t/ D exp.At/x.0/ D (1.29)
kŠ 0:15 0:35 1
kD0
X1 k    
t .0:5/k 1 0:5t 1
D De (1.30)
kŠ 1 1
kD0

which means that this orbit moves straightly in the direction of .1; 1/T .
Next, consider another example of a matrix A:
 
˛ ˇ
AD (1.31)
ˇ ˛

which has complex eigenvalues ˛ ˙ iˇ. In this case, the orbit is obtained as follows:
 
˛t cos ˇt  sin ˇt
x.t/ D exp.At/x.0/ D e x.0/: (1.32)
sin ˇt cos ˇt

Thus, when ˇ > 0, the orbit moves counter-clockwise at a speed proportional to

ˇ and approaches the origin if ˛ < 0. Figure 1.3d shows various orbits when
˛ D 0:3 and ˇ D 0:5. All orbits wind counter-clockwise toward the origin.
Panel (e) illustrates a special case: the ˛ D 0 case. All orbits do not approach
the origin but return to the initial state points. Since ˇ D 2, all orbits in (e) are
periodic orbits with period one. In panel (e), orbits fx.t/; 0  t  0:98g are shown.
1.3 Linearization, Stabilities, Coordinate Transformation 11

Exercise 1.2. The following is an example of MATLAB commands to draw the

phase plane of Fig. 1.3d:

%This is comment. Phase plane of 2D differential eqs.

clf; mm=5; axis([-mm, mm, -mm, mm]); axis(’square’);
xlabel(’x(1)’);ylabel(’x(2)’);
hold on
plot([-mm mm], [0 0], ’k’); plot([0 0], [-mm mm], ’k’);
global a b c d;
r=5;X=[-r:r/10:r];Y=[-r:r/10:r];
a=-0.3;b=-0.5; c=-b; d=a;
[XX,YY]=meshgrid(X,Y); F=( a*XX+b*YY );G=( c*XX+d*YY );
quiver(X,Y,F,G,’k’);
t0=0;tf=100;
x0=[3.5;3.5]; [t,x]=ode23(’linFlow’,[t0, tf], x0’);
plot(x(:,1),x(:,2),’-r’);
x0=[-3.5;3.5]; [t,x]=ode23(’linFlow’,[t0, tf], x0’);
plot(x(:,1),x(:,2),’-r’);
x0=[-3.5;-3.5]; [t,x]=ode23(’linFlow’,[t0, tf], x0’);
plot(x(:,1),x(:,2),’-r’);
x0=[3.5;-3.5]; [t,x]=ode23(’linFlow’,[t0, tf], x0’);
plot(x(:,1),x(:,2),’-r’);
hold off

The MATLAB command ode23 numerically solves the differential equations which
are described in the linFlow.m file:
%This is linFlow.m file.
function xdot=linFlow(t,x)
global a b c d
xdot=zeros(2,1);
xdot(1)=a*x(1)+b*x(2);
xdot(2)=c*x(1)+d*x(2);

Execute these commands on MATLAB and confirm that Fig. 1.3d can be obtained.
Modify these commands to make other phase planes of Fig. 1.3. 

1.3 Linearization, Stabilities, Coordinate Transformation

Although we have already explained about the stabilities of fixed points (equilibrium
points), let us summarize about the stabilities here. Consider the discrete-time linear
dynamical system (1.5). In the system, the origin 0 is the fixed point and is stable
(i.e. x.n/ ! 0, n ! 1) if absolute values of all eigenvalues of A are less than
unity. In the case of the continuous-time dynamical system (1.17), the equilibrium
point at the origin is stable if the real parts of all eigenvalues of A are less than zero
(negative).
12 1 A Very Short Trip on Dynamical Systems

1.3.1 Linearization and Stabilities

Next, consider nonlinear dynamical systems. Let x  be an equilibrium point

of the discrete-time dynamical system (1.1) or the continuous-time dynamical
system (1.16): f .x  / D x  in (1.1) or f .x  / D 0 in (1.16). The Taylor expansion
of the function f .x/ near the equilibrium point x  are obtained as follows:

f .x/ D f .x  / C Df .x  /.x  x  / C O.jjx  x  jj2 /; (1.33)

where O.jjxx  jj2 / denotes the higher-order terms (second-order terms and higher
terms) and Df .x  / is the Jacobian matrix:
0 1
@f1 .x/ @f1 .x/

B @x1 @x2 C
B C
B C
Df .x  / D B @f2 .x/ @f2 .x/    C ; (1.34)
B @x C
@ 1 @x2 A
:: ::
: :    x Dx 
x D .x1 ; x2 ;    /T ; f .x/ D .f1 .x1 ; x2 ;    /; f2 .x1 ; x2 ;    /;    /T :

Near the equilibrium point x  , we can neglect (under some conditions) the higher-
order terms O.jjx  x  jj2 / since O.jjx  x  jj2 / becomes small when jjx  x  jj is
small. Then, we can obtain a linearized system or linearization of (1.1) and (1.16),
respectively as follows:

z.n C 1/ D Az.n/; A D Df .x  /; (1.35)

d
z.t/ D Az.t/; A D Df .x  /; (1.36)
dt

where we have made use of the change of a variable z.n/ D x.n/  x  (or z.t/ D
x.t/  x  ).

1.3.2 Coordinate Transformations

In dynamical systems, a coordinate transformation or a change of variables is very

useful and often simplify our view about the dynamical systems. In this subsection,
we briefly explain such coordinate transformations.
Consider the linear discrete-time system (1.7). If we change the state variable
x to z by x D T z where T is an invertible (or regular) N  N matrix, the linear
discrete-time dynamical system (1.5) becomes

z.n C 1/ D T 1 x.n C 1/ D T 1 Ax.n/ D .T 1 AT /z.n/: (1.37)

1.4 Nonlinear Dynamical Systems and Bifurcations 13

Thus, if the matrix .T 1 AT / has a simple form such as diagonal matrices, the
coordinate transformation makes the analysis of the original system much simpler.
For the linear continuous-time dynamical system (1.17), the transformed system can
be similarly obtained as

d d
z.t/ D T 1 x.t/ D T 1 Ax.t/ D .T 1 AT /z.t/: (1.38)
dt dt
In the case of nonlinear dynamical systems (the discrete-time system (1.1) or
continuous-time system (1.16)), we can treat the coordinate transformation simi-
larly. Let us consider the case that the coordinate transformation is also nonlinear.
Let x D g.z/ be such a nonlinear transformation with an inverse g 1 , where x
denotes x.n/ or x.t/ (similarly z denotes z.n/ or z.t/). Then the transformations of
(1.1) and (1.16) are respectively obtained as

z.n C 1/ D g 1 .x.n C 1// D .g 1 ı f /.x.n// D .g 1 ı f ı g/.z.n// (1.39)

and
 
d d
z.t/ D g 1 x.t/ D .g 1 ı f /.x.t// D .g 1 ı f ı g/.z.t//; (1.40)
dt dt

where the symbol “ı” denotes a composition of maps.

1.4 Nonlinear Dynamical Systems and Bifurcations

Most examples of (discrete- or continuous-time) dynamical systems treated in pre-

vious sections are linear. In this section, we proceed to more general dynamical
systems: nonlinear dynamical systems, and discuss the bifurcation of a certain dy-
namical system. Roughly speaking, the word bifurcation means the change of the
number and/or stability of equilibrium points or periodic orbits.

1.4.1 Bifurcations of Discrete Dynamical Systems

Lift, Degree and Rotation Number

Let us consider the simplest case (N D 1 case) of the discrete-time dynamical

system (mapping) (1.1); consider the following one-dimensional map or 1D map:

x.n C 1/ D f .x.n//; x.n/ 2 S .n D 0; 1; 2; : : :/; (1.41)

14 1 A Very Short Trip on Dynamical Systems

where S is the unit circle (i.e. an interval Œ0; 1 identified unity with zero) and the
map f .x/ is the function from S to S with a certain smoothness (differentiability).
Such a map is often called a circle map.
We say that x  is a periodic point of period n if f n .x  / D x  and f i .x  / ¤ x 
for 1  i < n. The set ff i .x  /; 0  i < ng of periodic points is the periodic
orbit. The periodic point x  with unit period (f .x  / D x  ) is called the fixed point
particularly. The fixed point x  is asymptotically stable if there exists ı and for
any x 2 .x   ı; x  C ı/, f n .x/ ! x  as n ! 1. The asymptotic stability
of a periodic point or periodic orbit can be defined similarly. If jf 0 .x  /j < 1 (the
prime means d=dx), the fixed point x  is asymptotically stable while it is unstable if
jf 0 .x  /j > 1. When jf 0 .x  /j D 1, the stability is neutral and is not determined by
the coefficient jf 0 .x  /j of a linear term, and nonlinear terms or higher order terms
of the Taylor expansion of f .x/ are necessary for the determination of the stability.
Similarly, if the following relation:
ˇ ˇ
ˇ d n  ˇ ˇ 0  n1  ˇ ˇ 0  n2  ˇ ˇ 0  ˇ ˇ  ˇ
ˇ f .x /ˇ D ˇf f .x / ˇ  ˇf f .x / ˇ    ˇf f .x  / ˇ  ˇf 0 x  ˇ < 1
ˇ dx ˇ
(1.42)

holds for a periodic point x  , the periodic point x  and periodic orbit fx  , f .x  /,
. . . , f n1 .x  /g are asymptotically stable. When df n .x  /=dx D 0, the periodic
orbit is the most stable and is called super stable especially.
Next, we suppose that f W S ! S is continuous. A continuous map F W R ! R
is a lift of f if and only if:

F .x/ mod 1 D f .x mod 1/; x2R (1.43)

where “mod” means a modulus. Note that there are countably infinite lifts F for each
f . There exists an integer d such that F .x C 1/ D F .x/ C d for all real numbers x.
This number d is called the degree of F or of f . If F is a non-decreasing function,
the limit
F n .x/
 D lim (1.44)
n!1 n
can be well defined and is called the rotation number of F . Then, the number ( mod
1) is also called the rotation number of f . The rotation number  is very useful to
characterize the dynamics of one-dimensional maps. For example, f has a periodic
orbit if and only if  is a rational number. Also, if F is non-decreasing and  D p=q
with p and q being relatively prime integers, the periods of all periodic orbit of f
become q.
Figure 1.4 shows examples of linear maps on S:

x.n C 1/ D f .x.n//  x.n/ C a mod 1; x.n/ 2 S .n D 0; 1; 2; : : :/ (1.45)

which is a simple shift of x.n/ by a in a sense of modulo unity. Panel (a) is the
case of a D 0:4. Thick lines denote the graph of f .x/ and thin lines show the orbit
1.4 Nonlinear Dynamical Systems and Bifurcations 15

a b c
1 F(x) 1
3

2
xn+1 xn+1
1

–3 –2 –1 0 1 2 3 x

–1

–2
0
x0 x3 x1 x4 x2 1 0 1
xn xn

Fig. 1.4 Examples of the linear map (1.45) and its orbit. (a) Periodic case, a D 0:4. Thick lines
are the graph of map and thin lines show an example of a periodic orbit with period five. The
diagonal thin line is the line x.n C 1/ D x.n/ using which orbits of a map can be graphically
obtained. (b) Lift of (a). The map of (a) defined on S is extended to the one on the whole real axis
R. (c) Quasi-periodic case, a D =10. Note that the notation xn is used instead of x.n/, in this
figure

of the map (1.45): x.0/ D 0:1, x.1/ D 0:5, x.2/ D 0:9, x.3/ D 0:3, x.4/ D
0:7, x.5/ D 0:1 D x.0/ which is a periodic orbit with period five. Diagonal line
x.n C 1/ D x.n/ is also shown. It can be easily shown that all initial values x 2 S
have a period five. These periodic orbits or points are neither asymptotically stable
nor unstable since the derivative f 0 .x/  1. Panel (b) shows the continuous lift
F .x/ of f .x/ in panel (a), from which we can see that the map of (a) is of degree 1
and has a rotation number 0:4 D 2=5 since

F n .x/ x C n  0:4 2
lim D lim D 0:4 D :
n!1 n n!1 n 5

This result agrees with the result that the periods of all periodic orbits of (a) are five.
Panel (c) is the case of a D =10 whose rotation number is an irrational number
=10 and thus the map of (c) does not possess any periodic orbits: any orbits show
quasi-periodic behavior and never return to the previously visited points. Note, how-
ever, that digital computers can not realize an irrational number in a strict sense and
thus computer generated orbits may show periodic behavior with a long period.

Bifurcations of the Standard Circle Map

Next let us examine how the solution structure of a certain map is changed with a
continuous change of a parameter of the map, i.e. we study the bifurcation structure
of a typical circle map called the standard circle or sine-circle map:

x.n C 1/ D f .x.n//  x.n/ C a C b sin.2x.n// mod 1

x.n/ 2 S .n D 0; 1; 2; : : :/: (1.46)
16 1 A Very Short Trip on Dynamical Systems

As stated above, the stability of a fixed point x  changes at the critical value of
jf 0 .x  /j D 1. Thus, roughly speaking, a certain bifurcation (a change of qualitative
nature) may occur at a fixed point when jf 0 .x  /j D 1. There are only two cases
f 0 .x  / D 1 and f 0 .x  / D 1 corresponding to such a bifurcation since we are
considering one-dimensional dynamical systems.
Figure 1.5 shows examples of the sine-circle map (1.46) and its orbit. The pa-
rameter a is fixed to 0.1 and the value of b is slightly changed. The thick curve
is the graph of the function f .x/ which is continuous on S and has degree 1 for
all values of b shown in Fig. 1.5. Thin lines show an orbit with an initial value 0.1
and a diagonal line x.n C 1/ D x.n/ is also shown. Panel (a) is the case of small
b value in which there is neither fixed point nor stable periodic orbit and the map
shows a so-called quasi-periodic behavior. If b is slightly increased (b D 0:1), the
graph of f .x/ is tangent to the diagonal line at x D 3=4 where a saddle-node or
fold bifurcation occurs. After the fold bifurcation occurrence, a pair of fixed points
(stable one xN and unstable one x) O are generated (see Fig. 1.5c). If the value of b is
further increased (b  0:337), the slope at the stable fixed point xN becomes 1 and
a period-doubling or flip bifurcation occurs. After this bifurcation, the fixed point
xN becomes unstable and a stable periodic orbit with period two (the thick square of
Fig. 1.5e) appears (i.e. the unit period of the fixed point is doubled to two), and any
orbits asymptotically approach this periodic orbit.

a b 1
c 1
1

x^
xn+1 xn+1 xn+1
_
x

0 xn 1 0 xn 1 0 xn 1

d 1
e 1
f 1

xn+1 xn+1 xn+1

0 xn 1 0 xn 1 0 xn 1

Fig. 1.5 Examples of the sin-circle map (1.46). a D 0:1. (a) b D 0:08. (b) b D 0:1. (c) b D 0:12.
(d) b D 0:31. (e) b D 0:35. (f) b D 0:46. Note that the notation xn is used instead of x.n/, in this
figure
1.4 Nonlinear Dynamical Systems and Bifurcations 17

a b
1.00 1.0

0.90
PD3 FD
0.80 FD PD2
0.70 {Xn}
0.60 PD3
PD3
PD1
0.50 PD2
PD1 PD2
0.40
PD3
0.30 0.0
0.00 0.10 0.20 0.30 0.40 0.50 0.0 0.7
b b

Fig. 1.6 One-parameter bifurcation diagram of the sine-circle map for the bifurcation parameter
b when a D 0:1. (a) Bifurcation diagram obtained by AUTO (Doedel et al. 1995). (b) Bifurcation
diagram obtained by brute-force computer simulations of (1.46): Asymptotic sequences fx.n/g,
(n D 201, . . . ,700) produced by (1.46) were plotted for each of 1,000 equally spaced b values on
the interval Œ0; 0:7

These results are summarized in the bifurcation diagram of Fig. 1.6a where the
abscissa (the horizontal axis) is the bifurcation parameter b and the ordinate (the
vertical axis) shows the x values of fixed points or periodic orbits. Namely, the bi-
furcation diagram shows how the (stationary or asymptotic) solutions of the system
(1.46) change as the parameter b varies. The solid curves show the stable fixed
points or periodic points and the broken curves unstable ones. At the fold bifurca-
tion point FD (b D 0:1) of Fig. 1.6a, a pair of stable and unstable fixed points are
simultaneously generated. This stable fixed point undergoes a period-doubling bi-
furcation at the point PD1 (b  0:337) after which the fixed point becomes unstable
and a stable periodic orbit with period two is generated. This periodic orbit under-
goes a subsequent period-doubling bifurcation at PD2 (b  0:427) and a further
period-doubled periodic orbit is generated. A further period-doubling bifurcation
occurs at PD3 although a period-eight orbit is not shown in the figure. In fact, such
a cascade of period-doubling bifurcations continues infinitely and it is very difficult
to trace out all such bifurcations. Figure 1.6a is computed using the bifurcation anal-
ysis software called AUTO (Doedel et al. 1995) which is very useful to analyze the
nonlinear dynamical systems described by difference or differential equations.
Figure 1.6b is also a bifurcation diagram calculated by the computer simula-
tion of (1.46): Asymptotic sequences fx.n/g, (n D 201, . . . ,700) produced by (1.46)
were plotted for each of 1,000 equally spaced b values on the interval Œ0; 0:7. In
the range of b < 0:1, (1.46) shows a quasi-periodic behavior (many dots are plot-
ted). A cascade of period-doubling bifurcations explained in (a) is also seen. In the
right of the period-doubling cascade, chaotic solutions appear (deep-black region).
Figure 1.5f shows an example of such a chaotic solution. Although there exist many
deep-black regions in Fig. 1.6b, only black regions in b > 0:1 corresponds to chaotic
solutions. Roughly speaking, as the portion of f .x/ with jf 0 .x/j > 1 is increased,
the map f .x/ tends to create chaotic solutions.
18 1 A Very Short Trip on Dynamical Systems

Strictly speaking, in the calculation of Fig. 1.6b, no bifurcation analysis is used:

Only “brute-force simulations” of (1.46) are examined whereas several bifurca-
tion conditions such as jf 0 .x/j D 1 are numerically solved in the computation
of Fig. 1.6a obtained by AUTO. A bifurcation diagram obtained by brute-force sim-
ulations has an advantage of easy calculation but a disadvantage in that only stable
solutions can be obtained; the broken curves shown in Fig. 1.6a does not appear in
Fig. 1.6b. Although unstable solutions of a system cannot be observed in real ex-
periments or in numerical simulations, they sometimes offer missing links between
stable solutions, and therefore, are very helpful for the understanding of total behav-
ior of a system.
Exercise 1.3. Compute numerically the bifurcation diagram for the logistic map
f .x/ D ax.1  x/ with a as a bifurcation parameter. Next, compute the fixed points
of the map analytically (as a function of the parameter a). What are the stabilities
of these fixed points? Also, compute the periodic orbits with period two analytically
(A nonlinear coordinate transformation (variable change) from x to z such that z D
a.1  2x/=2 may be useful for such analytical calculations). 

1.4.2 Bifurcations of Continuous Dynamical Systems

Saddle-Node Bifurcation

Consider a (nonlinear) one-dimensional differential equation:

xP D f .x/; f .x/ D x 2 C ; x;  2 R; (1.47)

where xP denotes the time derivative: xP D dx=dt, and  is a parameter. Figure 1.7
shows the graph of f .x/ as a function of x for various values of . Panel (c) cor-
responds to the case  D 1. In this case, the graph of the function f .x/ intersects
with the x-axis at the two points: x D ˙1. These points are the equilibrium points

a 2
b 2
c 2

1.5 1.5 1.5

1 1 1

0.5 0.5 0.5

y = f (x)

y = f (x)

y = f (x)

0 0 0

−0.5 −0.5 −0.5

−1 −1 −1

−1.5 −1.5 −1.5

−2 −2 −2
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2
x x x

Fig. 1.7 Vector fields and orbits of the one-dimensional nonlinear dynamical system (1.47):
xP D  x 2 C . (a)  D 1. (b)  D 0. (c)  D 1
1.4 Nonlinear Dynamical Systems and Bifurcations 19

of the system (1.47) since fD1 .˙1/ D 0. In the figure, the x-axis corresponds to
the one-dimensional state space and the equilibrium points are shown by big dots.
The arrows show the direction of the one-dimensional vector field or orbits on the
x-axis. For example, in the range x > 1 of Fig. 1.7c, the direction is left since
f .x/ < 0. In the range 1 < x < 1, the direction becomes right, and thus all
orbits started from the range 1 < x move toward the equilibrium point at x D 1.
Therefore, x D 1 is a stable equilibrium point. On the other hand, x D 1 is a
unstable equilibrium point, and all orbits started from x < 1 diverge leftward.
In Fig. 1.7b, the value of  is decreased to zero. Thus, the graph of the function
f .x/ is tangent to the x-axis at the unique point: x D 0. There is only one equi-
librium point. All orbits started in the range x > 0 approach x D 0, whereas all
orbits started from the range x < 0 move leftward and finally diverge. If the value
of  is further decreased, the graph of f .x/ never cross the x-axis: there is no
equilibrium point. All orbits move leftward and diverge. As the value of the param-
eter  changes, the behavior of the dynamical system (1.47) changes, particularly
the number and/or the stability of equilibrium points change. This is a bifurcation
similarly to the discrete-time systems.  D 0 is the bifurcation point and this bifur-
cation is called a saddle-node (SN) or a fold bifurcation again (compare this with
the saddle-node bifurcation in the discrete-time systems).
Figure 1.8 is the bifurcation diagram of the dynamical system (1.47). The
abscissa denotes the bifurcation parameter  and the ordinate the x-value; the equi-
librium points of the system (1.47) are plotted as a function of . The solid and
broken curves denote the stable and unstable equilibria, respectively. At the point
labeled SN, a saddle-node bifurcation occurs and a pair of stable and unstable equi-
libria are generated.
Next, we consider the following two-dimensional system:
   
xP 1 x1 2 C 
D : (1.48)
xP 2 x2
p
It is apparent that this system has two equilibrium points .˙ ; 0/ if  > 0.
Figure 1.9 shows the state plane (phase plane) of this system. When  < 0 (panel a),

2.0
1.5
1.0
0.5
SN
x 0.0
–0.5
–1.0

Fig. 1.8 Bifurcation diagram –1.5

of the one-dimensional –2.0
nonlinear dynamical system –1.0 –0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0
(1.47): xP D x 2 C  λ
20 1 A Very Short Trip on Dynamical Systems

a b c
2 2
2

1 1
1

y
0 0
y

−1 −1
−1

−2 −2
−2
−2 −1 0 1 2 −2 −1 0 1 2 −2 −1 0 1 2
x x x

Fig. 1.9 Vector fields and orbits of the two-dimensional system (1.48). (a)  D 1. (b)  D 0.
(c)  D 1

all orbits diverge leftward approachingpthe horizontal axis. When  > 0 (panel c),
p with an initial value x1 .0/ > p
orbits   converge on the right equilibrium point
.C ; 0/ and orbits with x1 .0/ <   diverge leftward approaching the horizon-
tal axis. The right equilibrium
p point is stable and is called a node, whereas the left
equilibrium point . ; 0/ is called a saddle. A saddle is a point that it is stable in a
one direction (vertical, in this case) and is unstable in another direction (horizontal).
When  D 0 (panel b), both saddle and node collide. Therefore, this bifurcation is
called a saddle-node bifurcation. The bifurcation diagram is completely the same
as that of the one-dimensional system (1.47). This is apparent since the variable x2
does not affect the dynamics of x1 in the two-dimensional system (1.48). Although
the example (1.48) looks very artificial, the essential dynamics of general saddle-
node bifurcations is one-dimensional similarly to (1.48).

Hopf Bifurcation

Consider the following two-dimensional nonlinear dynamical system:

      
xP 1  1 x1 x1 .x1 2 C x2 2 /
D  : (1.49)
xP 2 1  x2 x2 .x1 2 C x2 2 /

This system has an equilibrium at the origin .x1 ; x2 / D .0; 0/. Consider the system
near this equilibrium point: .x1 ; x2 /  .0; 0/. Because jx1 j  jx1  x1 2 j and so on
near the origin, we can neglect the higher-order terms (the second term of the r.h.s.
of (1.49)) and obtain
      
xP 1  1 x1 x
D A 1 (1.50)
xP 2 1  x2 x2

which is a linearized system of (1.49) at the origin. Clearly the matrix A has
eigenvalues  ˙ i . The stability of the equilibrium .0; 0/ of the nonlinear
system (1.49) is determined by the linearized system (1.50). Thus, the equilibrium
1.4 Nonlinear Dynamical Systems and Bifurcations 21

a b c
1.5 1.5 1.5

1 1 1

0.5 0.5 0.5

0 0 0
y

y
−0.5 −0.5 −0.5

−1 −1 −1

−1.5 −1.5 −1.5

−1.5 −1 −0.5 0 0.5 1 1.5 −1.5 −1 −0.5 0 0.5 1 1.5 −1.5 −1 −0.5 0 0.5 1 1.5
x x x

Fig. 1.10 Vector fields and orbits of the two-dimensional dynamical system (1.49) which illus-
trates the super-critical Hopf bifurcation. (a)  D 0:5. (b)  D 0. (c)  D 0:5

point is stable or unstable if  < 0 or  > 0, respectively. When  D 0, the stability

cannot be determined by the linearized system but it is determined by the nonlinear
(higher) terms of the system (1.49). Later explanation will show that the equilibrium
point is still stable even when  D 0.
Figure 1.10 shows the vector field and orbits of the nonlinear system (1.49) when
(a)  D 0:5, (b)  D 0, (c)  D 0:5. Orbits in panel (a) move spirally toward the
origin; we can see that the origin is a stable equilibrium point. On the other hand,
when  D 0:5 (see panel c), the origin becomes unstable. Orbits near p the origin
leave the origin and move toward a certain circle with a radius of 1= 2. Orbits far
from the origin also move toward the circle. This circle itself is a special orbit (pe-
riodic orbit) called a limit cycle. Panel (b) illustrates a special case between panels
(a) and (c):  D 0. In the figure, we can see a circle, but it is not really a limit cycle.
The origin is still stable although the stability is weak. All orbits in panel (b) will
finally approach the origin if we prolong the simulation time.
To summarize, the equilibrium point at the origin changes its stability as the value
of  changes its sign from negative to positive. When the equilibrium becomes un-
stable, a stable limit cycle appears. This is called a Hopf bifurcation. Figure 1.11
is the bifurcation diagram which summarizes the dependency of the dynamics of
the system (1.49) on the parameter . The abscissa shows the bifurcation parame-
ter  and the ordinate the variable x1 of the system (1.49). The solid and broken
curves (lines) show the x1 component of the stable and unstable equilibrium points
as a function of , respectively. The filled circles show the maximum values of x1
of stable periodic orbits. At the point labeled by HB, the Hopf bifurcation occurs
and the stability of the equilibrium point is changed and the stable periodic orbit is
generated (bifurcated).
Consider a similar system to (1.49):
      
xP 1  1 x1 x .x 2 C x2 2 /
D C 1 12 ; (1.51)
xP 2 1  x2 x2 .x1 C x2 2 /
22 1 A Very Short Trip on Dynamical Systems

Fig. 1.11 Bifurcation 3.0

diagram of the 2.5
two-dimensional dynamical 2.0
system (1.49) which
1.5
illustrates the Hopf x1
1.0
bifurcation
0.5
0.0

–0.5 HB

–1.0
–5. –4. –3. –2. –1. 0. 1. 2. 3. 4. 5.
λ

a b c
1.5 1.5 1.5

1 1 1

0.5 0.5 0.5

0 0 0
y

y
−0.5 −0.5 −0.5

−1 −1 −1

−1.5 −1.5 −1.5

−1.5 −1 −0.5 0 0.5 1 1.5 −1.5 −1 −0.5 0 0.5 1 1.5 −1.5 −1 −0.5 0 0.5 1 1.5
x x x

Fig. 1.12 Vector fields and orbits of the two-dimensional dynamical system (1.51) which illus-
trates the sub-critical Hopf bifurcation. (a)  D 0:5. (b)  D 0. (c)  D 0:5

where only the sign of the nonlinear term is different from (1.49). Figure 1.12 shows
the vector field and orbits of this system (1.51) when (a)  D 0:5, (b)  D 0,
(c)  D 0:5. First, note that the stability of the equilibrium point .0; 0/ in both
cases (a) and (c) is the same as that of Fig. 1.10 since the stability is determined by
the linear terms. The stability in (b), however, differs between Figs. 1.10 and 1.12
since the nonlinear terms are different. Figure 1.12c is very similar to Fig. 1.10a.
Note that, however, the direction of orbits is different (the stability of the equilibrium
is different), and the orbits move outward and diverge. In Fig. 1.12a, the equilibrium
point is stable and thus orbits near the origin approach the equilibrium. The orbits
far from the origin move outward and diverge finally. There is a boundarypbetween
the inward and outward orbits. This boundary is a circle with a radius 2 and a
special orbit called a limit cycle also. This limit cycle, differently from Fig. 1.10c,
is unstable and nearby orbits do not approach it. When  D 0, the stability of the
equilibrium is not determined by the linear terms but by nonlinear terms, and it is
unstable as is clarified later. Thus all orbits nearby the origin moves outward and
finally diverge, although the speed of divergence is very slow (compare panels b
and c in Fig. 1.12).
Figure 1.13 is the bifurcation diagram of (1.51). The open circles show unstable
periodic orbits. The stability of the equilibrium points of Fig. 1.13 is the same as
those of Fig. 1.11 (except for the  D 0 case). The way of generation of periodic
1.4 Nonlinear Dynamical Systems and Bifurcations 23

Fig. 1.13 Bifurcation 3.0

diagram of the 2.5
two-dimensional dynamical
2.0
system (1.51) which
illustrates the sub-critical 1.5
Hopf bifurcation x
1 1.0

0.5

0.0

–0.5 HB

–1.0
–5. –4. –3. –2. –1. 0. 1. 2. 3. 4. 5.
λ

orbits, however, is different. Namely, in Fig. 1.13, the periodic orbits are unstable
and the region of  where the periodic orbits exist is the range below the bifurcation
point  D 0, whereas the region in Fig. 1.11 is the range above the Hopf bifurcation.
Therefore, the Hopf bifurcations in Figs. 1.11 and 1.13 are called super- and sub-
critical Hopf bifurcations, respectively. (Note that the words super and sub do not
really denote the direction of the branch of periodic orbits, but do denote the stability
of periodic orbits; there exists a sub-critical Hopf bifurcation that unstable periodic
orbits exist above the sub-critical Hopf bifurcation point. For example, if we change
 by  in (1.51), then the bifurcation diagram becomes the similar one to Fig. 1.13
where the right and left are reversed.)
In order to understand the dynamics of the system (1.49), it is very useful to
use a coordinate transformation, particularly to use a complex variable. Consider a
coordinate transformation from the two-dimensional variable .x1 ; x2 / to the one-
dimensional complex variable z: z D x1 C ix2 . Of course, we denote the complex
conjugate as zN D x1  ix2 and the absolute value as jzj2 D x1 2 C x2 2 . Then, we
obtain

zP D xP 1 C i xP 2
D x1  x2  x1 .x1 2 C x2 2 / C i.x1 C x2 /  ix2 .x1 2 C x2 2 /
D .x1 C ix2 / C i.x1 C ix2 /  .x1 C ix2 /.x1 2 C x2 2 /
D z C i z  zjzj2 : (1.52)

Furthermore, using the polar coordinate z D e i  , we have

P i  C ie i  P D e i  C ie i   e i  2 :

zP D e (1.53)

Comparing both sides of the second equality of this equation, we obtain

P D   3 D .  2 /;
P D 1: (1.54)
p
p that P D 0 for  D  if  0 (Note that  is the amplitude
Thus, we can see
and thus  D   is not allowed). Since the .; / coordinate system is the polar
24 1 A Very Short Trip on Dynamical Systems
p
coordinate, the solution .t/ D p corresponds to a periodic orbit of the original
system (1.49) with an amplitude  and with a unit angular velocity ( D 0 cor-
responds to the equilibrium point at the origin). From the first equation of (1.54),
we can see that the equilibrium point at the origin ( D 0) is stable when  p < 0
since P D    3   < 0 if  p D  is small.pAlso, the periodic orbit (pD )
is stable when  > 0 since P D .  ˙ /  .  ˙ /3 
2 if  D  ˙ .
When  D 0, the first equation of (1.54) becomes P D 3 and thus the equilibrium
point at the origin ( D 0) is (nonlinearly) stable. These results are consistent with
the above explanations on the super-critical Hopf bifurcation (Fig. 1.11).
Similarly, we obtain the polar-coordinate expression for the system (1.51):

P D  C 3 D . C 2 /;
P D 1: (1.55)

Only the difference from (1.54) is the sign of the nonlinear (cubic) term in thepfirst
equation. Thus, this system has an unstable periodic orbit with an amplitude 
when  < 0. Also, when  D 0, the equilibrium point at the origin ( D 0) is
(nonlinearly) unstable since the first equation of (1.55) becomes P D C3 . See the
bifurcation diagram of Fig. 1.13.

Exercise 1.4. Use the polar coordinate transformation x1 D  cos , x2 D  sin 

for (1.49) to directly obtain the same result as (1.54). 

1.4.3 Forced Dynamical Systems and Poincaré Maps

One-dimensional map of (1.41) is seemingly too simple to model and analyze real
systems. The one-dimensional map, however, arises commonly from general n-
dimensional dynamical systems. Consider the following “forced” dynamical system
(differential equations):

dx
D f .x/ C g.t/; x 2 Rn ; f W Rn ! Rn ; g W R ! Rn ; (1.56)
dt
where g denotes an external input (a forcing term) applied to the system and is a
time-periodic function with period T : g.t C T / D g.t/. Suppose that we make
a sampling (or an observation) of the state variable x.t/ of (1.56) at every time
interval T , and let us denote the state x.nT / at the n-th observation by x n . Also,
let us denote the solution of (1.56) with an initial state x.0/ D x 0 after a time t
as ˆ.x 0 ; t/. Then, the relation between x n and x nC1 is denoted by the following
discrete-time dynamical system:

x nC1 D F .x n /  ˆ.x n ; T /; F W Rn ! Rn .n D 0; 1; 2; : : :/: (1.57)

1.4 Nonlinear Dynamical Systems and Bifurcations 25

Usually, the dynamical system (1.56) without the external forcing (g.t/  0) has an
attractor (asymptotically stable set) and the state point x.t/ asymptotically moves
near the attractor. Even in the case that the forcing term g.t/ is present, if g.t/
is small enough relatively to the stability of the attractor or g.t/ is impulsive like
a neuronal action potential, the state point x.t/ can be assumed to move near the
attractor. Then the map F is (approximately) a map from the attractor to itself. If
the dynamical system (1.56) with g.t/  0 shows an oscillatory or periodic be-
havior, the attractor is a one-dimensional closed orbit called a limit cycle. Then the
n-dimensional map F can be reduced to a one-dimensional map on the limit cycle.
(Even if the system is not oscillatory but excitable like a neuron, F can become
one-dimensional. See Doi and Sato 1995.) By a suitable choice of a coordinate
(usually called a phase) on the limit cycle, the map becomes the one-dimensional
circle map (1.41).

1.4.4 Attractors and Basins of Attraction

So far, we have seen that dynamical systems can have various stable orbits such as
stable equilibrium points, stable limit cycles. Such a stable set or a geometric object
which consists of state points is called an attractor, since the set attracts nearby
orbits. Dynamical systems can have multiple attractors: two equilibrium points, an
equilibrium point and a limit cycle, and so on. Then, the final fate of an orbit depends
on its initial state. (Note that linear [discrete- or continuous-time] dynamical system
have only one equilibrium or fixed point, thus only nonlinear dynamical systems
can have multiple attractors.)
Consider a ball (considered to be a point particle) moving under the influence of
friction in a one-dimensional potential or a slope V .x/ as depicted in the upper panel
of Fig. 1.14 (Mcdonald et al. 1985). There are two points x1 and x2 where the po-
tential takes its minimal value. Thus, the ball eventually settles down to one of two
points. Since the movement of the ball is determined by its initial position x and ve-
locity dx=dt, the state space (phase plane) is two-dimensional as shown in the lower
panel. There are two stable equilibrium points denoted by filled circles in the phase
plane. There is also an unstable equilibrium point denoted by an open circle, which
corresponds to the potential peak in the center of the upper panel. If an initial state is
taken from the white region depicted in the x–dx=dt phase plane, the orbit starting
from the initial state eventually converges to the left equilibrium point x2 while the
orbits from shaded (yellow) region come to rest at the right equilibrium x1 . Each
region (a set of state points) which finally leads to an attractor (a stable equilibrium
point in this case) is called a basin of attraction or a basin simply. These two basins
are interweaved with each other, and thus a small change in the ball’s position and/or
velocity leads a big difference in its final fate. The basins in this simple example are
not actually so interweaved. In fact, the shape of basins can become more compli-
cated like fractal (Mcdonald et al. 1985). Anyway, we should pay much attention
26 1 A Very Short Trip on Dynamical Systems

Fig. 1.14 A simple ball V(x)

moving in a one-dimensional
potential illustrating a
complicated shape of a basin
of attraction. This is modified
x2* x1*
from Mcdonald et al. (1985)
x

dx/dt

Fig. 1.15 Example of multiple attractors and basins often encountered in neuronal models

on the shape of basins particularly in numerical simulations, since arbitrary choice

of an initial value leads to different simulation results.
Figure 1.15 schematically illustrates typical examples of multiple attractors and
basins often encountered in neuronal models. There are two stable equilibrium
points (filled circles) and one unstable equilibrium (open circle) in both left and
right phase planes. The dynamics, however, are much different in the two phase
planes. The center unstable equilibrium is a saddle; there are two special orbits
(thick solid curves with light-blue color) converge to the unstable equilibrium point
and other two special orbits (thick dotted curves with red color) converge to the
point “when time is reversed.” Such special orbits of thick solid (light-blue) and
thick dotted (red) are called stable and unstable manifold, respectively. Although
these orbits are special in that we should carefully choose an initial state in order to
find those orbits in numerical simulations, they play a very important role in order to
clarify the dynamics of dynamical systems. The shaded (yellow) region is the basin
of attraction of the right stable equilibrium in both phase planes. In the left figure,
the basin is separated by the stable manifolds of the saddle. (Note that orbits started
outside the basin approach the “left” equilibrium even if their initial states are closer
1.5 Computational Bifurcation Analysis 27

to the “right” equilibrium.) On the other hand, in the right phase plane of Fig. 1.15,
the basin of the right equilibrium is small and is separated by a closed curve (broken
curve). The closed circle is an unstable periodic orbit or limit cycle. In such a phase
plane, there occurs an interesting phenomenon; orbits started closely the unstable
limit cycle (but outside the limit cycle) will wind the limit cycle many times and
finally converge to the left stable equilibrium point.
In summary, the clarification of the geometric structure of such special (unstable)
orbits as stable and unstable manifolds, unstable limit cycles, etc., is very important
for the understanding of the dynamics of systems.

1.5 Computational Bifurcation Analysis

In this section, we explain how to compute numerically the solutions and bifurcation
diagrams of dynamical systems. First, note that usual numerical bifurcation analy-
ses do not use the direct numerical simulations of dynamical systems (differential
or difference equations), but find the solutions of such dynamical systems by dif-
ferent ways. Numerical bifurcation analyses trace continuously the solutions started
from a certain known solution (an equilibrium point or a periodic orbit) if such so-
lutions are “connected.” Thus, the numerical bifurcation analyses can find not only
stable solutions but also unstable solutions. Also, such bifurcation analyses can be
performed in relatively short computational time since they do not require lengthy
numerical simulations of dynamical systems.
There are several computer softwares for the numerical bifurcation analy-
ses, such as CONTENT (http://www.enm.bris.ac.uk/staff/hinke/dss/continuation/
content.html), MATCONT (http://www.matcont.ugent.be/), DDE-BIFTOOL (http://
twr.cs.kuleuven.be/research/software/delay/ddebiftool.shtml), Oscil8 (http://oscill8.
sourceforge.net/), BunKi (http://bunki.ait.tokushima-u.ac.jp:50080/), and so on.
The most famous computer software of bifurcation analysis is AUTO (AUTO97 and
AUTO07P) and many bifurcation diagrams in this book were made by using AUTO
(http://cmvl.cs.concordia.ca/auto/). In this section, however, let us explain another
software called XPPAUT or XPP (XPP-Aut: X-Windows Phase Planes plus Auto)
which adds the user-friendly graphical user interface (GUI) and several useful func-
tions to AUTO. XPPAUT can be obtained from http://www.math.pitt.edu/bard/
xpp/xpp.html. Note that the following examples of XPPAUT were performed on
both MacOS X and Windows XP. Some appearances and functions of XPPAUT
may be different between the operating systems. The usage of XPPAUT, however,
is essentially the same and does not depend on the operating systems.
In order to start the XPPAUT on MacOS with X-windows installed, just type
xppaut. On the Windows machine, start cygwin and type xinit or startx to
start the X-window, and type ./xpp.bat in the xppall folder (under the usual
installation of the XPPAUT).
When we start XPPAUT by typing xppaut, the small window shown in Fig. 1.16
will appear. We can see a list of files (in this case, only one file: hopf.ode) therein.
28 1 A Very Short Trip on Dynamical Systems

Fig. 1.16 Starting window of

XPPAUT (or simply XPP).
We can choose an ode file to
be analyzed

Fig. 1.17 The main window of the XPPAUT. On the top row and in the left column, there are many
buttons. The right big panel is the main graphic window

We can scroll up and down the list (if this includes many files) by using ^^ (or ^)
and vv (or v), and change directories. If we click a mouse button on the hopf.ode,
The big window shown in Fig. 1.17 comes up. This is the main window of XPPAUT
(note that at the start-up, such a drawing of a solution in the “Graphic Window”
which is the right panel in the main window does not appear). In order to exit the
XPPAUT, click on the File button in the left column of the XPPAUT main window
and then click on Quit.
1.5 Computational Bifurcation Analysis 29

The content of the hopf.ode file is as follows:

# Filename: hopf.ode
# This is comment. Example of the Hopf bifurcation
# in the differential equations
x’= L*x - y - x*( x^2 + y^2 )
y’= x + L*y - y*( x^2 + y^2 )
# parameters
par L=-0.5
# initial conditions
init x=0.5, y=0.5
done

Fourth and fifth lines describe the differential equations which are the same as
(1.49). Parameters and initial conditions are also described in the file.

Solving Differential Equations

In the leftmost of the top row of Fig. 1.17, we can see the ICs button where we can
set initial values. If we click on the button, a small Initial Data window shown in
Fig. 1.18 appears. We can see the initial “initial values” described in the hopf.ode
file have already set to the X and Y variables. If we click on the Go button in the
Initial Data window, a solution started from the initial condition will be drawn
as in Fig. 1.17 where the abscissa and the ordinate are the time T and the variable X,
respectively. If we change the initial values in the Initial Data window and press
Go button again, another solution started from a new initial condition will be drawn
(superimposed).
We can change the X vs. T plot (which is shown on the top of the “Graphic
Window” in the main window) to the Y vs. X plot (i.e. X-Y phase plane) as follows.
Click on the Viewaxes button and then 2D button in the left column of the XPP main
window, the 2D View window such as Fig. 1.19 comes up (note that Fig. 1.19 is the
window after some changes were made). Then, change *X-axis and *Y-axis to X
and Y, respectively. We can change the range of X (Y) axis by changing Xmin and
Xmax (Ymin and Ymax, resp.).
Clicking on the Initialconds button and then (M)ouse (or m(I)ce) allows
us to set initial values by using a mouse. By m(I)ce, we can draw many orbits
and only one orbit by (M)ouse. Figure 1.20 shows the X-Y phase plane and several

Fig. 1.18 The Initial

Data window in which we
can set initial values or initial
states of differential equations
30 1 A Very Short Trip on Dynamical Systems

Fig. 1.19 The 2Dview window. We can change the variables and their range to be plotted in the
X- and Y -axes

Fig. 1.20 The vector field and several orbits (solutions) of differential equations

orbits started at several initial values inputted by mouse clicks (note that the vector
field is also drawn by arrows. We will explain how to draw such a vector field later.)
We can see that the origin .0; 0/ is a stable equilibrium point and all orbits spirally
tend to the point. This figure is essentially the same as Fig. 1.10 (check the value of
the parameter L of the hopf.ode file). Note that the (M) and (I) are the keyboard
shortcuts of (M)ouse and m(I)ce, and we can type these keys on a keyboard instead
of mouse clicks.
In order to draw a vector field shown in Fig. 1.20, click on Dir.field/flow
and (D)irect Field. Then a message “Grid: 10” appears in the second line of
the window. This means the number of grids where the vector field is drawn by
arrows. If we do not want to change this default value, just hit return (or enter)
1.5 Computational Bifurcation Analysis 31

key on a keyboard. Then, a vector field will appear. We can draw various orbits
superimposed on this vector field by the method explained above. In order to clear
the graphic window, just click a mouse button on Erase.
Exercise 1.5. Change the value of the parameter as L D 0 and L D C0:5, and
draw several obits with different initial values and vector fields. Then obtain similar
X –Y phase plane to Fig. 1.10b,c. (In order to change the parameter value, use the
Param button on the top row of the XPP main window.) 

Bifurcation Diagrams

Next, let us draw a bifurcation diagram using XPPAUT. First, we note that the nu-
merical bifurcation analysis performs a continuation. Namely, changing parameter
values step by step, it seeks various solutions (equilibrium points and periodic or-
bits) from an initial solution. (Note that an initial solution does not mean an initial
condition.) Thus, at first, we must choose or fix an initial solution from which the
continuation is made. Often, we use a stable equilibrium point as an initial so-
lution. To obtain such a stable equilibrium point, use Initialconds and (G)o,
and then draw a solution, which approaches the stable equilibrium point, from any
initial value. Next, repeat both Initialconds and (L)ast (I and L keys) several
times. This command solves or integrates differential equations from the final (last)
value of previous solution. Thus we can obtain the approximate values of the stable
equilibrium point. We can see these values by clicking on Sing pts and (G)o.
File and Auto buttons create a new window such as Fig. 1.21 for bifurcation
analysis using the AUTO. (Note that results of bifurcation analysis have already

Fig. 1.21 The AUTO Window. Equilibrium points are drawn as a function of the parameter L.
Thick and thin lines show stable and unstable equilibrium points, respectively. The labels 1 and
3 are the starting and ending points, respectively. The label 2 denotes the Hopf bifurcation point
particularly
32 1 A Very Short Trip on Dynamical Systems

Fig. 1.22 AUTO Numerics Window. We can change the parameter values that AUTO uses in the
numerical bifurcation analysis

been drawn, which will not appear at start-up.) By Axes and hI-lo in the left
column of the AUTO window, we can set the variable to be plotted, the main bifur-
cation parameter (L here) (and the second parameter also) as well as the maximum
and minimum scales of periodic orbits. If we click on Numerics, a new window
such as Fig. 1.22 will appear. By this window, we set several parameters for nu-
merical bifurcation analysis. For example, set the value of Dsmax to 0.05 which is
the maximum step size of changing the bifurcation parameter. Also set the values
of Par Min and Par Max to 1 and 3, respectively. These values denote the range
where the main bifurcation parameter (L) should be changed. Then press on OK to
close the window.
Now we are ready to obtain a bifurcation diagram. To do so, just click on Run
and Steady state. Then the bifurcation diagram such as Fig. 1.21 can be obtained.
The line with labels 1, 2 and 3 denote equilibrium points. The differential equations
described in the hopf.ode file (or (1.49)) have an equilibrium point at the origin
.0; 0/ for all values of L (). Thus we can see that the X value of the line is zero
irrespective of the L value on the abscissa. The labels 1 and 3 denote the starting and
ending points of bifurcation analysis, respectively. The point labeled 2 is a special
(bifurcation) point detected automatically. The thick and thin lines denote stable and
unstable equilibrium points, respectively. We can see that the stability change at
the (Hopf bifurcation) point 2 from which a branch of periodic orbits should be
bifurcated (see (1.49) and its explanation).
Next we proceed to calculate the branch of periodic orbits from the Hopf bifur-
cation point 2. To do so, let us “grab” the point 2. If we click on Grab button, then a
cross (cursor) will appear on the one-parameter bifurcation diagram. We hit Tab (or
arrow) keys several times on a keyboard to move the cursor on the Hopf bifurcation
(HB) point labeled by 2. Then we grab the point by hitting the enter or return
key. Finally, click on Run and Periodic buttons to obtain the bifurcation diagram
shown in Fig. 1.23. The closed circles (dots) denote the maximum and minimum
1.5 Computational Bifurcation Analysis 33

Fig. 1.23 AUTO Window. The bifurcation diagram of both equilibrium points and periodic orbits
is shown. The closed circles (dots) bifurcated from the Hopf bifurcation point labeled by 2 show
stable periodic orbits

values of stable periodic orbits. We can see that the Hopf bifurcation labeled by 2 is
super-critical and stable periodic orbits are bifurcated through the Hopf bifurcation.
Note that, when we make an excursion along the curve of solutions (equilibrium
points or periodic orbits) by clicking on Grab and by hitting tab and arrow keys, we
can see several information (stabilities, parameter values, etc.) of the solution both
in the lower-left corner and below the bifurcation diagram of the AUTO window.

Exercise 1.6. Create an ode file for (1.51). For the parameter values such as L D
0:5, L D 0 and L D C0:5, draw several obits with different initial values and
vector fields to obtain similar X –Y phase planes to Fig. 1.12a–c. Also, obtain a
bifurcation diagram of (1.51) similar to Fig. 1.13. 

For a reference, Tables 1.1 and 1.2 show a partial list of XPP commands in the
XPP main window and in the AUTO window, respectively. For the installation and
the full list of XPP commands, visit
http://www.math.pitt.edu/~bard/xpp/xpp.html
34 1 A Very Short Trip on Dynamical Systems

Table 1.1 Commands in the XPP main window

Button Action and use
ICs A list of initial data will appear, and change the
initial data you want to change
Param A list of parameters will appear, and change the
numerics parameters you want to change
quit Quit the XPPAUT window
(I)nitialconds (G)o Use the initial data and the current numerics
parameters to solve the equation. The output
is drawn in the window and the data are saved
for later use. The solution continues until
either the user aborts by pressing Esc, the
integration is complete, or storage runs out
(L)ast Use the end result of the most recent integration
as the starting point of the current integration
(M)ouse Allows you to specify the values with the mouse.
Click at the desired spot in a phase-plane
picture. You must have a two-dimensional
view and only variables along the axes
m(I)ce Allows you to choose multiple points with the
mouse. Click Esc when done
(D)ir.field/flow (D)irection Choosing the direction field option will prompt
fields you for a grid size. The two-dimensional
plane is broken into a grid of the size
specified and lines are drawn at each point
specifying the direction of the flow at that
point
(E)rase Erasing the screen
(S)ing pts (G)o A new window about eigenvalue and its stability
will appear
(V)iewaxes (2)D Set the variables to place on the axes, upper and
lower limits of the two axes and the labels for
the axes
(F)ile (A)uto Bring up the AUTO window if you have
installed it
(Q)uit Quit the XPPAUT window
1.5 Computational Bifurcation Analysis 35

Table 1.2 Commands in the XPP AUTO window

Button and parameter Action and use
Close Close the AUTO window
(P)arameter A list of parameters will appear. Type in the
names of the parameters you want to use
(A)xes h(I)-lo Plot both the max and min of the chosen variable
(convenient for periodic orbits)
(T)wo par Plot the second parameter versus the primary
parameter for two-parameter continuations
(N)umerics Ntst This is the number of mesh intervals for
discretization of periodic orbits. If you are
getting bad results or not converging, it helps
to increase this
Nmax The maximum number of steps taken along any
branch. If you max out, make this bigger
Ds This is the initial step size for the bifurcation
calculation. The sign of Ds tells AUTO the
direction to change the parameter
Dsmin The minimum step size (positive)
Dsmax The maximum step size. If this is too big, AUTO
will sometimes miss important points so if it
seems to miss a stability transition, or if the
diagram is jagged, decrease this
Par Min This is the left-hand limit of the diagram for the
principle parameter. The calculation will stop
if the parameter is less than this
Par Max This is the right-hand limit of the diagram for the
principle parameter. The calculation will stop
if the parameter is greater than this
(R)un (S)teady state Starting at a new steady state
(P)eriodic Compute the branch of periodics emanating from
the Hopf point
(T)wo Param Compute a two parameter diagram of the selected
points
(G)rab Select a special point
(U)sr period Number (0-9) Choose the number of points you want to fix, and
the fixed points will be labeled
(C)lear Clear the diagram
re(D)raw Redraw the diagram
(F)ile (P)ostscript Make a hard copy of the bifurcation diagram
Chapter 2
The Hodgkin–Huxley Theory of Neuronal
Excitation

Hodgkin and Huxley (1952) proposed the famous Hodgkin–Huxley (hereinafter

referred to as HH) equations which quantitatively describe the generation of action
potential of squid giant axon, although there are still arguments against it (Connor
et al. 1977; Strassberg and DeFelice 1993; Rush and Rinzel 1995; Clay 1998). The
HH equations are important not only in that it is one of the most successful math-
ematical model in quantitatively describing biological phenomena but also in that
the method (the HH formalism or the HH theory) used in deriving the model of a
squid is directly applicable to many kinds of neurons and other excitable cells. The
equations derived following this HH formalism are called the HH-type equations.
The dynamical system theory is very useful to analyze and understand the
dynamics of the HH equations. On the other hand, the HH equations have rich math-
ematical structures and give many insights to mathematics also. For example, there
are still many studies and mathematical findings on the “classical” HH equations
(Plant 1976; Rinzel 1978; Hassard 1978; Troy 1978; Rinzel and Miller 1980;
Labouriau 1985, 1989; Hassard and Shiau 1989; Shiau and Hassard 1991; Bedrov
et al. 1992; Guckenheimer and Labouriau 1993; Labouriau and Ruas 1996; Fukai
et al. 2000a,b). This chapter gives an overview of the dynamics of the HH equations
and of the mechanism of the action potential generation.

2.1 What is a Neuron? Neuron is a Signal Converter

Figure 2.1 illustrates a shape and a function of neurons schematically. Our brain is
a complicated network of a tremendous number of neurons. Right figure shows a
small network of three neurons. A neuron has a very special shape which is much
different from usual sphere-shaped or disklike cells. A soma is the main body of
neuron from which a long cable called an axon is extended. Neurons transmit and
exchange electric signals called action potentials or spikes, each other. (The gener-
ation of a spike is also called as the excitation or the firing of a neuron.) Neurons
receive the spikes at a synapse which is a connection between neurons. Then, the
electric signals or information is transmitted in the direction from a dendrite to an
axon. The upper-left panel of Fig. 2.1 illustrates the waveform of action potentials.
Action potential or a spike has an amplitude of about 100 mV.

S. Doi et al., Computational Electrophysiology, 37

DOI 10.1007/978-4-431-53862-2 2,  c Springer 2010
38 2 The Hodgkin–Huxley Theory of Neuronal Excitation

v (mV)
100 Neurons
80
Action potential
60
40
20
0
–20 Inputs to a neuron
–40 Axon
0 10 20 30 40 terminal
time (ms) Axon

Soma Synapse
Dendrite
?
Neuron

Fig. 2.1 Diagrams illustrating: A network of three neurons which exchange electric signals called
action potentials, each other (right). Waveform of action potentials (upper left). Neuron as a device
which converts input signals to output signals (lower left)

Typical neurons do not generate any spikes without input signals (i.e. spikes from
other neurons). A sufficiently large input pulse causes a neuron to generate an output
spike, as illustrated in the upper-left panel, whereas no output spike is generated by
a small input (the first pulse in the lower trace of the upper-left panel). Therefore, a
neuron possesses a threshold or all-or-none characteristic. There is a special period
or timing called the refractory period (the timing of the downstroke of the action po-
tential) in which the neuron cannot produce any output spike even though sufficient
amount of inputs (the third and fourth pulses in the panel) were put in the neuron.
Thus, we can consider a neuron as a device which transforms or converts the train
of input spikes to a train of output spikes (see the lower-left panel of Fig. 2.1).

2.2 The Hodgkin–Huxley Formulation of Excitable

Cell Membranes

This section briefly explains the framework of the HH formalism to model the action
potential generation of neurons and of other excitable cells.
Biological cells, including neurons, are enclosed by a plasma membrane or sim-
ply membrane which separates the intracellular and extracellular water-containing
media. The cell membrane consists of lipid bilayer, as shown in Fig. 2.2. There
are various ions in both the intra- and extra-cellular regions. The concentrations
of ions, however, are much different between the intra- and extra-cellular regions.
For example, the concentration of KC ion is high and low in the intra- and extra-
cellular regions, respectively. On the contrary, that of NaC ion is low and high
2.2 The Hodgkin–Huxley Formulation of Excitable Cell Membranes 39

Ion-selective filter
Membrane potential

K+ chanel
Na+ 460 [mM]
Cell membrane
K+ 10 K+ 400 Gate Cell membrane
(Lipid bilayer)
–
Cl 540 Na+ 50 Cl– 70

– 70 [mV] Outside the cell membrane

+
Na channel Leak channel
ENa EK

Capacitance
Concentraton distribution of ions ......
inside and outside the cell
gNa gK

Inside the cell membrane

Fig. 2.2 The equivalent-circuit formulation of a cell membrane and ionic channels by Hodgkin
and Huxley

Na channel K channel Open and close of a gate

dm
= (1 – m) am(V ) – mbm(V )
dt
m gates n gates
bm(V )
Open Close
h gate am(V )

Fig. 2.3 Diagrams explaining the gate dynamics

in the intra- and extra-cellular regions, respectively. Usually, the resistance of the
membrane is very high and the membrane acts as an insulator to the movement of
ions. If the electrical potential at the inside surface of the cell membrane is com-
pared to the potential at the outside surface, there is a potential difference or voltage
called the transmembrane potential or simply the membrane potential.
In the membrane, there are holes through which ions can move in and out. Such
a hole is called an ion channel and consists of membrane proteins. Ion channels
are not simple holes (pores) or passive resistors through which ion flux flows. Ion
channels are selective for a particular ion. For example, an ionic channel named NaC
channel can pass only NaC ions. Also, ion channels are dynamic and sensitive to the
membrane potential and to other factors. Namely, they open and close depending
on such factors. An ion channel has several gates and the opening and closing of
the ion channel are controlled by the gates as shown in Fig. 2.3. Note that, there
40 2 The Hodgkin–Huxley Theory of Neuronal Excitation

are various types of ionic channels which pass a specific ion. Particularly, there are
many variants of KC channels classified by their various characteristics (Adams
1982; Crill and Schwindt 1983; Llinas 1988; Hille 1992).
The basic idea of the HH formalism is to just recognize the cell membrane as a
simple electric circuit as shown in the lower-right panel of Fig. 2.2. The capacitive
property of the cell membrane is denoted by the capacitor with a certain capaci-
tance in the circuit. NaC and KC channels are modeled by the resistors which have
conductances gNa and gK , respectively. Note that the resistors are not linear but non-
linear, and also are dynamic: the values of the conductances vary temporally, which
are explained later. There is a tendency that NaC ions flow inward and that KC ions
flow outward the cell membrane because there are differences in the concentrations
of the ions between inside and outside of the membrane. Namely, ions have a ten-
dency to move down their concentration gradients. Such a tendency is denoted by
the batteries with voltages ENa and EK in the circuit. These voltages depend on the
inside–outside concentration difference in each ion. Notice that the polarities of the
batteries ENa and EK are reversed.

2.3 Nonlinear Dynamical Analysis of the Original

HH Equations

The Hodgkin–Huxley equations (Hodgkin and Huxley 1952) of a squid giant axon
are simply the differential equations of the electric circuit shown in Fig. 2.2 and are
described as follows:

@v @2 v
C D a C G.v; m; n; h/ C Iext ; (2.1a)
@t 2 @x 2
@m
D ˛m .v/.1  m/  ˇm .v/m; (2.1b)
@t
@n D ˛ .v/.1  n/  ˇ .v/n; (2.1c)
n n
@t
@h
D ˛h .v/.1  h/  ˇh .v/hI (2.1d)
@t

G.v; m; n; h/ D INa .v; m; h/ C IK .v; n/ C IL .v/

D gN Na m3 h.VNa  v/ C gN K n4 .VK  v/ C gN L .VL  v/I (2.2)

C D 1 F cm2 ; gN Na D 120 mS cm2 ; gN K D 36 mS cm2 ; gN L D 0:3 mS cm2 ;

VNa D 115 mV; VK D 12 mV; VL D 10:599 mVI
2.3 Nonlinear Dynamical Analysis of the Original HH Equations 41

0:1.25  v/
˛m .v/ D ; ˇm .v/ D 4e v=18 ;
expŒ.25  v/=10  1
0:01.10  v/
˛n .v/ D ; ˇn .v/ D 0:125e v=80 ;
expŒ.10  v/=10  1
˛h .v/ D 0:07 expŒv=20; ˇh .v/ D 1 I
expŒ.30  v/=10 C 1
where v (mV) is the membrane potential. Equation (2.1a) simply denotes the
Kirchhoff’s law. (In (2.1), a neuron is considered as a cylinder-shaped cell and
the dependence of v on its position x is also taken into account. The term
.a=2/.@2 v=@x 2 / denotes the diffusions of ions along the axis of the cylinder.
However, either the shape of neurons or the x-dependence are not considered in
this book.) INa and IK are the currents through NaC and KC channels, respectively.
The current IL is the leak current and denotes all residue currents through a cell
membrane other than NaC and KC currents.
As seen from (2.2), the NaC current INa is denoted by gN Na m3 h.VNa  v/ which
takes a form of (Conductance)  (Voltage): Ohm’s law. The voltage VNa is called
the Nernst potential or the equilibrium potential or sometimes the resting potential
(do not confuse with the resting potential of whole membrane) of NaC ion. The
Nernst potential is the potential where the tendency of ions to move down their
concentration gradient is exactly balanced with the force by the electric potential
difference; no NaC current flow through the NaC channel when v D VNa . gN Na m3 h
(D gNa in the circuit of Fig. 2.2) denotes the conductance of NaC channel where the
constant gN Na is called the maximum conductance of the channel and m3 h denotes
dynamic or temporal change of the conductance. In the KC current IK , the term n4
denotes the temporal change of KC channel conductance.
The variables m, n and h take a (dimensionless) value between zero and unity,
and are called the gate variables. As seen from the left panel of Fig. 2.3, it is as-
sumed that NaC channel possesses three m-gates and single h-gate whereas KC
channel four n-gates. In the HH formalism, it is also assumed that the variables m,
n and h denote the probabilities that corresponding gates are open. The dynamic
opening and closing of gates obey a simple (linear) process described by (2.1b–d)
where ˛m .v/ (ˇm .v/) is the rate “constant” for changing from closed (opened) state
to opened (closed, resp.) state, as shown in the right panel of Fig. 2.3. Note that the
rate “constant” ˛m .v/ and ˇm .v/ are not actually the constants but the functions
which depend on the membrane potential v. m and h are also called the activation
and inactivation variables of NaC ionic channel, respectively, while n the activation
variable of KC channel. The reason of this naming is because m and n are the in-
creasing functions of v while h the decreasing one. Iext ( A cm2 ) is the constant
current externally applied to a neuron. The constants a and  are the radius and
resistivity of the “cylindrical” axon, respectively. Throughout this book, we do not
treat either such a cylindrical axon or the partial differential equation (2.1) which is
an infinite-dimensional dynamical system. For the rich and complicated dynamics of
such neuronal cable equations, see Carpenter (1977), Horikawa (1994), Kepler and
Marder (1993), Rinzel and Keener (1983), Poznanski (1998), and Yanagida (1985,
1987, 1989).
42 2 The Hodgkin–Huxley Theory of Neuronal Excitation

2.3.1 Simple Dynamics of Gating Variables

In the following, we assume that the membrane potential v is spatially constant and
omit the spatial derivative in (2.1); we consider the following HH equations for a
space-clamped squid giant axon:

dv
C D G.v; m; n; h/ C Iext ; (2.3a)
dt
dm 1
D .m1 .v/  m/; (2.3b)
dt
m .v/
dn
D 1 .n1 .v/  n/; (2.3c)
dt
n .v/
dh
D 1 .h1 .v/  h/; (2.3d)
dt
h .v/

where

1 ˛x .v/

x .v/ D ; x 1 .v/ D ; x D m; n; h: (2.4)
˛x .v/ C ˇx .v/ ˛x .v/ C ˇx .v/

(The word “space-clamped” means that both the shape of a neuron and the depen-
dence of the membrane potential v on the spacial position x are ignored while both
were taken into account in (2.1).)
Figure 2.4 shows an example of a numerically solved solution of the HH
equations (2.3). Panel (a) is a waveform of the membrane potential. A pulsatile
input applied at a time t D 5 ms induces an action potential. Panel (b) is the wave-
forms of the gating variables m, n and h. Total membrane current, Na current, K
current and leak current are shown in (c)–(f), respectively.
The HH equations (2.3) are nonlinear differential equations with four variables
and apparently look very complicated. Equations (2.3b–d) which describe the dy-
namics of gating variables, however, share a simple common structure. Functions

x .v/ and x 1 .v/, x D m; n; h depend on the membrane potential v and thus vary
temporally with the temporal change of v. If we assume that the functions do not de-
pend on v (
x .v/ 
x , x 1 .v/  x 1 ), then (2.3b–d) reduce to a linear differential
equation
dx 1
D
.x 1  x/; x D m; n; h;
dt x

whose analytical solution with an initial value x D x0 is

x.t/ D exp.t=
x /.x0  x 1 / C x 1 :
2.3 Nonlinear Dynamical Analysis of the Original HH Equations 43

a b
v (mV) Iext (µA/cm2)
100 50 1.0
80 40 0.8
60
30 0.6
40
20 20 0.4
0 10 0.2
–20
0
0 5 10 15 20 25 0 5 10 15 20 25
c 2
d
G (µA/cm ) INa (µA/cm2)
250
200 600
150
100 400
50
0 200
–50
0
0 5 10 15 20 25 0 5 10 15 20 25
e f
2 2
IK (µA/cm ) IL (µA/cm )
0 5
0
–200
–5
–400 –10
–15
–600 –20
–25
–800
0 5 10 15 20 25 0 5 10 15 20 25
t (ms) t (ms)

Fig. 2.4 Solution of the Hodgkin–Huxley equations

The variable x.t/ approaches x 1 at a speed depending on a time constant

x .
In (2.3b–d), although they are function of v and we cannot solve the equation
analytically,
x .v/ still preserves a role of “time constant” and x 1 .v/ is the
steady-state function to which the variable x asymptotically approaches in a steady
(or stationary) state.
Figure 2.5a shows the functions x 1 .v/; x D m; n; h. In spite of the complicated
functional forms, these functions are all monotonic functions and have the shape of
so-called “sigmoid function.” m1 .v/ and n1 .v/ are increasing functions and thus
variables m and n are activation variables while h1 .v/ is a decreasing function and
h is an inactivation variable. Figure 2.5b shows the functions
x .v/; x D m; n; h
which change depending on v. The function
m .v/, however, much smaller than

n .v/;
h .v/ in the whole range of v. In the following, let us explore the dynamics of
the HH equations regarding this time-scale difference.
44 2 The Hodgkin–Huxley Theory of Neuronal Excitation

a b
1.0
n
0.8 h 8 τh

0.6 6
m τn
0.4 4
τm
0.2 2
0.0 0
–100 –50 0 50 100 –50 0 50 100 150
v (mV) v (mV)

Fig. 2.5 Various functions in the Hodgkin–Huxley equations

2.3.2 FitzHugh’s Subsystem Analysis of the HH Equations

In this subsection, following the pioneering paper FitzHugh (1960), let us see from
what dynamics the threshold property of a neuron or the HH equations comes.
The HH equations have four variables and it is difficult to observe the full (four-
dimensional) state space directly. The separation of the full state space to several
subspaces, however, resolves this difficulty. As shown in Fig. 2.5, the “times con-
stants” of variables n and h are much bigger than that of m; n and h change their
values more slowly than m. Thus we (temporarily) ignore the dynamics of n and h
in the HH equations and consider the following system:

dv
C D gN Na m3 h.VNa  v/ C gN K n4 .VK  v/ C gN L .VL  v/; (2.5a)
dt
dm
D 1 .m1 .v/  m/: (2.5b)
dt
m .v/

(We call this system as the v–m subsystem.) In the v–m subsystem, v and m are the
dynamic variables while h and n are set in suitable values as a “parameter.”
By using the v–m subsystem, let us explain the firing process in the HH equations
in the following order:

quiescent state ! depolarization ! decrease of h ! increase of n

! repolarization :

Figure 2.6a shows the v–m phase plane of (2.5) when the values of h and n are fixed
to that of the quiescent state (a stable equilibrium point) of the HH equations (2.3).
The m-nullcline (a curve in which d m=dt D 0: m D m1 .v/) is a sigmoidal mono-
tonically increasing function. (For more explanation of nullclines, see Chap. 3.) We
can see that the v-nullcline (dotted curve) and m-nullcline (solid curve) intersect in
the three points v1 , v2 and v3 which are equilibrium points of the v–m subsystem
(panel b is the magnification of lower-left region of a). The leftmost equilibrium
2.3 Nonlinear Dynamical Analysis of the Original HH Equations 45

a m b m
1.0 100
v∗
dm =0
3 v∗2 dv
=0
0.8 80
dt dt
0.6 60

×10–3
v∗1
0.4 40
dm
0.2 v∗2 dv 20 =0
v∗1 =0 dt
dt
0.0 0
0 20 40 60 80 100 0 2 4 6 8 10
v (mV) v (mV)

c m
d m
1.0 dm 1.0
=0 v∗
dt 3 dm = 0
0.8 0.8 dt
dv =0
0.6 0.6 dt
dv
v∗2 =0
dt
0.4 0.4

0.2 ∗ 0.2 v∗1

v1
0.0 0.0
0 20 40 60 80 100 0 20 40 60 80 100
v (mV) v (mV)

Fig. 2.6 Phase plane of the v–m subsystem (2.5) of the HH equations. (a) n D 0:317677,
h D 0:596120. (b) Magnification of (a). (c) n D 0:317677, h D 0:02. (d) n D 0:5, h D 0:02

point v1 corresponds to the quiescent state (a stable equilibrium point) and the mid-
dle point v2 is a saddle point whose stable manifold (the broken curve tending to
v2 ) forms a threshold between exciting and non-exciting, which means that the HH
model when h and n are fixed to the values of quiescent state is the type-I neuronal
model with a strict threshold. If a sufficiently large stimulus is applied to a neuron
in a quiescent state v1 , the state point moves rightwards beyond the stable manifold
of v2 and then goes towards the rightmost equilibrium v3 which corresponds to the
depolarized state of a neuron. If a membrane potential v increases, h is decreased
(after a slight delay) because the variable h tends to the function h1 .v/ which is a
decreasing function of v.
Figure 2.6c is the phase plane with a smaller value of h. In the v–m subsystem
(2.5), the v-nullcline does depend on both h and n while the m-nullcline does not
depend on them. From (2.5a), the v-nullcline is obtained by

gN K n4 .v  VK / C gN L .v  VL /
m3 D (2.6)
gN Na h.VNa  v/

from which we can see that the decrease of h moves the v-nullcline upward
(the shape of v-nullcline is also changed). Note that we consider this equation in the
range of 0  m  1, thus the right-hand side is positive. As a result of displacement
of the v-nullcline, three intersections of the v-, m-nullclines become more clearly
46 2 The Hodgkin–Huxley Theory of Neuronal Excitation

Fig. 2.7 v–m Phase plane of dv dn dh

= = =0
the HH equations (2.3) dt dt dt
1.0

0.8
dm
m
0.6 =0
dt
0.4

0.2

0.0
0 20 40 60 80 100
v (mV)

distinguishable. As is seen from Fig. 2.5b, in the depolarized (high-voltage) region

of v,
n .v/ is slightly larger than
h .v/. Thus n increases subsequently to the de-
crease of h (practically, these changes occur almost simultaneously) since n1 .v/ is
a increasing function of v.
Figure 2.6d shows the case that n is increased in addition to the decrease of h. As
is seen from (2.6), the further increase of n moves the v-nullcline upward. Then two
equilibria v2 and v3 disappear by a saddle-node bifurcation and only equilibrium
v1 remains. As a result of disappearance of those equilibria, the state point near the
equilibrium v3 (depolarized state) cannot stay there and then changes its direction
toward the equilibrium v1 (quiescent state). After this process, h and n change to
increase and decrease, respectively, and then return to the state of Fig. 2.6a.
Figure 2.7 shows the v–m phase plane of the full HH equations (2.3) (projec-
tion of the four-dimensional state space to v–m phase plane). The sigmoid-like
curve is the m-nullcline and is same as the case of v–m subsystem. The
other curve denotes the intersection of v; n; h-nullclines; a curve such that
d v=dt D d n=dt D dh=dt D 0:

gN K fn1 .v/g4 .v  VK / C gN L .v  VL /
m3 D :
gN Na h1 .v/.VNa  v/

The intersection of these two nullclines corresponds to the equilibrium point of the
original HH equations (2.3) and thus the HH equations have a unique equilibrium.
This implies that in a strict sense the original HH equations are a type-II neuronal
model without a distinct threshold. As is seen above, however, in a very short time
range (i.e. if we ignore the temporal change of h and n) the HH equations behave
like a type-I neuronal model which has a distinct threshold. Thus we can understand
why the HH equations have a relatively sharp threshold although they do not have
any threshold in a strict sense.
2.3 Nonlinear Dynamical Analysis of the Original HH Equations 47

2.3.3 Dynamic I–V Relation of the Squid Giant Axon Membrane

In this subsection, let us investigate the HH equations from more global viewpoint
without entering into its detailed dynamics. Figure 2.8 shows dynamic or transient
current–voltage relation of the HH equations; total membrane currents G when a
time t elapses after the membrane voltage is instantaneously changed to v (mV)
from the quiescent state are plotted for various v values. These current–voltage
relations are shown for various values of the time t. At a time t D 0:02 ms, the
relation is almost linear. After some time elapses, negative-resistor characteristics
appear and the current–voltage relation becomes N-shaped. After sufficiently long
time elapses, the relation shows a rectifier characteristics in which only outward
current can be flowed.

G (µA/cm2) t = 0.02 t = 0.05

20 20
0
0
–20
– 20 –40
– 40 –60
– 60 –80
– 100
– 80 – 120
– 40 0 40 80 120 – 40 0 40 80 120
v (mV)

t = 0.07 t = 1.0

0 1000
500
– 50
0
– 100 – 500
– 1000
– 150
– 1500
–40 0 40 80 120
– 40 0 40 80 120

t = 4.0 t = 10.0
0 0

– 1000 – 1000

– 2000 – 2000

– 3000 – 3000

– 4000 – 4000
–40 0 40 80 120 –40 0 40 80 120

Fig. 2.8 Transient current–voltage relations of the Hodgkin–Huxley membrane

48 2 The Hodgkin–Huxley Theory of Neuronal Excitation

In order to see how these current–voltage relations are generated, let us

decompose the total membrane current to its components: INa and IK where the
(small) leak current is ignored. Figure 2.9 is the similar voltage–current relations to
Fig. 2.8 for Na and K channels. When t D 0:02, the relation of K channel (dotted
curve) is almost linear. After time elapses, the K current in a low-voltage range be-
comes small and finally, the current–voltage relation becomes a rectifier. In the case
of Na channel (solid curve), when t D 0:02, it does not flow much current (almost
flat). After suitable time elapses, it shows large nonlinear characteristics and then
finally becomes almost flat again. The combination of these two current–voltage
relations makes the I –V relation of the total current in Fig. 2.8.
The above current–voltage relations are obtained by numerically solving the
HH equations (2.3). The relation when t D 10:0 is considered as a steady-state
current–voltage relation. This relation can be obtained directly (without numerical
simulation) from the HH equations (2.3) as

G.v; m1 .v/; n1 .v/; h1 .v//:

INa, IK(µA/cm2) t = 0.02 t = 0.05

10 40
0
20
–10 Na
0
–20
–30 –20
K
–40 –40
–50 –60

–40 0 40 80 120 –40 0 40 80 120

v (mV)

t = 1.0 t = 0.07

1000 100
500 50
0
–500 0
–1000 –50
–1500
–40 0 40 80 120 –40 0 40 80 120

t = 4.0 t = 10.0
0 0

–1000 –1000

–2000 –2000

–3000 –3000

–4000 –4000
–40 0 40 80 120 –40 0 40 80 120

Fig. 2.9 Transient current–voltage relations of Na and K channels

2.3 Nonlinear Dynamical Analysis of the Original HH Equations 49

Fig. 2.10 Steady-state G (µA/cm2)

current–voltage relation of 0
the Hodgkin–Huxley
–1000
equations
–2000
–3000
–4000
–5000

–50 0 50 100 150

v (mV)

a b
4000 1.0
3000

G (mA/cm2)
0.5
G (mA/cm2)

2000
1000 0.0
0
–1000 – 0.5
–2000
–1.0
–50 0 50 100 150 –50 –40 –30 –20 –10 0 10 20
v (mV) v (mV)

Fig. 2.11 (a) Steady-state current–voltage relation of the v; m-subsystem of the HH equations.
(b) Magnification of (a)

This steady-state current is plotted as a function of v in Fig. 2.10. Similarly to the

t D 10 case of Fig. 2.8, we can see the rectifier characteristics.
Figure 2.11 shows the steady-state current of the v–m subsystem (2.5):

G.v; m1 .v/; n ; h /

as a function of v where n and h denote the quiescent-state values of n and h,

respectively. Panel (b) is the magnification of (a), from which the current–voltage
curve intersect with the horizontal line at three points which correspond to the three
equilibrium points of the phase plane of Fig. 2.6a. From this observation, we can
also see the threshold property of the HH equations.
Comparing Fig. 2.11 to the transient current–voltage relation of Fig. 2.8, we con-
firm that the v–m subsystem approximates the dynamics of the full HH equations
(2.3) in the time scale of t  1 ms.

2.3.4 Dimension Reduction of the HH Equations

In the v–m subsystem analysis, we investigated the v–m phase plane of the HH
equations by fixing the values of n and h. Namely, we explored the dynamics of
HH equations by decomposing the four-dimensional full phase space into several
n; h-fixed slices.
50 2 The Hodgkin–Huxley Theory of Neuronal Excitation

a b
1.0 1.0
0.8 0.8
0.6 0.6
m n
0.4 0.4
0.2 0.2
0.0 0.0
– 50 0 50 100 150 0.0 0.2 0.4 0.6 0.8 1.0
v (mV) h

Fig. 2.12 Projections of the orbit (solution) of the HH equations to (a) a v–m phase plane and to
(b) a h–n phase plane

This subsection briefly describes the method which reduces the full
four-dimensional system into two-dimensional system (FitzHugh 1961; Krinskii
and Kokoz 1973; Kokoz and Krinskii 1973; Rinzel 1985). Although such dimen-
sion reduction, differently from the v–m subsystem analysis, loses some information
on original dynamics, it is useful to catch the essential feature of the whole four-
dimensional dynamics of the HH equations on a reduced phase plane.
Figure 2.12 shows the projections of an orbit (solution) .v.t/; m.t/; h.t/; n.t//
of the HH equations to (a) a v–m phase plane and to (b) a h–n phase plane. From
panel (a), we can see that the orbit in the region of m  0 or m  1 moves close to
the m-nullcline m D m1 .v/ (broken curve). Panel (b) shows that the orbit moves
restricted in a certain line on the h–n phase plane.
From these observations, we reduce the HH equations (2.3) in two steps:
1. Suppose that the variable m which follows (2.3b) is settled in its steady-state
value: m D m1 .v/ since m is the fast-changing variable (
m is small). Thus we
ignore (2.3b) and substitute m1 .v/ for m in (2.3a).
2. Approximate the orbit on the h–n phase plane by a line n D 0:8.1  h/; we
consider that the variable n linearly depends on h. Thus we ignore (2.3c) and
substitute 0:8.1  h/ for n in (2.3a).
From these reduction steps, we obtain the reduced equations:

dv
C D G.v; m1 .v/; 0:8.1  h/; h/ C Iext ; (2.7a)
dt
dh
D ˛h .v/.1  h/  ˇh .v/h: (2.7b)
dt

This model has only two dynamic variables v and h, and thus has an advantage
that we can analyze the neuronal dynamics on a phase plane rather than the four-
dimensional phase space of the original HH equations.
Figure 2.13a shows an example of the membrane potential waveforms of the orig-
inal HH equations (2.3) (dotted curve) and of the reduced model (2.7) (solid curve).
The upstroke of the membrane potential of the reduced model is slightly faster than
that of the HH equations and the peak value is also bigger than the HH equations.
2.3 Nonlinear Dynamical Analysis of the Original HH Equations 51

a 50
100
80 40

Iext ( mA/cm2)
60 30
v (mV) 40
20 20

0 10
–20
0
–40
0 10 20 30 40 50 60
t (ms)

b
0.7 dh
=0 dv
dt =0
0.6 dt

0.5

0.4
h

0.3

0.2

0.1

0.0
0 20 40 60 80 100
v (mV)

Fig. 2.13 Comparison of (a) membrane potential waveforms and (b) solution orbits in the phase
plane, between the dimension-reduced model (solid curve) and the original HH model (dotted
curve)

This is because we substituted m1 .v/ for m and thus the activation process of Na
channel has been slightly accelerated. Except for these small differences, the two
waveforms of the original and the reduced model are quite similar. This similarity
is surprising since the dimensions of the original and the reduced model are much
different.
Figure 2.13b shows the solution orbit of both models on the v–h phase plane.
The solid curve denotes the orbit of the reduced model (2.7), and the dotted curve
the original HH equations (2.3). The v-nullcline

G.v; m1 .v/; 0:8.1  h/; h/ D gN Na .m1 .v//3 h.VNa  v/

CgN K .0:8.1  h//4 .VK  v/ C gN L .VL  v/ D 0

and the h-nullcline

˛h .v/
h D h1 .v/ D
˛h .v/ C ˇh .v/
are also shown by broken curves. We can see the similarity of orbits of the two
models except for the slight difference in the action potential upstroke.
52 2 The Hodgkin–Huxley Theory of Neuronal Excitation

Topological features (one nullcline is N-shaped and the other is monotonic) of

the phase plane of the reduced model resemble to that of the Bonhoeffer–van der Pol
(BVP) or FitzHugh–Nagumo (FHN) model (FitzHugh 1961; Nagumo et al. 1962).
In fact, the assumptions (i) and (ii) which are used to derive the reduced model (2.7)
(Krinskii and Kokoz 1973; Rinzel 1985) is essentially the same as the one used by
FitzHugh to derive the BVP model (FitzHugh 1961). The reduced model (2.7) by
Krinskii and Kokoz (1973) and by Rinzel (1985) is derived by some logical process
while the BVP model is derived a priori. Thus, the relation of physiological parame-
ters of the original and the simplified models are much clear in the model (2.7) rather
than in the BVP model. For more systematic reduction of general HH-type models,
see Abbott and Kepler (1990), Golomb et al. (1993), and Kepler et al. (1992).

2.3.5 Bifurcation Analysis

In this subsection, we consider the dependence of the HH equations’ behavior on the

parameter Iext ; the constant-current-transfer characteristic of the HH neuron (Rinzel
1978; Guttman et al. 1980). The neuron model produces an action potential with
response to an external pulsatile stimulus. We can expect that the neuron generates
action potentials persistently when a continuous current is applied externally.
Figure 2.14 shows the example of such repetitive firings (oscillation or periodic
orbit) of the HH neuron when Iext D 7. At t D 62, a pulsatile stimulus is applied to
the neuron in addition to the constant current injection. After the pulse stimulus, the
repetitive firing is stopped in spite of the persistent injection of the constant current.
This means that an repetitive firing (stable limit cycle) and a quiescent state (stable
equilibrium) coexist when Iext D 7. The state point of the HH neuron is moved from
one attractor (limit cycle) to an another one (equilibrium) by the external pulse.
We note that the applied pulse is depolarizing current; not only inhibitory input
but also excitatory input can inhibit such repetitive firing since it is a nonlinear
oscillation. The timing (phase) when the pulse is applied is important.

100 20

80 18
60 16
Iext ( mA/cm2)

40 v
v (mV)

14
20 12
0 10
Iext
–20
8
–40
6
–60
20 40 60 80 100
t (ms)

Fig. 2.14 Example of a membrane potential waveform of the HH equations when an external
constant current is applied: Iext D 7 A cm2 . A pulse is applied to the model in addition to the
constant current at t D 62
2.3 Nonlinear Dynamical Analysis of the Original HH Equations 53

a b
v (mV)
100. 125.

DC3 100. DC3

75.
75.
50.
50.
DC2
25. DC1 DC2
25. DC1
HB1 HB1
HB2
0. 0.
0. 50. 100. 150. 200. 0.0 5.0 10.0 15.0 20.0
25. 75. 125. 175. 2.5 7.5 12.5 17.5
2 2
Iext (µA/cm ) Iext (µA/cm )
c v (mV)
25.0
22.5
A PD2
20.0
DC2
17.5
15.0 PD1
DC1
PD3
12.5
10.0
7.80 7.85 7.90 7.95 8.00 8.05
Iext (µA/cm2)
d e
15.0 25.
12.5 DC1 20.
10.0 15.
7.5
10.
5.0
5. A
2.5
0.0 0.
–2.5 –5.
–5.0 –10.
0.00 0.20 0.40 0.60 0.80 1.00 0.00 0.20 0.40 0.60 0.80 1.00
0.10 0.30 0.50 0.70 0.90 0.10 0.30 0.50 0.70 0.90
t (ms) t (ms)

Fig. 2.15 (a) One-parameter bifurcation diagram of the HH equations. (b) Magnification of (a).
(c) Magnification of (b). (d, e) Membrane potential waveforms at the points DC1 and A of (c)

Figure 2.15a shows the dependence of the solution of the HH equations on the
parameter Iext ; the v values of the stationary solution of the HH equations are plot-
ted for various values of Iext where the maximum value of v is plotted for a periodic
(oscillatory) solution. Solid and dotted curves denote stable and unstable equilib-
ria, respectively. The filled (open) circles denote stable (unstable, resp.) periodic
solutions.
Panel (b) is the magnification of left part of (a) and we can verify the multi-
stability of an equilibrium and a periodic solution when Iext D 7 (ref. Fig. 2.14). At
the point HB2 of panel (a), a stable periodic solution bifurcates from a equilibrium
point by the (super-critical or stable) Hopf bifurcation. An unstable periodic solution
54 2 The Hodgkin–Huxley Theory of Neuronal Excitation

is bifurcated by the (sub-critical or unstable) Hopf bifurcation at the point HB1. As

is seen from panel (b), a multistability occurs near the sub-critical Hopf bifurcation.
At the point DC3, a double-cycle bifurcation or saddle-node bifurcation of periodics
occurs and a pair of stable and unstable periodic solutions is generated.
At both points DC1 and DC2, double-cycle bifurcations occur also. Near Iext D
7:9, four periodic solutions (one stable solution and three unstable solutions) coexist
(Rinzel and Miller 1980). The fact is that there are more coexisting (unstable) so-
lutions. Panel (c) is the magnification of the region near the points DC1 and DC2 of
(b). (Note that the periodic solutions are denoted by curves rather than circles.) At
the points PD1 and PD2, period-doubling bifurcations of unstable periodic solution
occur. Panel (d) shows the membrane potential waveform (only one-period length
of periodic solution is shown and abscissa is the time normalized by its period) of
such unstable periodic solution at the point DC1 of panel (c). Panel (e) is the similar
waveform of the period-doubled solution at the point A of panel (c).
We can not observe such unstable solutions in real experiments. The unstable
solutions, however, connect “the missing links” between stable solutions and much
help us to understand the total behavior of neuronal models. The bifurcation analysis
of Fig. 2.15 is made by the use of numerical bifurcation analysis software called
AUTO (Doedel et al. 1995). AUTO is very useful software which can detect several
bifurcation points automatically and can trace both stable and unstable branches of
equilibria and periodic solutions.
Figure 2.16a shows the period of the periodic solutions shown in Fig. 2.15a.
Panel (b) is the magnification of panel (a). The period of stable periodic solutions
(closed circle) varies in the range from several milliseconds to 20 ms. The period
does not change much totally although the variation is comparatively large in the
small Iext range.

a (ms)
b (ms)
30. 30.

25. 25.
DC3 DC2
20. 20. DC1
DC2
15. DC1 15. DC3

10. 10.

5. 5. HB1
HB1 HB2
0. 0.
0. 25. 50. 75. 100. 125. 150. 175. 6.0 7.0 8.0 9.0 10.0
6.5 7.5 8.5 9.5
Iext(µA/cm2) Iext(µA/cm2)
.

Fig. 2.16 (a) Period of the periodic solution shown in Fig. 2.15a and (b) Magnification of (a)
Chapter 3
Computational and Mathematical Models
of Neurons

What are models? The HH equations (2.3) are often called a physiological model,
whereas the models appeared in the following sections are simplified models or
abstract models. However, there is no model in which any simplifications or ab-
stractions have not been made. Of course, many features of real neurons are ignored
even in the HH equations. All models have their applicability and limits to describe
natural phenomena. Therefore, all types of models whatever simplified or physio-
logical, have their own values to model real phenomena. Starting with the BVP or
FHN model which is a simplification of the HH equations, this chapter proceeds
to several neuronal models with higher abstractions which are useful to tract some
essential features of neurons.

3.1 Phase Plane Dynamics in the Context of Spiking Neuron

Neurons generate action potentials or spikes as seen in the above chapter. In this
section, let us consider the one of the simplest models of such spike generation. The
Bonhoeffer–van der Pol (BVP) model is derived from the well-known van der Pol
equation by FitzHugh (1961) which mimics well the qualitative behavior such as
excitability, refractoriness, repetitive activity of the Hodgkin–Huxley model. The
BVP model is also called as the FitzHugh–Nagumo (FHN) model because the model
is equivalent to the electronic circuit with a tunnel diode proposed by Nagumo et al.
(1962).
The (space-clamped) BVP model is expressed by the following set of differential
equations:

xP D c.x  x 3 =3  y C Iext /; (3.1a)

yP D .x  by C a/=c; (3.1b)

where xP means dx=dt. The variable x denotes the membrane potential of a neuron,
y is a “recovery” variable which corresponds to the combination of NaC inactivation
and KC activation of the HH model (2.3), and Iext the current stimulus applied

S. Doi et al., Computational Electrophysiology, 55

DOI 10.1007/978-4-431-53862-2 3,  c Springer 2010
56 3 Computational and Mathematical Models of Neurons

externally. For the equations to mimic the behavior of a neuron, the parameters a, b
and c are set in a certain range (FitzHugh 1961; Nagumo et al. 1962; Hadeler et al.
1976), and the conventional values are a D 0:7, b D 0:8, c D 3:0.
In the following, we explore the neuronal dynamics of the BVP equations using
XPPAUT.

3.1.1 Excitability and Flows of the BVP Neuron Model

The following is the XPPAUT ode file of the BVP equations.

# bvp.ode
init x=1 y=0
x’ = c*(x - x^3/3 - y + Iext)
y’ = (x - b*y + a)/c
par Iext=0, b=0.8, c=3.0, a=0.7
@ xplot=t,yplot=x
@ total=100,dt=.03,xlo=0,xhi=10, ylo=-2,yhi=2
@ meth=runge-kutta
@ bound=200
@ autoxmin=0,autoxmax=1.8,autoymin=-3,autoymax=3
@ ntst=15 nmax=75 npr=200 parmin=-5 parmax=5 dsmax=0.1
ds=0. 1 done

In this ode file, not only the differential equations but also the initial (default)
values of several parameters for the execution of the numerical bifurcation analysis
by AUTO are described.
Figure 3.1 shows the x-waveforms of solutions of the equations with various
initial values: y value is fixed to zero, and x values are x D ˙1, ˙0:5, ˙0:4,
˙0:3, ˙0:2, ˙0:1, 0. We can see that the solutions are mainly classified to two
groups: Solutions with a positive initial value of x grow up and make a big wave
at first, then they converge to a stable equilibrium point while other solutions de-
crease without such a big wave, then they also converge to the equilibrium. The
variable x corresponds to the membrane potential of a neuron and the big wave cor-
responds to an action potential. Therefore, we can see that the solutions of the BVP
equations are very sensitive to their initial values of x. The initial value of x corre-
sponds to a current injection to a neuron; sufficiently large current injection to the
BVP neuron model can produce an action potential, while small current injection
cannot. These results demonstrate the excitability and the threshold characteristics
of the BVP model. Strictly speaking, there is an intermediate solution between the
solutions with and without action potential. This means that the threshold is not
strict or that the response of the BVP neuron model is not all-or-none. We may
call such a threshold a quasi-threshold.
Figure 3.1 can be obtained by using the XPPAUT as follows: Start the XPPAUT
as shown in Fig. 1.16, and select or open the bvp.ode file. Then the XPPAUT main
window will appear. Click on the ICs button which is the leftmost button on the top
3.1 Phase Plane Dynamics in the Context of Spiking Neuron 57

Fig. 3.1 Solutions of the BVP equations (3.1). Iext D 0. Initial values are y D 0, x D ˙1, ˙0:5,
˙0:4, ˙0:3, ˙0:2, ˙0:1, 0

Fig. 3.2 Initial Data

window to set initial values
of the BVP equations

row of the XPP main window. Then the window as Fig. 3.2 will appear. If we set
initial values and then click on (G)o, a waveform will be drawn. Iterations of this
process may produce the figure.
As already explained in Sect. 1.5, in order to create a phase plane which corre-
sponds to Fig. 3.1, click on the Viewaxes button and then 2D button, the 2D View
window will comes up. If we change the axes and their ranges to such values as
Fig. 3.3, then we can obtain the phase plane shown in Fig. 3.4.
To draw the vector field (or direction field) as Fig. 3.5, just click on the
Dir.field/flow and (D)irect Field buttons. Then, in the second row of
the XPP main window, a message such as “Grid:10” will appear. If you accept this
default value, just hit an enter (or return) key, if not accept, modify the value as
you want and hit enter.
58 3 Computational and Mathematical Models of Neurons

Fig. 3.3 The 2Dview window. We can change the variables and their range to be plotted in the
X- and Y-axes

Fig. 3.4 Phase plane and orbits of the BVP equations. Initial values are y D 0, x D ˙1, ˙0:5,
˙0:4, ˙0:3, ˙0:2, ˙0:1, 0

In Fig. 3.5, nullclines are also drawn. The x-nullcline is the curve such that
xP D 0. As seen from (3.1), this curve is the cubic curve: y D x  x 3 =3 C Iext .
The y-nullcline is the curve such that yP D 0 and thus the line x  by C a D 0.
Since xP D 0 (yP D 0) on the x- (y-) nullcline, trajectories cross the nullcline verti-
3.1 Phase Plane Dynamics in the Context of Spiking Neuron 59

Fig. 3.5 Phase plane, vector field and nullclines of the BVP equations. Initial values of orbits are
y D 0, x D ˙1, ˙0:5, ˙0:4, ˙0:3, ˙0:2, ˙0:1, 0

cally (horizontally, resp.). To draw the nullclines by XPP, just click on Nullcline
and (N)ew. The intersection point of x- and y-nullclines is the equilibrium point.
We can see that all trajectories approach the equilibrium point since this point is sta-
ble. We can verify this stability by the linearization and simple linear algebra (see
Sect. 1.3).
Figure 3.5 shows the (quasi-) threshold characteristics of the BVP model
again. We can see that the threshold exists near the “middle” branch of the cu-
bic x-nullcline. The middle branch is, in a certain sense, unstable and thus it makes
the sensitivity on initial values and the threshold characteristics, although we do
not explain this in detail. Anyway, the geometric analysis of a phase plane (phase
space) is very useful to analyze various neuronal characteristics of the BVP model.

3.1.2 Bifurcations in the BVP Neuron Model

Next, we proceed to the bifurcation analysis of the BVP equations. Let us restart
XPPAUT and choose the bvp.ode file again, then an XPP main window will appear.
If we click on the Initialconds button and then click on the (G)o button, a wave
60 3 Computational and Mathematical Models of Neurons

Fig. 3.6 XPP main window and the X vs. T plot of the BVP equations (3.1)

of X vs. T will be drawn as Fig. 3.6. After the integration (drawing of waveform)
finished, we click on both Initialconds and (L)ast buttons (several times) to
obtain enough convergence to a stable equilibrium point. We click on the File
button and then click on the Auto button, an AUTO window will appear.
In the AUTO window, if we click on Run button and click on Steady state
button, then a one-parameter bifurcation diagram will be drawn as Fig. 3.7. The ab-
scissa is the bifurcation parameter Iext and the ordinate is the x variable of the
BVP equations (3.1). The x-coordinate of the equilibrium points (steady states) of
the BVP equations are plotted as a function of the external stimulus current Iext .
Thick and thin curves shows stable and unstable equilibrium points, respectively.
As Iext increases, the stable equilibrium point becomes an unstable one at the point
labeled by 2. This is the Hopf bifurcation point. If the stimulus current Iext is fur-
ther increased, the unstable equilibrium point recovers its stability at another Hopf
bifurcation point labeled by 3.
In order to clarify what is happening between the two Hopf bifurcation points,
we trace solutions which may be bifurcated through the Hopf bifurcation point 2.
To do so, let us “grab” the point 2. If we click on Grab button, then a cross (cursor)
will appear on the one-parameter bifurcation diagram. We use the Tab (or arrow)
key on a keyboard to move the cursor on the Hopf bifurcation (HB) point labeled
by 2. Then we grab the point by hitting the enter or return key. Finally, if we
click on Run and Periodic buttons, the one-parameter bifurcation diagram on both
steady states and periodic orbits shown in Fig. 3.8 will be drawn.
3.1 Phase Plane Dynamics in the Context of Spiking Neuron 61

Fig. 3.7 One-parameter bifurcation diagram on steady states (equilibrium points) of the BVP
equations (3.1)

Fig. 3.8 One-parameter bifurcation diagram on both steady states and periodic orbits of the BVP
equations (3.1)

The magnification of this bifurcation diagram near the left Hopf bifurcation
point 2 is shown in Fig. 3.9. The open and closed circles denote unstable and
stable periodic orbits, respectively. Unstable periodic orbits are bifurcated from
62 3 Computational and Mathematical Models of Neurons

Fig. 3.9 Magnification of the one-parameter bifurcation diagram of Fig. 3.8

the Hopf bifurcation point 2: this Hopf bifurcation point is sub-critical. Those
unstable periodic orbits change their stability at the point labeled by 5 which is
called the double-cycle bifurcation or saddle-node bifurcation of periodic orbits.
The same phenomenon occurs near another Hopf bifurcation point 3 also. There-
fore, between two double-cycle bifurcation points 5 and 6, there exist stable periodic
orbits: the BVP model shows repetitive spiking there. Since both Hopf bifurcation
points are sub-critical, in the range of Iext between the double cycle bifurcation
point 5 and the Hopf bifurcation point 2 (and also between 3 and 6), the BVP model
shows multi-stability: both a stable equilibrium point and a stable periodic orbit co-
exist. In this parameter range, the final fate (state) of a solution depends on its initial
values.
These bifurcation phenomena are very similar to those of the HH model shown
in Fig. 2.15. Of course, there are slight differences between HH and BVP models.
In the HH model, the right Hopf bifurcation is super-critical while the left one is sub-
critical. In the BVP model, however, both Hopf bifurcations are sub-critical. This
difference is caused by the “symmetry” of the BVP equations (3.1); the equations
are composed by only terms of odd-numbered power. If we add some terms of
even-numbered power such as x 2 , the “stabilities” of two Hopf bifurcations can
be different. We can say that, in spite of the difference, the very simple BVP model
well mimics the neuronal behavior of the HH model.
In order to make the magnification shown in Fig. 3.9, click on Axes and hI-lo
buttons, and change the scales in the window as shown in Fig. 3.10. Then, the click-
ing on Clear and reDraw buttons will give us Fig. 3.9.
Next, let us explore more detailed behavior of bifurcations of the BVP
model (3.1). In particular, we perform the two-parameter bifurcation analysis.
3.1 Phase Plane Dynamics in the Context of Spiking Neuron 63

Fig. 3.10 The AutoPlot

window for the scale change

Fig. 3.11 The Parameters

window for the change of
parameter values

To do so, let us “restart” the XPP and choose the bvp.ode file again (Note that we
do not actually need to restart the XPP. We can continue the analysis. But the restart
is much simpler than the continuation). At first, in the XPP main window, we click
on Param button and change the default value of b to 2, as shown in Fig. 3.11. If we
click on Initialconds and (G)o buttons, and subsequently Initialconds and
(L)ast buttons (a few times), then Fig. 3.12 will be drawn.
We click on File and Auto buttons to show a AUTO window. Then, if we click
on Run and Steady state buttons, a one-parameter bifurcation diagram will be
drawn as Fig. 3.13. The x-coordinates of equilibrium points are plotted as a function
of Iext . Differently from the previous bifurcation diagram of Fig. 3.7 where b D 0:8,
the curve of equilibrium points is not monotonic increasing. Two points labeled by
3 and 4 at the “knees” of the S-shaped curve, are the saddle-node bifurcation points.
Therefore, in the range of Iext between the two points, there are three equilibrium
points (two stable and one unstable equilibrium points), whereas only one stable
equilibrium point exists in the other range. There are also two Hopf bifurcation
points labeled by 2 and 5 which locate near the saddle-node bifurcation points.
Similarly to the previous analysis, let us trace periodic solutions which may bi-
furcate from both Hopf bifurcation points 2 and 5. To do so, clicking on Grab button
and using tab key, we grab the Hopf bifurcation (HB) point with label 2. Then, we
64 3 Computational and Mathematical Models of Neurons

Fig. 3.12 X vs. T plot of the BVP equations (3.1)

Fig. 3.13 One-parameter bifurcation diagram on steady states of the BVP equations (3.1) with
different parameter value

click on Run and Periodic buttons to obtain a branch of periodic orbits bifurcated
from the Hopf bifurcation point 2. If we repeat these operations (Grab the point 5,
and so on) for another Hopf bifurcation point 5, we can obtain the bifurcation
3.1 Phase Plane Dynamics in the Context of Spiking Neuron 65

Fig. 3.14 One-parameter bifurcation diagram on both steady states and periodic orbits of the BVP
equations (3.1)

diagram on both equilibrium points and periodic orbits as shown in Fig. 3.14. Note
that the two branches of periodic orbits from two points 2 and 5 are very short. The
reason of these short terminations of branches of periodic orbits may be clarified by
the following two-parameter bifurcation analysis.
In order to investigate how the locus of the Hopf bifurcation changes when the
other parameter value is changed, we make two-parameter bifurcation diagrams.
To do so, we Grab the Hopf bifurcation point 2 again. Then we click on Axes
and Two par buttons. A new window as shown in Fig. 3.15 will appear, where
we change as XMIN=-1, YMIN=-1, XMAX=2, YMAX=4. If we click on Run and
Two param buttons, a (part of) two-parameter bifurcation diagram will be obtained
as Fig. 3.16, where both axes denote the bifurcation parameters: the abscissa is Iext
and the ordinate is another parameter b of the BVP equations (3.1). The short curve
is the Hopf bifurcation curve which denotes the loci of the Hopf bifurcation. If the
two parameters Iext and b are chosen on the Hopf bifurcation curve, then a Hopf
bifurcation occurs. Namely, the two-parameter bifurcation diagram illustrates how
the locus of the Hopf bifurcation changes with the change of two parameters.
The Hopf bifurcation curve in Fig. 3.16 is not complete. In order to complete the
Hopf bifurcation curve, we trace the Hopf bifurcation curve in “another direction.”
To do so, we click on Numerics button and change the value of Ds to -0.1 as
shown in Fig. 3.17. Then, we again Grab the Hopf bifurcation point 2 of Fig. 3.13.
Note that, although we cannot see the Hopf bifurcationpoint 2 on the present AUTO
66 3 Computational and Mathematical Models of Neurons

Fig. 3.15 AutoPlot window

Fig. 3.16 A part of two-parameter bifurcation diagram on Hopf bifurcation

window, we can move to the Hopf bifurcation point (by using tab key) following
the information below the bifurcation diagram, where several information including
the Hopf bifurcation label 2, will be shown. Then, if we hit enter or return key,
we can grab the bifurcation point labeled by 2. Finally the clicking on Run and
Two param buttons gives us the whole two-parameter bifurcation curve of Hopf
bifurcation as shown in Fig. 3.18.
The bifurcation curve in Fig. 3.18 is the curve traced from the Hopf bifurca-
tion point 2 of Fig. 3.13. Let us draw another bifurcation curve which started from
3.1 Phase Plane Dynamics in the Context of Spiking Neuron 67

Fig. 3.17 AutoNum window

Fig. 3.18 Whole two-parameter bifurcation diagram on Hopf bifurcation

another Hopf bifurcation point 5 of Fig. 3.13. To do so, we Grab the point 5
following the information below the diagram, similarly to the above case. Then, if
we click on Run and Two param buttons, then “a part of another” Hopf bifurcation
curve will appear. Completely similarly to the above case, we change the tracing
direction by changing the Ds value back to 0.1 (at first, click on Numerics button).
Then, if we Grab the Hopf bifurcation point 5 and click on Run and Two param
buttons, then whole bifurcation curve of another Hopf bifurcation can be obtained
as Fig. 3.19.
68 3 Computational and Mathematical Models of Neurons

Fig. 3.19 Whole two-parameter bifurcation diagram on “another” Hopf bifurcation

In Fig. 3.19, the two Hopf bifurcation curves terminate at the points labeled by
9 and 11. These terminations are strange, and thus we clarify the reason of the
termination by making bifurcation curves of the other bifurcation. Let us make a
two-parameter bifurcation diagram on the different bifurcation: the saddle-node bi-
furcation labeled by 3 in Fig. 3.13. The computational process is completely the
same as that of Hopf bifurcation: main ingredients are the twice executions of Grab
of point 3 and the “Run C Two param” with changing the tracing directions by the
Ds value change. This process gives us the final two parameter bifurcation diagram
on both Hopf and saddle-node bifurcations as shown in Fig. 3.20. Note that in order
to obtain the saddle-node bifurcation curve, we do not need to Grab another saddle-
node bifurcation point labeled by 4 in Fig. 3.13. This is because the saddle-node bi-
furcation curve is a connected single curve, differently the Hopf bifurcation curves.
In Fig. 3.20, there are special points labeled: the point 14 is a special bifurcation
point called the cusp point. The two Hopf bifurcation curves terminate on the saddle-
node bifurcation curve at two points labeled by 9 and 11 which are also special
bifurcation points called the Bogdanov–Takens bifurcation point. In both cases of
the cusp and Bogdanov–Takens bifurcation points, the equilibrium point of the BVP
equations (3.1) possess double zeroes as eigenvalues. However, the behavior of the
BVP model may be different between the two special bifurcation points.

Exercise 3.1. Choosing several sets of parameter values of b and Iext on the two-
parameter bifurcation diagram of Fig. 3.20, explore the dynamics of the BVP model
on x–y phase plane. What is the difference from the BVP model with default
parameter values? Consider geometrically what makes these differences if any. 
3.2 Simple Models of Neurons and Neuronal Oscillators 69

Fig. 3.20 Final two-parameter bifurcation diagram on both Hopf and saddle-node bifurcations

As we have explored so far, one- and two-parameter bifurcation analyses

extremely help us to understand the whole behavior of the BVP neuronal model
which cannot be obtained by the numerical integrations (simulations) of the BVP
equations (3.1) solely. For the more detailed explorations on the dynamics and bifur-
cations of the BVP or FHN equations, see Braaksma (1993, 1998), Guckenheimer
(1986), Koper (1995), Nomura et al. (1994a), and Rocşoreanu et al. (2000).

3.2 Simple Models of Neurons and Neuronal Oscillators

In this section, let us consider very simple or formal models of neurons and neuronal
oscillators (pacemakers) and examine the input–output (IO) translation characteris-
tics of these models. Surprisingly, we will show that these simple models arisen
in different contexts are all systematically described by a simple one-dimensional
discrete-time dynamical system (difference equation) using a concept of the phase
transition curve (PTC), although the IO characteristics (or response characteristics)
of these simple models are very complicated.

3.2.1 Integrate-and-Fire Neuron

Integrate-and-fire (IF) neuronal model is the simplest model in which a neuron is

considered to be a single switch with a membrane capacitance and is a special case
of the so-called leaky integrator treated in the next subsection (Rescigno et al. 1970;
Stein et al. 1972; Glass and Mackey 1979, 1988). The capacitor integrates input
70 3 Computational and Mathematical Models of Neurons

stimuli and if a membrane potential reaches a threshold, the switch is closed and
the discharge of the capacitor occurs instantaneously and this process is repeated.
The dynamics of the IF neuron model is described by

dv
D  C I.t/; if 0  v  ; (3.2a)
dt
v.t C / D 0; if v.t/ D ; (3.2b)

where v is the membrane potential,  the constant bias of an external (synaptic)

input to the neuron, I.t/ the time-dependent component of the external input, and 
the firing threshold. The resting potential is set to zero without loss of generality.
The IF neuron generates an action potential if the membrane potential v.t/ reaches
the threshold  and then v.t C / is reset to zero and this process is repeated. Note that
if  is considered as a certain internal attribute rather than a bias of the external
input, the IF neuron model fires repeatedly without the external input I.t/ and then
the IF neuron is considered as a model of a neuronal oscillator or a pacemaker cell.
Consider the case that the external input I.t/ is a periodic pulse trains with period
T and amplitude A: 1
X
I.t/ D Aı.t  kT /; (3.3)
kD0
where ı is a Dirac’s delta function; if a single pulse is injected, the membrane
potential v is increased by A immediately (if v exceeds the threshold  by the pulse,
the neuron emits an impulse or a spike, and then v is reset to zero):
(
C v.t/ C A if 0  v.t/ <   A
v.t / D .t D 0; T; 2T; 3T; : : :/:
0 if   A  v.t/  

Figure 3.21 shows an example of a phase locking of the IF neuron (3.2) to

the periodic stimulus (3.3). Thick lines in the upper trace shows the value of mem-
brane potential v.t/ as a function of t and the lower trace shows the periodic pulsatile

firing
v
θ

Fig. 3.21 Phase locking

of the integrate-and-fire
neuron model.  D 1,
 D 0:18, A D 0:2, T D 1.
Thick lines of the upper trace 0 5 10 t
shows the value of membrane
potential v.t / as a function of
t and the lower trace shows
the periodic pulsatile stimulus t
3.2 Simple Models of Neurons and Neuronal Oscillators 71

stimulus. The initial value v.0/ is zero and the first pulse is injected at t D 0;
i.e. v.0C / D A D 0:2. The IF neuron generates the first action potential (fires) a
short time later the third pulse input (t  2:2) and also fires at t D 5 by the sixth
pulse input. After this firing, the variation of v.t/ becomes periodic generating one
action potential for every three input pulses: 3:1 phase locking occurs. The mecha-
nism of phase lockings of the IF neuron to pulse stimuli will become clearer if we
observe or sample the v.t/ value at t D 0T , 1T , 2T , . . . , as follows.
Suppose that vn denotes the value of membrane potential v.t/ immediately after
the n-th pulse injection and vnC1 that of (n C 1)-th pulse. Then we can obtain the
relation between vnC1 and vn as follows:
(
vn C A C T mod  if 0  f.vn C T / mod g <   A ;
vnC1 D p.vn / 
0 if   A  f.vn C T / mod g   :
(3.4)

If we otherwise consider vn as the value of the membrane potential immediately

before the n-th pulse injection, then the above relation should be replaced by the
following:
(
vn C A C T mod  if 0  vn <   A ;
vnC1 D p.vn /  (3.5)
T mod  if   A  vn   :

If we consider the value of v as the “phase” of the neuron, the maps p.v/ of (3.4) and
(3.5) describe the transition rule of the phase and thus is called the phase transition
curve (PTC) (Kawato and Suzuki 1978; Winfree 1980; Kawato 1981).
In order to study the phase-locking patterns of the IF neuron subjected to periodic
pulsatile inputs, it is sufficient to explore the dynamics of the one-dimensional map
(circle map) of (3.4) or (3.5). Figure 3.22a,b show the examples of (3.4) and (3.5),

a b

vn+1 vn+1

0 v0 vn 0 vn
v0 v6 v7 v5v8

Fig. 3.22 PTC of the IF neuron. A D 0:2, T D 0:18 . Panels (a) and (b) are different each
other only by their coordinates, and correspond to (3.4) and (3.5), respectively
72 3 Computational and Mathematical Models of Neurons

respectively. Thick lines are the graph of p.v/ and thin lines show an orbit. Both
(a) and (b) have the essentially same dynamics: The difference is just the choice of
coordinate (the value of v.t/ after or before the input). Let us explain the dynamics
of the map p.t/ using Fig. 3.22b. The orbit fvn g of Fig. 3.22 corresponds to the
solution v.t/ shown in Fig. 3.21. The initial value v0 of the orbit in Fig. 3.22b is zero
since vn is the value of v.t/ immediately before the pulse injection (see Fig. 3.21).
As stated above, the IF neuron fires by the sixth pulse injection at t D 5T which
corresponds to v5 in Fig. 3.22b. After this firing, the orbit becomes periodic: v6 , v7 ,
v8 , v9 D v6 , v10 D v7 , . . . . In Fig. 3.22a, the coordinate or phase vn is the value
of v.t/ immediately after a pulse injection and thus the initial value of orbit fvn g of
Fig. 3.22a is 0.2 while it is zero in (b). The dynamics of (a) is completely same as (b).
The main part of the maps of Fig. 3.22 is just the shift of vn by A C T (linear
part with a unit slope) and such a linear map hardly produces any phase-locking
without A C T being a rational number which is a rare case. The flat (zero-slope)
part of the map, however, can produce a phase locking for almost all initial val-
ues and parameter values of A and T except for special values in a measure zero
set (Doi 1993). Note that a zero slope corresponds to a super stability as stated in
Sect. 1.4.1. See the fact that the orbit fvn g becomes periodic after it passes the flat
segment in Fig. 3.22. In terms of neurons, the existence of threshold in the IF neuron
produces phase lockings.

3.2.2 Leaky Integrate-and-Fire Neuron

Next, we take into account a (linear) resistivity of a membrane in addition to the

membrane capacitance of the IF neuron. Then, we obtain a so-called leaky integrate-
and-fire (LIF) neuron:

dv v.t/
D C  C I.t/; if 0  v  ; (3.6a)
dt

v.t C / D 0; if v.t/ D : (3.6b)

Only the first term of the r.h.s. of (3.6a) is different from the IF neuron (3.2). Simi-
larly to the IF neuron, if we consider the periodic pulsatile input (3.3), we have the
PTC of the LIF neuron:
(
ˆ.vn C A; T / if 0  vn <   A ;
vnC1 D p.vn /  (3.7)
ˆ.0; T / if   A  vn   ;
3.2 Simple Models of Neurons and Neuronal Oscillators 73

where ˆ.x0 ; t/ denotes the solution of the LIF neuron at time t with an initial value
x0 and without the periodic stimuli:
8 
 x0
ˆ
<.x0  
/e t = C 
if 0  t 
ln ;
ˆ.x0 ; t/ D 
 
 = 
 x0
:̂
e t C 
if t >
ln ;

 
   
 
 x0 

t D t 
ln mod
ln :

  
 

The phase vn is the value of the membrane potential immediately before the pulse
input and we supposed 
>  (the LIF neuron can repetitively fire without the
pulse-train stimuli).
Figure 3.23 shows an example of the PTC and its orbit of the LIF neuron. Simi-
larly to the IF neuron, the PTC consists of three linear segments. The different aspect
of the LIF neuron from the IF neuron is the slope of the two segments other than the
flat (zero-slope) segment. The slope of the left most segment is less than unity which
means that the introduction of the membrane resistivity makes the LIF neuron more
stable (easier to phase-locked) than the IF neuron. The effect of resistivity, however,
is not always the stabilization of the neuron: The slope of the middle segment of
the three is greater than unity which means that the resistivity sometimes makes a
neuronal dynamics more unstable. The LIF neuron, however, is totally very stable
and easy to be phase-locked to periodic inputs owing to the flat segment (super sta-
bility). See the fact that the orbit in Fig. 3.23 with an initial value v0 D 0 becomes
periodic (phase locked) after its passage (v6 ) of the flat segment of the map. The flat
segment of the PTC is generated by the existence of a firing threshold (and a reset
to the resting potential) similarly to the case of IF neuron.
In this section, we have considered the LIF neuron model stimulated by peri-
odic pulse trains. There are, however, vast literatures on the forced LIF neuron: e.g.
(to name just a few) the LIF neuron forced by a sinusoidal input (Rescigno et al.
1970; Scharstein 1979; Keener et al. 1981) and the one forced by the more general

vn+1
Fig. 3.23 PTC of the LIF
neuron.  D 0:5,
D 4:0,
A D 0:2, T D 0:5. Note that,
differently from the IF neuron
shown in Fig. 3.22, the slopes
of the left and middle linear
segments are less than and
greater than unity, 0 vn
v0 v6 
respectively
74 3 Computational and Mathematical Models of Neurons

periodic input (Bulsara et al. 1996; Torikai and Saito 1999; Nakano and Saito 2002;
Pakdaman 2001). Also see Mirollo and Strogatz (1990) for the synchronization
behavior of mutually coupled LIF neuronal oscillators.

3.2.3 Radial Isochron Clock as a Model of Pacemaker Cells

When the parameter  is considered as an intrinsic aspect of the IF neuron, the IF

neuron is considered as a model of pacemaker cells or biological clocks. Such a
clock, so to say, corresponds to a sand glass (just an integration or accumulation
of a charge). The simplest model which is known as a model of the other type of
clocks which correspond to a pendulum, is the so-called radial isochron clock (RIC)
or Poincaré oscillator (Guevara and Glass 1982; Hoppensteadt and Keener 1982;
Keener and Glass 1984; Glass and Mackey 1988; Glass and Sun 1994; Nomura
et al. 1994b):

dr
D ar.1  r/; a>0 (3.8a)
dt
d
D 2 (3.8b)
dt

which are expressed in the polar coordinate. These equations possess a unique limit
cycle with a radius r D 1 and with a unit period. Figure 3.24a shows the phase
plane of (3.8) in Cartesian coordinate. (Do not confuse the “phase” of the phase
plane with the phase of the above subsections.) The unit circle is the stable limit
cycle and solution curves started from any points of the phase plane (except for
the origin) asymptotically approach the limit cycle. The parameter a controls the
stability of the limit cycle (the strength of the attraction of the solution curves by
the limit cycle). The reason that this oscillator is called the RIC is that its isochrons
(a set of state points in the phase plane which asymptotically converge to the same
point on the limit cycle) are radial (Guckenheimer 1975).
Consider the RIC (3.8) stimulated by a periodic pulse trains (3.3) where a sin-
gle pulse instantaneously shifts the state point in Fig. 3.24a in a right horizontal
direction by an amount A. In the absence of such external stimuli, we can consider
that the state point moves in the neighborhood of the stable limit cycle (attractor)
and thus the state of the RIC oscillator can be denoted by a phase  (do not confuse
this term with a phase of the phase plane). If a pulse is injected, the state point leaves
the limit cycle by the pulse for a moment. After an enough long time, however, the
state point comes back to the neighborhood of the limit cycle again and thus the
state of the RIC oscillator can be denoted by the phase  (the angle  in the polar
coordinate). Therefore, similarly to the case of the IF neuron with periodic stimuli,
let us denote the phase of the RIC immediately before the n-th pulse by n and that
3.2 Simple Models of Neurons and Neuronal Oscillators 75

Pulse
θ stimulus

b c
1 1

n+1 n+1

0 n 1 0 n 1

Fig. 3.24 (a) Phase plane of the radial isochron clock (RIC). (b, c) PTC of the RIC. Parameter
values are: (b) A D 0:9, T D 0:1; (c) A D 1:1, T D 0:5

of .n C 1/-th pulse by nC1 . From a simple calculation, we can obtain the relation
(PTC of the RIC) between nC1 and n as follows:

sin n
nC1 D p.n /  arctan C 2T mod 2: (3.9)
A C cos n

Figure 3.24b,c show the examples of the PTC p./ of the RIC. Panel (b) corre-
sponds the case that the amplitude of the external pulse is less than the radius of the
limit cycle (A < 1) and (c) the case of A > 1. The map of (b) and (c) are continu-
ous maps on S with degree unity and zero, respectively (see Sect. 1.4.1). The phase
locking of the RIC can be studied by examining the one-dimensional map (3.9) sim-
ilarly to the case of the IF neuron. The map of (b) has the similar dynamics to that
of the sine-circle map (1.46) with small b values. The map of (c) has a typical fea-
ture as a bimodal map with two increasing segments and one decreasing segment,
and the characteristics of this bimodal map or the RIC will be clarified later. The
essential difference between the phase-locking behavior of the RIC and that of the
IF neuron is that RIC can produce chaotic behavior which is not appeared in the IF
neuron (we can easily expect chaotic behavior in the RIC from the bimodal feature
of PTC p./ of the RIC).
76 3 Computational and Mathematical Models of Neurons

3.2.4 Caianiello’s Neuron Model

So far, we have considered the simple neuron models expressed by differential

equations and reduced them to difference equations using PTC. In this subsection,
we will consider the simple neuron models expressed by difference equations.
Caianiello’s neuron model is described by (Caianiello 1961):
" #
X
n
xnC1 D 1 S.n/  ˛ b k xnk   ; (3.10)
kD0

where 1Œx D 1 (x 0), D 0 (x < 0). The neuron’s state xn at a discrete time n
takes 1 or 0 which correspond to the firing state and non-firing state of a neuron,
respectively. S.t/ is an input to a neuron and  is the threshold of a firing. The
summation term of the r.h.s. of (3.10) denotes the refractoriness which depends on
the past firing fxnk g. If the neuron fired before (xnk D 1), the summation term
becomes negative and thus the neuron must have a bigger input S.n/ for a firing at
the next time n C 1. The parameter b.> 1/ controls the decay rate of refractoriness
depending on the past state fxnk g. The unit step function 1Œx corresponds to the
threshold property of neurons and thus the Caianiello’s neuron model is an abstract
neuron model which incorporates only two characteristics of threshold property and
refractoriness of a neuron.
Nagumo and Sato (1972) investigated the Caianiello’s neuron model in detail.
When a constant input (S.n/  A) is applied to the neuron, changing the variable
xn to yn as follows:
Xn
yn D .A  /=˛  b k xnk (3.11)
kD0

(3.10) is reduced to

ynC1 D p.yn /  yn =b  1Œyn  C a; (3.12)

where we have set as a  .A  /.1  1=b/=˛. Again, we have obtained a one-

dimensional map p.y/ of (3.12). Figure 3.25a is an example of the map p.y/ which
is a piecewise linear map with two increasing segments with a slope 1=b. Whole
dynamics of such a piecewise linear map is well known (Leonov 1959; Nagumo
and Sato 1972; Hata 1982): the map has a stable periodic orbit for almost all pa-
rameter values (except for a measure zero set). As is seen from the definition of the
variable yn , the right branch (yn > 0) and the left branch (yn < 0) correspond to a
firing and a non-firing, respectively. In Fig. 3.25a, an orbit fyn g with y0 D 1 is also
drawn. This orbit asymptotically converges to a periodic orbit fy1 , y2 , y3 , y4 , y5 ,
y1 , y2 , . . . g which corresponds to a firing pattern 10101 since y1 , y3 and y5 pass
the right branch and y2 and y4 pass the left branch (where a firing and non-firing
are denoted by the symbols “1” and “0”, respectively).
3.2 Simple Models of Neurons and Neuronal Oscillators 77

a b
1 1
y4*
y2*

yn+1 yn+1

0 y5* 0
y3*
y1*

–1 0 1 –1 0 1
yn yn

Fig. 3.25 (a) Caianiello’s neuron model (a D 0:7, b D 2). (b) Chaos neuron model ( D 0:05,
a D 0:5, b D 2)

a b
1.0 0.6

g g

0.0 0.5
–0.2 a 1.2 0.666 a 0.680

Fig. 3.26 Mean firing rate (MFR) of the Caianiello’s neuron model (3.12) as a function of the
parameter a. The other parameter is set as b D 2. (b) is the magnification of a portion of (a): a
similar structure to (a) is nested between the steps of (a). This nesting structure (fractals) continues
infinitely and constitutes a Cantor function or devil’s staircase (zero differential coefficient almost
everywhere, non-decreasing and continuous function)

Let us define a mean firing rate (MFR) for the solution orbit fyn g of the
one-dimensional map (3.12) as follows:

1X
n1
 D lim 1Œyk : (3.13)
n!C1 n
kD0

The MFR of Fig. 3.25a is 3/5 since its firing pattern is 10101. Figure 3.26 shows the
MFR  as a function of the parameter a which corresponds to the amplitude A of
the input to the Caianiello’s neuron. Panel (b) is the magnification of a portion of
(a); between the steps or stairs of (a), many small stairs are nested and this nesting
structure continues infinitely. The complicated structure (fractals) of the MFR is
mathematically a Cantor function (zero slope almost everywhere, non-decreasing
and continuous function) and also is called a devil’s staircase. Mathematical ver-
ification of this complicated structure is not so difficult, see Leonov (1959) or
78 3 Computational and Mathematical Models of Neurons

Mira (1987) for detail. Note that the MFR  is equivalent to the rotation number 
defined in Sect. 1.4.1 (Guckenheimer and Holmes 1983) although the rotation num-
ber was defined for continuous maps in Sect. 1.4.1 and the map of Fig. 3.25a is not
continuous.

3.2.5 Chaotic Neuron Model

In the Caianiello’s neuron model, a unit step function is used for the representa-
tion of a threshold property of neuronal firings. The real neurons and also realistic
neuronal models such as the Hodgkin–Huxley equations, however, do not possess
such a rigorous threshold (i.e. not all-or-none) and do possess intermediate firing
states (neither firing nor non-firing). Aihara et al. (1990) incorporate this property
into the Caianiello’s neuron model. They replaced the step function by the following
sigmoidal (or logistic) function:

1
f .x/ D : (3.14)
1 C exp.x=/

In the limit of  ! 0, this function converges to the step function. Similarly to the
Caianiello’s neuron model, we can obtain a one-dimensional map:

ynC1 D p.yn /  yn =b  f .yn / C a: (3.15)

Figure 3.25b shows an example of the graph of p.y/. The Caianiello’s neuron
model of Fig. 3.25a has two discontinuous segments. In the map of Fig. 3.25b, the
two segments are connected continuously and the map has a typical bi-modal struc-
ture and thus can produce chaotic behavior differently from the Caianiello’s neuron
model. Thus, this neuron model is called the chaotic neuron model. (Real neurons
and the HH equations also present a chaotic behavior, Aihara et al. 1984; Matsumoto
et al. 1984; Takahashi et al. 1990.) For the chaotic neuron model, we can (numer-
ically) calculate the MFR similarly to the Caianiello’s neuron model (3.10). The
graph of MFRs of the chaotic neuron model, however, does not constitute the com-
plete devil’s staircase but does constitute the partial or incomplete staircase owing
to the difference of the topological feature of the two maps shown in Fig. 3.25a,b
(Yellin and Rabinovitch 2003).
Comparing Figs. 3.25b and 3.24c, we can see the similarity of topological fea-
tures of the two graphs although their functional forms are different. Thus, it is
apparent that two neuronal models (RIC and chaos neuron model) arisen in different
contexts and with different intrinsic dynamics have the essentially same phase-
locking or input–output characteristics. This is a very advantage of the usage of
such simplified or abstract models.
3.3 A Variant of the BVP Neuron Model and Its Response to Periodic Input 79

3.3 A Variant of the BVP Neuron Model

and Its Response to Periodic Input

3.3.1 Piecewise-Linearized BVP Model

In the following, we consider a variant of the BVP model (3.1):

 xP D f .x; y/  y  x C 5=6fjx C 1j  jx  1jg; (3.16a)

yP D g.x/  x C a; (3.16b)

where the cubic polynomial in the original BVP model (3.1a) is approximated by
a piecewise linear function and the parameters are also simplified. Note that the
externally applied constant current Iext of (3.1) is now considered to be included
in the new parameter a of (3.16) (by a change of variables). Also, note that the
sign of the variable x in (3.16) is reversed from the previous BVP model (3.1); the
negative direction of x or the left direction on the x–y phase plane corresponds to
the depolarization or the excitation of a neuron. We consider that this simplified
BVP model still preserves the neuronal features of the original BVP model and
of the HH model although this equation is much simpler than (3.1). Throughout the
following sections of this chapter, (3.16) will be referred to as the BVP model unless
otherwise specified in the text.
Let us explain the “neuronal” features of the BVP model (3.16). Figure 3.27
shows the x–y phase plane of (3.16). The N-shaped line is the x-nullcline on which
xP D 0 and orbits (solutions) of (3.16) move vertically, and the vertical line is the
y-nullcline (yP D 0). So the intersection point P of the two nullclines is an equilib-
rium or a fixed point of (3.16) which corresponds to the resting state or the quiescent
state of a neuron. Typical orbits .x.t/; y.t// with different initial values of (3.16)

y
1.5
Absolutely Refractory

1.0 .
y =0
ry
e

to
tiv

0.5 c
fra
Ac

e
R
0.0 ely
tiv
ela
Regen R
–0.5 . er ative
x =0
Resting Point P
QTP Separatrix
–1.0
–2 –1 0 1 2 x

Fig. 3.27 The x–y phase plane of (3.16) with the N-shaped x-nullcline (xP D 0) and the linear
y-nullcline (yP D 0). The parameters are  D 0:1, a D 1:1
80 3 Computational and Mathematical Models of Neurons

are also displayed. If the neuron is in its resting state and a pulse-like current is
injected, the neuron responds and settles down back to the resting state. For example,
an orbit starts at the equilibrium or the resting point P, and a large single-pulse
perturbation in the negative (depolarizing) direction of the x axis displaces it left-
wards, it goes through both the regenerative and active regions, and finally comes
back to P. (Note that, in the modified BVP equations (3.16), the negative direction
of x corresponds to the depolarization or the action potential differently from the
original BVP equations (3.1).) If the perturbation is small, the orbit does not enter
the regenerative region, i.e. the BVP neuron model is not excited. A phase-plane
analysis is very useful so that we can get the intuitive understanding of the total be-
havior of the model without need to make brute-force computer simulations of the
model with many different initial values.
The response of the BVP model to a single pulse perturbation is not all-or-none
(living neurons are not all-or-none also). There are neutral orbits which we cannot
decide whether they excite or not. The orbits around QTP (quasi-type) separatrix are
such type’s orbits. In this region, the orbit changes its amplitude drastically (but con-
tinuously) depending on its initial value. Thus, the QTP separatrix is considered as
a (quasi-) threshold of the neuron model. Discussions on the geometric property of
this threshold curve and on the excitability of the BVP neuron model (3.1) are made
by Okuda (1981). The time waveforms “x.t/” of the orbits shown in Fig. 3.27 are
plotted in Fig. 3.28a. Depending on the strength of the input pulse current, various
amplitude of responses are generated.
Figure 3.28b shows a response characteristics of the BVP neuron model to a
single pulse stimulus. The abscissa shows the amplitude of such a depolarizing
stimulus, and the ordinate shows the response amplitude (the maximum values of
“x.t/”) of the neuron model. In the case of  D 0:1 (solid curve), the response
amplitude grows rapidly in the range [0.3,0.4] of the abscissa. So the BVP model is
considered to have a certain (quasi-) threshold property. If the parameter  decreases
to 0.05 (dashed curve), this threshold property becomes more strict; the response
amplitude changes abruptly (but still continuously) near a value 0.28. In the limit
case of  D 0, the BVP model responds ideally (all-or-none) and the response char-
acteristics becomes a step function.
Let .x  ; y  / D .a; a  5=3/ denote the resting point P. This equilibrium point
is stable when jaj > 1 and unstable when jaj < 1. In the latter case, (3.16) has a
stable limit cycle and the BVP model excites repetitively. So, we can consider the
BVP model (3.16) with jaj > 1 a neuron (excitable cell) model and the one with
jaj < 1 a neuronal oscillator (pacemaker cell) model.
A neuron shows a repetitive activity when an external constant current is ap-
plied to. Mathematically this repetitive activity corresponds to a limit cycle solution
of a model. The HH model and the original BVP model (3.1) have such a limit cy-
cle which is “bifurcated” from an equilibrium point (a resting state) by the Hopf
bifurcation (Hadeler et al. 1976; Hassard 1978; Rinzel 1978). Strictly speaking, no
such a bifurcation occurs in the piecewise-linearized BVP model (3.16) since the
right hand side (r.h.s) of (3.16) at the equilibrium point P is not differentiable at the
critical parameter value a D 1. This difference between (3.1) and (3.16) does not
3.3 A Variant of the BVP Neuron Model and Its Response to Periodic Input 81

a
1
–x(t)
0

–1

–2

0 1 2 3 4 5
t
b
1.5
Response amplitude

excitation
1.0

0.5

0.0

–0.5
no excitation
–1.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Stimulus amplitude

Fig. 3.28 (a) Response waveforms x.t / to various amplitude of single pulses. The time wave-
forms of various orbits which are shown in Fig. 3.27 are plotted. (b) Response characteristics of
the BVP model (3.16) to a single pulse input, with parameter values  D 0:1 (solid curve) and
 D 0:05 (dashed curve). The amplitudes (the maximum values) of x.t / are plotted against the
amplitudes of single pulse inputs

affect the total behavior of the model: The model (3.16) has a stable equilibrium
point when jaj > 1 and has both an unstable equilibrium point and a stable limit
cycle when jaj < 1, although a slight delicate phenomenon occurs at the critical
value jaj D 1. The amplitude of the limit cycle grows as the value of jaj decreases
away from unity. This qualitative behavior of (3.16) is the same as that of (3.1).
Figure 3.29 shows a repetitive activity of the BVP model with different parameter
values: a D 0:0 (dashed curve) and a D 0:9 (solid curve). As the value of a in-
creases, the asymmetry of the waveform increases and the depolarized (x.t/ > 0/
fraction of time in one cycle (period) decreases. This fraction of time is called an
activity of a neuron model and the importance of asymmetry is already noted in the
context of coupled neuronal oscillators (Kepler et al. 1990; Meunier 1992). In the
figure, the parameter  is set as much smaller than one used in Figs. 3.27 and 3.28.
As the value of  approaches zero, the waveform of x.t/ loses its smoothness.
As stated above, the parameter a may be considered as an “external” current ap-
plied to a neuron. If we regard a as an “intrinsic” property of a neuron, then the BVP
82 3 Computational and Mathematical Models of Neurons

1
−x(t)
0

–1

–2

10 12 14 t 16 18 20

Fig. 3.29 Repetitive activities of the BVP model (3.16) with parameter values: a D 0:0 (dashed
curve) and a D 0:9 (solid curve). The parameter  is set to 0.001 which is much smaller than one
used in Fig. 3.27 and the waveforms have a sharp shape. As the a value increases, the fraction time
such that x.t / > 0 decreases and the waveform becomes asymmetric (neuron-like)

model (3.16) with a parameter jaj < 1 oscillates spontaneously and thus we can
consider the neuron as a neuronal oscillator or a pacemaker cell. Note that the BVP
model (3.16) with a D 0 is equal to the piecewise-linearized version of the van-der
Pol oscillator (van der Pol 1926; Grasman and Jansen 1979; Xu and Jiang 1996).

3.3.2 Singular Version of the Model and Its Response

to Sinusoidal Input

In the following, we consider the BVP model (3.16) in the limit of  D 0.

Figure 3.30 shows the x–y phase plane of (3.16) with  D 0. The parameter is
set as jaj < 1, so the equilibrium point is unstable and the model shows a repetitive
activity. The rectangle ABCD is a stable limit cycle.
In the limit of  D 0, the orbit with an initial value which is not on the
x-nullcline (xP D 0) instantaneously jumps to the x-nullcline in the horizontal
direction since the velocity xP D f .x; y/= becomes infinity unless f .x; y/ D 0.
Thus, all orbits are considered to move on the x-nullcline. On the right (left) branch
of the N-shaped x-nullcline, an orbit moves toward the point B (D, resp.). At the
point B (D), the orbit instantaneously jumps to the point C (A, resp.), since we
consider the limit of  D 0. This limit case is called a singular limit since orbits are
not differentiable at such a jump point. Note that all orbits move on the x-nullcline
even in the case of a non-oscillatory parameter value jaj > 1 although all orbits
tend to the equilibrium (resting) point after a suitable time in this case.
Let us introduce a sinusoidal external input to the BVP model (3.16):

 xP D f .x; y/ D y  x C 5=6fjx C 1j  jx  1jg; (3.17a)

yP D g.x/ C v.t/  x C a C A sinf2.t=T C 0 /g; (3.17b)
3.3 A Variant of the BVP Neuron Model and Its Response to Periodic Input 83

y
y-nullcline (y = 0)

D(−1, 2/3) A(7/3, 2/3)

x
x-nullcline
(x˙ = 0)

C(−7/3, –2/3) B(1, –2/3)

Fig. 3.30 The x–y phase plane of (3.16) in the limit of  D 0 and the N-shaped x-nullcline
(xP D 0). The closed orbit ABCD is a limit cycle. The parameter a shifts the y-nullcline in the
horizontal direction (a D 0:4 in this figure)

where A, T , 0 are the amplitude, the period, the initial phase of the sinusoidal
input, respectively.
Note that the external stimulus v.t/ is applied to the variable y rather than x.
It is usual to apply the stimulus to the variable x since x is considered as a voltage
and the dimension of the r.h.s. of (3.17a) corresponds to a current. Mathematically,
however, the application of a current stimulus to y is equivalent to the one to x as
follows (Alexander et al. 1990): R If .x.t/; y.t// is a solution of (3.17) and define
.t/  x.t/ and .t/  y.t/  v.t/dt, then . .t/; .t// is a solution of
Z
 P D f . ; / C v.t/dt;
P D g. /:
R
Thus the addition of v.t/ to the r.h.s. of (3.17b) is equal to that of v.t/dt to the
r.h.s. of (3.17a).
In the singular limit, (3.17) is reduced to

yP D g.x/ C v.t/; f .x; y/ D 0

or more explicitly, to

xP D yP D x C a C A sinf2.t=T C 0 /g; (3.18a)


x  5=3 .x 1/;
yD (3.18b)
x C 5=3 .x  1/:

This differential equation is a one-dimensional (piecewise) linear equation. Thus

we can obtain its analytical solution piecewise (in each region x 1 or x  1).
84 3 Computational and Mathematical Models of Neurons

However, a total solution which connects two solutions of both region x 1 and
x  1 can not be obtained since we have to solve a transcendental equation which
includes both exponential and trigonometric functions.
Note that all orbits move on the x-nullcline of the x–y phase plane (see Fig. 3.30)
even in the case of the presence of the sinusoidal input; the sinusoidal input v.t/
changes the velocity of orbits on the x-nullcline. Also note that (3.18a) is equivalent
to the LIF neuron (3.6) when a sinusoidal signal is applied to. Thus the singular
BVP neuron model can be considered as a “combination” of two LIF neurons (i.e.
corresponding to the right and left branches of the x-nullcline of Fig. 3.30).

3.3.3 Phase Lockings

Figure 3.31 shows examples of responses of the BVP model (3.17) with  D 0:001
to a sinusoid. Panel (a) corresponds to a neuronal oscillator (a D 0:9) and (b)–(d)
a (non-oscillatory) neuron (a D 1:1). In (a) and (b), the neuron model excites once
during one period of a sinusoid. This response pattern is called a 1:1 phase locking.
Generally, a response in which m cycles of an input correspond to n spikes of a neu-
ron is called an m:n phase locking. Panels (c) and (d) show a 2:1 phase locking and
a 1:2 phase locking, respectively. Note that we do not see any essential difference
between an oscillatory neuron in (a) and a non-oscillatory neuron in (b).
In order to study the relation between the sinusoidal forcing and the forced BVP
model, let us observe the sinusoidal input whenever the orbit .x; y/ visits the point
A of the x–y phase plane (see Fig. 3.30). If the parameter a is set in the range of
jaj > 1 (non-oscillatory neuron), orbits will stay near the resting point P and may
not visit the point A without any input v.t/. If the amplitude of v.t/ is sufficiently
large, an orbit can excite (move to point C) and then visits the point A even in the
case of non-oscillatory neuron.

a b
−x(t) 2 2

v(t) 1 1
0 0
–1 –1
–2 –2
20 25 30 35 40 20 25 30 35 40
t t
c d
−x(t) 2 2

v(t) 1 1
0 0
–1 –1
–2 –2
20 25 30 35 40 20 30 40 50 60
t t

Fig. 3.31 Examples of phase lockings of the forced BVP model (3.17). The amplitude A of the
sinusoidal input v.t / is set to 1.0 and other parameters are: (a) a D 0:9, T D 3:0 (b) a D 1:1,
T D 3:0 (c) a D 1:1, T D 2:0 (d) a D 1:1, T D 8:0. The solid curves are waveforms of x.t /
and dashed curves are that of v.t /
3.3 A Variant of the BVP Neuron Model and Its Response to Periodic Input 85

a
θ0 θ1 θ2 θ3 θ4

6
v(t)
4

2
x(t)
0

–2
t=0 t
b
1.0 θ0

p(θ)

θ∗

θ2
θ1

0.0 θ 1.0

Fig. 3.32 (a) Phase of the sinusoidal input (a 1:1 phase locking). Waveforms of both x.t / and the
input v.t / are plotted (v.t / is shifted upward and its scale is arbitrary). A D 0:3, T D 1:5. The
peaks (maximum values) of x.t / corresponds to the point A of the x–y phase plane (Fig. 3.30).
At the time t D 0, the initial phase of the sinusoidal input is 0 and the phase is changed to 1 ,
2 , . . . as the time elapses. In this case, the sequence of the phases converges to a certain fixed value
  ; a 1:1 phase locking occurs. (b) Phase transition curve (PTC) p. / (thick curve) and its orbit
fn g (thin lines). A D 0:3, T D 1:5. The orbit 0 , 1 , . . . asymptotically converges to a fixed point
  which corresponds to a 1:1 phase locking of (a)

Figure 3.32a is an example of waveforms of both v.t/ and x.t/. The point A of
the x–y phase plane corresponds to the peaks of the waveform x.t/. The phase of
the sinusoidal input (at the point A) is a number between 0 and 1 normalized by the
period T . Note that this terminology “phase” is irrelevant to the “phase” of the x–y
phase plane. Suppose that the orbit starts at the point A of the x–y phase plane with
an initial phase 0 of the sinusoidal input and that the orbit returns to the point A
again after a time t1 . Then the second phase 1 at the second visit to A is obtained by

def
1 D p.0 / D 0 C t1 =T .mod 1/: (3.19)

The relation p.0 / between 0 and 1 is called a return map or a phase transition
curve (PTC). In Sect. 3.2, the phase of a forced neuron was defined using the state of
the neuron when the state of the forcing was fixed (i.e. the membrane potential was
86 3 Computational and Mathematical Models of Neurons

sampled when t D kT ) whereas in this BVP model the phase is defined using the
state of the sinusoidal forcing when the neuron’s state is fixed (the sinusoidal forc-
ing is observed whenever the state point .x.t/; y.t// passes the point A). Note that
this difference of definitions of phases, however, is just the difference of coordinate
choice and does not matter at all.
Using the return map, phases 2 , 3 , . . . are recursively defined by the one-
dimensional mapping:

nC1 D p.n /; n D 0; 1; 2; : : : : (3.20)

The sequence fn g generated by this equation is also called the orbit of (3.20).
Thus, in order to analyze the phase lockings of the forced BVP model (3.17),
we investigate the asymptotic properties of the sequence (orbit) fn g of phases. In
Fig. 3.32a, the sequence fn g asymptotically approaches a certain values   ; in this
case a 1:1 locking occurs. Figure 3.32b shows an example of the return map p./
and its orbit fn g which corresponds to Fig. 3.32a. The graph of p./ intersects with
a diagonal line in two points (indistinguishable in this figure) and the lower intersec-
tion point   is a stable fixed point of (3.20) (see the fold bifurcation described in
Sect. 1.4.1). Thus an orbit 0 , 1 , . . . , with any initial phase 0 asymptotically con-
verges to this fixed point   , which shows a 1:1 phase locking occurs. Generally, in
the case of m:n locking, an orbit or sequence fn g asymptotically approaches to an
n-periodic sequence:
 .1/ ;  .2/ ; : : : ;  .n/ ;  .1/ ; : : : :
Note that we can not obtain the explicit form of p./ analytically by the same
reason as stated in Sect. 3.3.2. So, we numerically calculated p./ solving some
transcendental equations by the Newton method. The computation time of p./ is
very short since the convergence of the Newton method is much faster than the
numerical simulation of differential equations.

3.3.4 Bifurcation Diagrams

Patterns of phase lockings depend on both the amplitude and the period of the sinu-
soidal input. One phase-locking pattern can be considered to be bifurcated from the
other phase-locking pattern as a (bifurcation) parameter is changed. A bifurcation
diagram is a very useful tool to systematically investigate the effect of a parameter
change on a system’s behavior.
Figure 3.33 shows how the asymptotic value(s) of the sequence fn g is changed
depending on the input period T . Panel (a) is a bifurcation diagram of a D 0:0
(original van der Pol oscillator). For example, in the T D 2 case, only one dot is
plotted in the vertical direction (   0:14); a 1:1 locking occurs. In the T D 7
case, three dots are plotted and a 1:3 locking occurs; one cycle of the sinusoidal in-
put synchronizes with three cycles of the oscillator. In the T D 4 case, many points
3.3 A Variant of the BVP Neuron Model and Its Response to Periodic Input 87

a b
1
1
1:1 locking 1:3

1:1 lockingt 2:3 1:2 1:3 1:4

{θn} {θn}

0 0
1 T 11 1 T 11
c d
1 1

1:1 1:2 1:3 2:1 1:1 1:2 1:3

{θn} {θn}

0 0
1 T 11 1 T 11

Fig. 3.33 Bifurcation diagrams of the return map or the PTC p. /. A D 1:0. A stationary (or
asymptotic) sequence after some transient initial sequence fn g, n D 101, . . . , 600 produced by
(3.20) were plotted for each of 600 equally spaced T values on the interval Œ1; 11. (a) a D 0
(the van der Pol oscillator) (b) a D 0:5 (the BVP oscillator) (c) a D 0:9 (the BVP oscillator)
(d) a D 1:1 (non-oscillatory neuron)

are plotted; no locking occurs (in this case, the sequence fn g approaches a so-called
quasi-periodic sequence rather than a periodic sequence). In a quasi-periodic case,
the oscillator is not phase-locked to the input and the phase of the input takes in-
finitely many values between 0 and 1 although the oscillator seems to be locked for
a short time range (quasi-periodic orbits are different from chaotic ones).
In panel (b), the value of a is set to 0:5. As stated in Sect. 3.3.1, the asymme-
try of the waveform increases if the absolute value of a increases. The larger jaj
becomes, the narrower the T range of quasi-periodic orbits becomes, although the
total bifurcation structure (1:1, 1:3 lockings of (a)) does not change much. From this
observation, we can see that the asymmetry of an oscillator increases the tendency
to phase lockings. In panels (c) and (d), the value jaj is increased more. We can still
see the similarity in all the figures (a)–(d) although the bifurcation diagrams seem
to be prolonged to the right direction as jaj increases. A notable point is that we can
see no essential difference between (c) and (d) although the panel (c) corresponds
to a neuronal oscillator while the panel (d) a non-oscillatory neuron.
We note that the intrinsic period TaInt of the singular BVP model which has a
repetitive activity (jaj < 1/ without the sinusoidal input (v.t/  0) is given by

.7=3/2  a2
TaInt D ln
1  a2
88 3 Computational and Mathematical Models of Neurons

and does not change much for a parameter range 0  jaj  0:5:
Int
T0:0 D 1:69; Int
T˙0:3 D 1:77; Int
T˙0:5 D 1:94; Int
T˙0:9 D 3:19:

Thus we conjecture that the difference between (a) and (b) is caused not by the
difference of intrinsic period but by that of the asymmetry.

3.4 Stochastic Neuron Models

So far, we have considered deterministic models of single neurons and their response
characteristics to also deterministic inputs. A real neuron, however, has stochastic
natures in itself, such as spontaneous releases of a synaptic vesicle which cause the
random fluctuations of the order 0.5 mV in membrane potential (miniature end-plate
potentials), the random opening and closing of ion channels which cause conduc-
tance changes, and so on. Furthermore, a single neuron such as cortical neurons
receives input signals from many other neurons and the input sequence is never pe-
riodic but can be approximated by a stochastic or random signal in a certain sense.
(For the stochastic nature of neurons and the stochastic neuron models, see Ricciardi
1977 and Tuckwell 1988, 1989.)
In this section we briefly summarize the input–output characteristics of the IF
neuron and the LIF neuron to a Gaussian white noise input with noise intensity .
The noise is interpreted as a random component of the total inputs to a single neuron
or as an internal fluctuations as stated above.
In the following, we focus on the mean interspike interval (ISI) and the coefficient
of variation (CV) of the ISIs as functions of the input parameters, the constant bias 
and the noise intensity . Let us denote the ISI as a random variable T in this section.
Then, the mean EŒT  of the ISI is the inverse of the mean firing rate, and the CV,
which describes the relative width of the ISI histogram, is defined as follows:
p
VarŒT 
CV  :
EŒT 

The dimensionless CV is often used as a measure of spike train irregularity. For

a very regular spike train (pacemaker), the ISI histogram will have a very narrow
peak and CV  0. When a neuron fires periodically with period 2 and the observed
series of ISIs are f5; 10; 5; 10; 5; 10; : : :g, the resultant CV is 1=3. If a spike train is a
completely random process (i.e. a Poisson process), the corresponding ISI histogram
is an exponential distribution and CV D 1.
After the report of Softky and Koch (1993), a high CV at high firing rates ob-
served in the cortex has led to numerous speculations on the nature of the neural
code (Shadlen and Newsome 1994, 1998; Konig et al. 1996).
3.4 Stochastic Neuron Models 89

3.4.1 Stochastic Integrate-and-Fire Neuron

Let the time-variable external input I.t/ in the IF neuron (3.2) be the Gaussian white
noise with noise intensity . Then, the time variation of the membrane potential
V .t/ of a neuron becomes the well-known Wiener process and is described by a
stochastic differential equation as follows:

dV.t/ D dt C dW.t/; (3.21)

where W .t/ is the standard Wiener process or the Brownian motion (Gardiner 1983).
The value of  is the bias of the total sum of many excitatory and inhibitory inputs
and the term d W .t/ is the (random) deviation from the bias. This stochastic IF
neuron has been originally proposed by Gerstein and Mandelbrot (1964) as an ap-
proximation of the random walk model.
The ISI is represented by the random variable

T D infft 0 W V .t/ g; V .0/ D 0 <  (3.22)

which is the first-passage time (FPT) through  of the stochastic process V .t/. In
this framework, the ISIs are generated in accordance with a renewal process and
therefore the nature of the ISI sequence is completely described by the probability
density function (pdf) of the T which is the theoretical counterpart of the ISI his-
togram. For the Wiener process with a constant threshold, the distribution of T is
the well-known Inverse Gaussian Distribution (IGD) (see Chhikara and Folks 1988
for details of the distribution) and the pdf is as follows:
" #
 .  t/2
g.t/ D p exp  ; t > 0: (3.23)
2 2 t 3 2 2 t

When the constant bias  is zero or toward the threshold  ( > 0), an action
potential is generated in a finite time with probability one. On the other hand, when
it is away from the threshold ( < 0), there is a probability that no action potential
is ever generated. Also, for  D 0 which is the case that a neuron gets completely
same amount of excitatory and inhibitory inputs, the mean ISI becomes infinite.
Therefore only the IF neuron with the positive constant bias  is adequate for a
model of active neurons. Figure 3.34 shows the examples of ISI densities for the
IF neuron with various values of the parameters. From this figure, it is evident that
the ISI densities of the stochastic IF neuron can have a wide variety of shapes and
thus can be fitted to the ISI histograms obtained experimentally for various types
of neurons. The stochastic IF neuron, however, has a definite deficiency as will be
stated below.
The mean and the variance of the ISI for the positive bias  are obtained as
follows (Chhikara and Folks 1988):


EŒT  D   2
; VarŒT  D ;  > 0:
3
90 3 Computational and Mathematical Models of Neurons

a 6
b
6
α=1 β=1
5 5

4
a
4
a
pdf

pdf
3 3
d
2 2
b
b c
1 1 c
d
0 0
0 1 2 3 0 1 2 3
t t

Fig. 3.34 ISI densities of the IF neuron. The function g.t / of the IGD (3.23) are plotted for various
values of parameters: (a) ˛ D 1, a ˇ D 0:1; b ˇ D 1; c ˇ D 10; d ˇ D 100, (b) ˇ D 1, a ˛ D 0:2;
b ˛ D 0:4; c ˛ D 0:8; d ˛ D 1:6, where we have set ˛ D = and ˇ D  2 = 2 . (The IGD is the
distribution family with the parameters ˛ and ˇ)

The mean ISI is the same as the ISI of the IF neuron without any random inputs
( D 0). Noise intensity  does not affect the mean ISI. The coefficient of variation
(CV) is
CVŒT  D p ;  > 0 (3.24)

which shows that CV can take the value from 0 to infinity depending on the model
parameters and that it is proportional to the noise intensity  and also to the recipro-
cal of the square root of the input bias . By choosing the value of  appropriately,
the IF neurons can fire with a variety of irregularity holding the value of mean ISI.
However, when  is very small, which is the case that a neuron receives almost
similar amount of excitatory and inhibitory inputs, both the mean and the CV of ISI
become extremely big values. Therefore, the IF neuron with a large mean ISI does
not show any exponential behaviour (i.e. CV  1) which is often observed in the
spontaneous activities of single neurons.

3.4.2 Stochastic Leaky Integrate-and-Fire Neurons

The membrane potential of the LIF neuron (3.6) stimulated by a Gaussian white
noise is described by the Ornstein–Uhlenbeck (OU) process:

dV.t/ D .V =
C /dt C dW.t/; V .0/ D 0 < : (3.25)

Differently from the IF neuron, the first-passage through  is a sure event (i.e. an
event with probability one) for any values of  and  > 0. The LIF neuron will
3.4 Stochastic Neuron Models 91

500

400
Firing Rate [1/s]

300

200
σ = 10

100
σ=5 σ=2 σ=0
0
–4 –2 0 2 4
μ θ/τ

Fig. 3.35 Firing rates of the LIF neuron. Dotted curve shows the firing rate of the LIF neuron
without a random input ( D 0). Solid curves show the mean firing rate of the LIF neuron with
a Gaussian white noise for various noise intensity. For the evaluation of these quantities, we have
used the formula of Ricciardi and Sato (1988)

fire even when inhibitory inputs are superior to excitatory inputs. Figure 3.35 shows
the firing rate of the stochastic LIF neuron as a function of the constant bias  for
various values of noise intensity . We can clearly see that the random inputs ensure
firings of the LIF neuron for any value of  and accelerate them. It is also shown that
the firing rate as a function of  becomes linearized as the noise intensity increases,
see Clay (1976) and Yu and Lewis (1989) for the linearization by noise.
Figure 3.36 shows the ISI densities for the LIF neurons with various input pa-
rameters. We can see that the ISI densities cover typical features of experimentally
obtained ISI histograms of single neurons such as the gamma and the exponential
distributions. In particular, the densities of the LIF neurons can show a density shape
with a steep rising and a high peak and also with an exponentially long tail which
cannot be obtained by the ISI densities (IGD) of the IF neurons.
Figure 3.37 shows the mean and the CV of ISI as a function of the noise intensity
 with various values of . We can see that the values of the mean ISI for a positive
bias ( D 2) are too small as a neuron model (Fig. 3.37a). When  D 2, the attractor
of the deterministic part of the dynamics is above the threshold  and therefore pas-
sage to the threshold is essentially driven by deterministic forces, where the model
neuron can fire without fluctuation. On the other hand, we notice that negative bi-
ases can produce ISIs in the physiologically plausible range when noise intensity 
is enough large (see also Fig. 3.35). It tells us that the LIF neuron produces a spike
train with an adequate firing rate when the fluctuation is dominant for the firing than
the deterministic force by the term of .
92 3 Computational and Mathematical Models of Neurons

a c
0.30 σ=1 μ = –3
μ = 4.0 0.20
0.25
σ = 13
0.15
0.20
μ = 3.5

pdf
pdf

0.15 0.10 σ = 10
0.10 μ = 3.0
μ = 2.5 0.05 σ=7
0.05
0.00
0.00
0 10 20 30 40 50 0 10 20 30 40 50
t msec t msec
b d
0.25 σ = 10 0.25 μ=3
σ = 10
μ=3
0.20 0.20
σ=7
0.15 μ=0 0.15 σ=4
pdf
pdf

0.10 μ = –3 0.10 σ=1

0.05 μ = –6 0.05

0.00 0.00
0 10 20 30 40 50 0 10 20 30 40 50
t msec t msec

Fig. 3.36 Examples of ISI densities (histograms) of the stochastic LIF neurons. We have set
 D 15 mV and
D 5 ms. They have been obtained numerically by means of the algorithm pro-
posed in Buonocore et al. (1987)

a b
107 2.2
2
106
1.8
105 1.6
E [Tθ] (msec)

104 1.4
μ = –4
CV

1.2
103 μ=0
μ = –4 1 μ=2
μ = –2 μ=4
102 0.8
μ=0
μ=2 0.6
101
0.4
100
0.2
0 5 10 15 20 0 5 10 15 20
σ σ

Fig. 3.37 (a) The mean EŒT  and (b) CV of ISIs for the OU model as a function of
the noise intensity  for various values of the parameter . We have set  D 10 mV and
 D 4; 0; 2; 4 mV ms1 . For the evaluation of these quantities, we have used the formula of
Ricciardi and Sato (1988)

Figure 3.37b shows that the value of CV approaches zero and the LIF neuron acts
as a pacemaker when  ! 0 for the cases of  D 2; 4. It is also shown that the value
3.5 Stochastic Phase-Lockings and Bifurcations in Noisy Neuron Model 93

Fig. 3.38 Plots of the pairs 4

of the mean EŒT  and CV of
ISIs obtained from the LIF 3.5
neuron. The firing threshold 3
is  D 10 mV. The input
parameters have been 2.5
changed in the

CV
2
physiologically reasonable
parameter ranges and 1.5
the parameters which give
EŒT =
< 1 are excluded. 1
The parameters are changed
0.5
in the following range:
1    30 mV ms1=2 , 0
10    10 mV ms1 0 50 100 150
E[Tθ]

of CV of the LIF neuron ranges from 0 to 1 mainly in accordance with the input
variability . Low CVs are attained only when the bias  is positively large enough
to reach the threshold and the value of  is small. For neurons which are not able
to fire without random noise ( D 0; 4 in the figure), the values of CV are mostly
above unity.
Figure 3.38 plots the pairs of the mean EŒT  and the CV of the LIF neuron for
various parameter values. The parameter values have been changed in the physio-
logically reasonable parameter ranges and the parameters which give the mean ISI
less than the membrane time constant
are excluded. The ranges of CV for the LIF
neuron is 0:5158  CV  3:381. High CVs with a high firing rate, namely with
a short mean ISI, as well as the exponential behaviour (CV  1) with a low firing
rate are observed. The lower limit of CV  0:5 coincides with the result of Softky
and Koch (1993) which reported that the CVs of the ISIs for neocortical neurons of
awake behaving monkey ranged between 0.5 and 1.0. It is shown that the extremely
high CVs are much decreased by introducing a lower boundary of the membrane
potential or by introducing reversal potentials (Inoue and Doi 2007).

3.5 Stochastic Phase-Lockings and Bifurcations

in Noisy Neuron Model

Next, we consider the effect of both periodic and stochastic forcing on the response
characteristics of a neuronal model. Let us study the noise effects on the forced BVP
model (3.17) and thus we consider a stochastic differential equation:

 xP D y  x C 5=6fjx C 1j  jx  1jg; (3.26a)

dW.t/
yP D x C a C A sinf2.t=T C 0 /g C  ; (3.26b)
dt
94 3 Computational and Mathematical Models of Neurons

where W .t/ is the standard Wiener process or the Brownian motion (Gardiner 1983)
and the term dW.t/=dt denotes a Gaussian white noise with noise intensity . We
also suppose the singular limit  D 0 and then this equation is reduced to

dW.t/
xP D x C a C A sinf2.t=T C 0 /g C  ; (3.27a)
dt

x  5=3 .x 1/;
yD (3.27b)
x C 5=3 .x  1/:

Note that, as stated in Sect. 3.3.2, this stochastic (singular) BVP neuron model
forced by a sinusoidal signal can be considered as a combination of two stochas-
tic LIF neuron (3.6) when a sinusoidal input is applied to.
In order to investigate the noise effects in detail, the parameter a of the BVP
model (3.26) or (3.27) is set as a D 0:0 in the following numerical examples. Thus
we are specifically considering the noise effects on the forced (piecewise-linearized)
van der Pol oscillator. All the following analysis frameworks, however, do not de-
pend on this specific choice of the parameter value a and are valid for all parameter
values a, even for the case jaj > 1 (a non-oscillatory neuron).

3.5.1 Stochastic Phase Lockings

In the presence of noise, both variables 0 and 1 defined in (3.19) fluctuate by noise
and are thus random variables ‚0 , ‚1 . We extend the deterministic map p./ to the
case with noise as follows. Define a kernel function g.0 ; 1 / using conditional
probability density functions:

g.0 ; 1 /d1 D Prf1  ‚1  1 C d1 j ‚0 D 0 g: (3.28)

The function g.0 ; 1 / can be calculated numerically without simulations of the

stochastic differential equations (3.26) or (3.27) (Tateno et al. 1995; Doi et al. 1999).
Figure 3.39 is an example of the kernel function g which corresponds to the deter-
ministic map (PTC) of Fig. 3.32b. The function g takes relatively high values along
the curves of the map p./. Thus g can be considered as the stochastic extension
of the deterministic PTC. The heights and widths of the peaks of g depend on the
values of (0 , 1 ) and thus we can see that the effects of blurring of the deterministic
map by noise are not uniform.
Using the kernel function g, we extend the system (3.20) to the noisy case as
follows. Let S denote a unit circle [0,1) and D the set of absolutely integrable func-
tions with a unit L1 norm on S. An operator P on D is defined by
Z
Ph ./ D g.0 ; /h.0 /d0 ; h 2 D: (3.29)
S
3.5 Stochastic Phase-Lockings and Bifurcations in Noisy Neuron Model 95

Fig. 3.39 Stochastic kernel

function g.0 ; 1 /. a D 0:0, g(θ0, θ1)
A D 0:3, T D 1:5,  D 0:03.
This figure corresponds to the
deterministic return map of 1.0
Fig. 3.32b

θ1

0.0 θ0 1.0

This operator is a Markov operator since it transforms a probability density function

to an another probability density function (Lasota and Mackey 1994). Let h0 ./ 2 D
denote the probability density function of the initial phase ‚0 when an orbit starts
at the point A of the x–y phase plane of Fig. 3.30. Then the density function of ‚1
when the orbit returns to the point A again is obtained by h1 ./ D Ph0 ./. Thus,
the deterministic mapping (3.20) is extended to the system

hnC1 ./ D Phn ./; n D 0; 1; 2; : : : : (3.30)

Thus we investigate the asymptotic behavior of the sequence fhn ./g of probability
density functions rather than the sequence fn g of phases. The operator P is called a
stochastic phase transition operator since it is a direct extension of the deterministic
PTC to the stochastic (noisy) case.
Let us list several preliminary definitions (Lasota and Mackey 1994). A function
h ./ is called the invariant density function of an operator P if the relation Ph D
h holds. The invariant density is asymptotically stable if for any initial density
function h0 2 D
lim jjP n h0  h jj D 0;
n!1

where jj  jj is the L norm on S.

1

As is easily seen from its definition, the function g has the property
Z
g.0 ; 1 / 0; g.0 ; 1 /d1 D 1
S

and is called a stochastic kernel. The inequality

Z
inf g.0 ; 1 /d1 > 0 (3.31)
S 0
96 3 Computational and Mathematical Models of Neurons

is assumed to hold (Tateno et al. 1995). The operator with this property has a unique
asymptotically stable invariant density (Lasota and Mackey 1994). Thus the se-
quence fhn ./g produced by the operator P asymptotically approaches a unique
invariant density as n ! 1. The invariant density is considered as the stationary
(asymptotic) probability density function of the phase ‚n .
Figure 3.40a is an example of the evolution of the density sequence fhn ./g. The
uniform initial density function h0 ./ changes its shape and approaches an invariant
density function h ( h30 ) with one sharp peak, which shows a 1:1 phase locking
does occur in a stochastic sense. The peak of the density functions fhn ./g is highest
near n D 7 and is relatively lower in the invariant density h , which means that
fluctuations of phases are bigger in the stationary state than in the transient state.
This phenomenon is due to the fact that the peaks of the kernel function g.0 ; 1 /
near the deterministic fixed point 0 D 1 D   are relatively lower than the other
(cf. Figs. 3.39 and 3.32b).
Figure 3.40b is also an example of the density sequence in the case of stochastic
5:3 phase locking. The initial density function h0 ./ has a high peak near  D 0:42

a b
10.0 33.3

hn(θ) hn(θ)

0 0

n n

30 10
0.0 θ 1.0 0.0 θ 1.0

h∗(θ)

0.0 θ 1.0

Fig. 3.40 Density evolution (sequence) fhn . /g. (a) Stochastic 1:1 phase locking. A D 0:3,
T D 1:5,  D 0:03. A uniform initial density function evolves to an invariant density function
h  h30 . (b) Stochastic 5:3 phase locking. A D 2:0, T D 0:85,  D 0:02. (c) Invariant density
h  h200 of (b)
3.5 Stochastic Phase-Lockings and Bifurcations in Noisy Neuron Model 97

which is one of the three phases to which oscillator is locked in the noise-free case.
The functions h1 and h2 have a high mode near the other two phases of the three
phases. This sequence seems to initially vary with period 3, but finally converges to
an invariant density function (panel (c)) with three sharp peaks, which shows a 5:3
locking occurs in a stochastic sense.
In the noise-free case, the amplitude A of the sinusoid should be sufficiently
large in order to define the deterministic map p./ in the case of jaj > 1 (non-
oscillatory neuron), since the neuron may not excite (and thus orbits of the BVP
model may not visit the point A of the x–y phase plane) otherwise. Note that this
condition is not necessary in the noisy case since there are always positive possibil-
ities (probabilities), for all values of a, that a non-oscillatory neuron excites owing
to noise.

3.5.2 Stochastic Bifurcation Diagrams

Patterns of deterministic phase lockings depend on both the amplitude A and the
period T of the input. Figure 3.41 shows a deterministic bifurcation diagram of
the (PTC) map p./ for the noise-free ( D 0) case with a bifurcation parameter
A rather than T (cf. Fig. 3.33). In this case, the period of the sinusoidal input is about
a half of the intrinsic period of the van der Pol oscillator. So, for a wide range of
the amplitude A, 2:1 phase lockings occur (two cycles of the input synchronize with
one cycle of the oscillator). In particular, two different patterns of 2:1 phase lockings
coexist; the pattern depends on both the initial phase of the input and the initial state

Coexistence of
5:3 1:1
two 2:1 lockings
{θn}

0
0 2.5
A

Fig. 3.41 Deterministic bifurcation diagram of the return map p. /. The bifurcation parameter is
the amplitude A rather than the period T , which is fixed to T D 0:85. A sequence fn g, n D 101,
. . . , 600 produced by (3.20) were plotted for each of 600 equally spaced A values on the interval
Œ0; 2:5
98 3 Computational and Mathematical Models of Neurons

a b
h∗(θ) h∗(θ)

0.0 0.0

A A

2.5 2.5
0.0 θ 1.0 0.0 θ 1.0

Fig. 3.42 Stochastic (invariant density) bifurcation diagram. T D 0:85. Invariant densities h . /
were plotted for each of 50 equally spaced A values on the interval Œ0; 2:5. (a)  D 0.05 (b)  D 0.2

of the oscillator. Various phase-locking patterns are bifurcated as the value of A

varies. If A is large enough (A > 2:4), one cycle of the input can synchronize with
the one cycle of the oscillator although the intrinsic periods differ by twice.
Figure 3.42 is a stochastic or a density bifurcation diagram which corresponds to
the deterministic one (Fig. 3.41). Invariant density functions are plotted with various
value of A for two different noise intensities: (a)  D 0:05 and (b)  D 0:2. In the
panel (a), corresponding to the deterministic bifurcation diagram, the modes (peaks)
and the shape of the invariant density are changed as the bifurcation parameter A
varies. Particularly, in the range of 1:8 < A < 2:4, the invariant densities have
several peaks and complicated shapes although the shapes are not so complicated as
the deterministic one.
In the case of large noise (panel b), the stochastic bifurcation diagram is very
simple; the invariant densities have at most two peaks. The shapes of invariant den-
sities do not depend on the amplitude A; only the heights of the two peaks depend
on A. Thus noise, as is easily expected, washes out the dependency of the density
shapes (the number of modes) on the amplitude of the input.
In the deterministic dynamical systems, the word “bifurcation” means qualitative
change (the number and stability) of solutions of a system. As stated above, the
equation
Ph D h
which the invariant density satisfies, always has a unique asymptotically stable
solution and thus no bifurcation of this equation occurs in such a sense. So what
is a stochastic bifurcation? This is a newly developing field (Arnold 1995, 1998;
Doi et al. 1998).
Chapter 4
Whole System Analysis of Hodgkin–Huxley
Systems

In Chap. 2, we have explored the dynamics of the original Hodgkin–Huxley

equations of a squid giant axon where only the parameter Iext was changed. The
HH equations, however, include various constants or parameters whose values were
determined based on physiological experiments, and thus the values inherently
possess a certain ambiguity. Also, the “constants” are not really constant but change
temporally. Thus, in this chapter, we study the effects of the change of the constants
or parameters on the dynamics of the HH equations and consider the robustness and
sensitivity of the equations; we study the bifurcation structure of the HH equations
by changing their various parameters. To do so, in this chapter, we consider a slight
modification of the original HH equations since the modification is mathematically
more tractable.

4.1 Changing the Parameters: Sensitivity and Robustness

Consider a slight simplification of the HH equations (2.3):

dv
C D G.v; m; n; h/ C Iext ; (4.1a)
dt
dm
D 1 .m1 .v/  m/; (4.1b)
dt
m .v/
dn 1
D .n1 .v/  n/; (4.1c)
dt
n .v/
dh 1
D .h1 .v/  h/; (4.1d)
dt
h .v/

where the nonlinear functions

x .v/ and x 1 .v/, (x D m; n; h) are denoted as
follows:
1 tNm
m1 .v/ D ;
m .v/ D  ; (4.2a)
.1 C expŒNsm .v  vN m;1 // sNm .v  vN m;2 /
cosh
2

S. Doi et al., Computational Electrophysiology, 99

DOI 10.1007/978-4-431-53862-2 4,  c Springer 2010
100 4 Whole System Analysis of Hodgkin–Huxley Systems

1 tNn
n1 .v/ D ;
n .v/ D  ; (4.2b)
.1 C expŒNsn .v  vN n;1 // sNn .v  vN n;2 /
cosh
2
1 tNh
h1 .v/ D ;
h .v/ D  : (4.2c)
.1 C expŒNsh .v  vN h;1 // sNh .v  vN h;2 /
cosh
2

Note that, differently from the original HH equations (2.3), all functions x 1 .v/
(or all
x .v/), x D m; n; h have the same functional form (only the constants sNx ,
tNx , vN x;1 , vN x;2 , x D m; n; h, are different). Figure 4.1 shows the functions x 1 .v/ and

x .v/ (x D m; n; h) of the simplified HH equations (4.1) in solid curves and those

of the original HH equations (2.3) in broken curves. First, note that, as seen from
(4.2), the functions in the simplified HH equations are symmetric with respect to
v D vN x;i (x D m; n; h, i D 1; 2), although the original HH equations do not have
such symmetry. Secondly, note that the functions
h .v/ and
n .v/ in (4.2), where the
values of the parameters sNx , tNx , vN x;2 (x D n; h) are given in the caption of Fig. 4.1,
are much different from those of the original HH equations.

1 1 1
a-i b-i c-i
0.8 0.8 0.8

0.6 0.6 0.6

minf

hinf

ninf

0.4 0.4 0.4

0.2 0.2 0.2

0 0 0
−100 0 100 −100 0 100 −100 0 100
v v v
0.8 12 6
a-ii b-ii c-ii
10 5
0.6
8 4
mtau

ntau
htau

0.4 6 3

4 2
0.2
2 1

0 0 0
−100 0 100 −100 0 100 −100 0 100
v v v

Fig. 4.1 Comparison of the functions (a-i) m1 .v/, (b-i) h1 .v/, (c-i) n1 .v/, (a-ii)
m .v/,
(b-ii)
h .v/, (c-ii)
n .v/ of the slight simplified HH equations (4.1) (solid curves) with those of
the original HH equations (2.3) (broken curves). sNm D 0:1, vN m;1 D vN m;2 D 24:0, tNm D 0:5,
sNh D 0:09, vN h;1 D vN h;2 D 2:0, tNh D 12:0, sNn D 0:06, vN n;1 D vN n;2 D 10:0, tNn D 5:0
4.1 Changing the Parameters: Sensitivity and Robustness 101

1
100

m
50 0.5
v

0
0
0 20 40 60 0 20 40 60
t t
1 1

0.5 0.5
h

n
0 0
0 20 40 60 0 20 40 60
t t

Fig. 4.2 Comparison of the solution of the slight simplified HH equations (4.1) (solid curves) with
that of the original HH equations (2.3) (broken curves). Iext D 15. sNm D 0:1, vN m;1 D vN m;2 D 24:0,
tNm D 0:5; sNh D 0:09, vN h;1 D vN h;2 D 2:0, tNh D 12:0; sNn D 0:06, vN n;1 D vN n;2 D 10:0, tNn D 5:0

Different Models Show a Similar Behavior

Figure 4.2 shows the comparison of the solutions between (4.1) and (2.3).
Panels (a)–(d) compare the waveforms of v.t/, m.t/, h.t/, n.t/, respectively. In
spite of the big differences of the functions
h .v/ and
n .v/ shown in Fig. 4.1, the
waveforms of v.t/ are very similar each other, although the periods are different
slightly. (Note that a constant current is injected [Iext D 15], and thus there is a pe-
riodic [repetitive] firing.) The waveform of h.t/ in the two cases are much different
while the waveforms of m.t/ and n.t/ are very similar. Anyway, can we say that
such neuronal systems described by (2.3) or (4.1) are robust (i.e. small change of
functions or parameters do not affect the total behavior of the system)? Answer is
“No” as shown in the following.

Slight Difference Leads a Big Difference of Solutions

Figure 4.3 illustrates the effect of the change of the parameter vN n;1 of (4.1) when
Iext D 46:87. Solid and broken curves show the cases vN n;1 D 5:0 and vN n;2 D 10,
respectively. Other parameters are the same as those of Fig. 4.2. As seen from
panel (c-i), this change of vN n;1 leads a very small change in the function n1 .v/
(other functions are completely the same in the two cases and overlapped, and thus
broken curves are not seen). However, the solutions show completely different be-
havior. The period of periodic firing in solid curve is hundred times longer than that
in broken curve (note the scale difference of abscissae in panels d and e).
As we have seen in Figs. 4.1 and 4.2, in spite of the difference of functional
forms, the membrane-potential waveforms were very similar in the original HH
102 4 Whole System Analysis of Hodgkin–Huxley Systems

1 1 1
a-i b-i c-i
0.8 0.8 0.8
minf 0.6 0.6 0.6

hinf

ninf
0.4 0.4 0.4

0.2 0.2 0.2

0 0 0
−100 0 100 −100 0 100 −100 0 100
v v v

0.5 12 5
a-ii 10
b-ii c-ii
0.4 4
8
0.3 3
mtau

htau

ntau
6
0.2 2
4
0.1 2 1

0 0 0
−100 0 100 −100 0 100 −100 0 100
v v v

100 d 100 e

50 50
v

0 0

0 20 40 60 80 100 0 1000 2000 3000 4000 5000

t t

Fig. 4.3 Comparison of the functional forms and the solution of (4.1) in two cases: vN n;1 D 5 (solid
curves) and vN n;1 D 10 (broken curves). Functions (a-i) m1 .v/, (b-i) h1 .v/, (c-i) n1 .v/, (a-ii)

m .v/, (b-ii)
h .v/, (c-ii)
n .v/, and (d, e) membrane potential waveforms. Iext D 46:87. Other
parameters are the same in both cases: sNm D 0:1, vN m;1 D vN m;2 D 24:0, tNm D 0:5; sNh D 0:09,
vN h;1 D vN h;2 D 2:0, tNh D 12:0; sNn D 0:06, vN n;2 D 10:0, tNn D 5:0

equations (2.3) and their modification (4.1). Next, we choose the parameter values
so that the functions of the two equations become closer each other. Similarly to
Figs. 4.1 and 4.2, Fig. 4.4 compares the two models: the original HH equations (2.3)
and the simplification (4.1). (Note that Fig. 4.3 compared two solutions of the sim-
plification (4.1) only.) Now the two models are very close each other; functional
forms in both broken and solid curves are very similar. However, the behaviors
of solutions are completely different. The membrane potential of the original HH
equations (broken curve) fires repeatedly, while that of the simplification shows no
firing after the initial firing. What has happened? We note the slight difference in
the functional forms of n1 .v/ between solid and broken curves. The variable n is
the activation variable of outward potassium current. Thus, we guess that the slight
enhancement of the activation in low voltage region (solid curve) induces a fast
outward current flow and thus inhibits a repetitive firing.
4.1 Changing the Parameters: Sensitivity and Robustness 103

1 1 1

0.8 0.8 0.8

minf 0.6 0.6 0.6

hinf

ninf
0.4 0.4 0.4

0.2 0.2 0.2

0 0 0
−100 0 100 −100 0 100 −100 0 100
v v v

0.8 10 6

8 5
0.6
4
6
mtau

htau

ntau
0.4 3
4
2
0.2
2
1
0 0 0
−100 0 100 −100 0 100 −100 0 100
v v v

100 100

50 50
v

0 0

0 20 40 60 80 100 0 20 40 60 80 100
t t

Fig. 4.4 Comparison of the functional forms and the solution of the slight simplified HH equa-
tions (4.1) (solid curves) with that of the original HH equations (2.3) (broken curves). Iext D 15.
sNm D 0:1, vN m;1 D vN m;2 D 24:0, tNm D 0:5, sNh D 0:13, vN h;1 D vN h;2 D 2:0, tNh D 8:5, sNn D 0:055,
vN n;1 D 5:0, vN n;2 D 12:0, tNn D 5:7

Figure 4.5 shows the similar figure to Fig. 4.4, but the value of the parameter vN n;1
of the simplified HH equations (4.1) has been increased from 5.0 to 10.0. Now both
broken and solid graphs of n1 .v/ are overlapped more than Fig. 4.4. Repetitive fir-
ing has been recovered! However, we note that the amplitudes of two membrane
potential waveforms are different much. (Compare with the resemblance of the
waveforms in Fig. 4.2.)
Through these examples, we have demonstrated:
The closeness of functional forms and/or parameter values of nonlinear dynamical systems
does not mean the closeness of the solutions. Completely different behavior can be induced
by a small change of model parameter and/or function. Conversely, apparently different
systems often present a very similar behavior.

Therefore, we should pay much attention to the sensitivity of system’s behav-

ior on such parameter values and functional forms in model-based researches.
104 4 Whole System Analysis of Hodgkin–Huxley Systems

1 1 1

0.8 0.8 0.8

minf 0.6 0.6 0.6

ninf
hinf
0.4 0.4 0.4

0.2 0.2 0.2

0 0 0
−100 0 100 −100 0 100 −100 0 100
v v v

0.8 10 6

8 5
0.6
4
6
mtau

htau

ntau
0.4 3
4
2
0.2
2 1
0 0 0
−100 0 100 −100 0 100 −100 0 100
v v v

100 100

50 50
v

0 0

0 20 40 60 80 100 0 20 40 60 80 100
t t

Fig. 4.5 Comparison of the functional forms and the solution of the slight simplified HH equa-
tions (4.1) (solid curves) with that of the original HH equations (2.3) (broken curves). Iext D 15.
The value of the only parameter vN n;1 has been increased from 5.0 to 10.0 and other parameter values
are the same as those of Fig. 4.4

The analysis of the dependence of system’s dynamics on parameters is the

bifurcation analysis. In the following, we do such bifurcation analyses in detail,
in order to clarify the “whole” behavior of neuronal systems.

Exercise 4.1. Using MATLAB, try the following exercises:

 Figure 4.6 is a MATLAB script (commands) file to compare the forms of func-
tions x 1 .v/ and
x .v/ (x D m; n; h) between the original HH equations (2.3)
and its modification (4.1), as shown in Fig. 4.1. Try to run this script file. Try
also to run it with changing several parameters.
 Figure 4.7 is also a MATLAB script file to compare the solutions between the
original HH equations (2.3) and its modification (4.1), as shown in Fig. 4.2. Try
to execute this script file (with the two script files shown in Figs. 4.8 and 4.9) on
MATLAB. Try also to run it with changing several parameters and confirm the
results described in this section (Figs. 4.3, 4.4 and 4.5). 
4.2 Bifurcations of the Hodgkin–Huxley Neurons 105

%HHtypeFncs_1.m
clear all;
v=[-100+0.001:0.5:100];
am = 0.1*(v-25)./(1-exp(-(v-25)/10));
bm = 4*exp(-v/18);
ah = 0.07*exp(-v/20);
bh = 1./(1+exp(-(v-30)/10));
an = 0.01*(v-10)./(1-exp(-(v-10)/10));
bn = 0.125*exp(-v/80);
minf = am./(am+bm); mtau = 1./(am+bm);
hinf = ah./(ah+bh); htau = 1./(ah+bh);
ninf = an./(an+bn); ntau = 1./(an+bn);
%
Sm=0.1; Vm=24.0; Tm=0.5;
Sh=-0.13; Vh=-2.0; Th=8.5;
Sn=0.055; Vn1=10.0; Vn2=-12.0; Tn=5.7;
Minf = 1./(1+exp(-Sm*(v-Vm))); Mtau = Tm./cosh(Sm*(v-Vm)/2);
Hinf = 1./(1+exp(-Sh*(v-Vh))); Htau = Th./cosh(Sh*(v-Vh)/2);
Ninf = 1./(1+exp(-Sn*(v-Vn1))); Ntau = Tn./cosh(Sn*(v-Vn2)/2);
%
subplot(2,3,1); plot(v,minf,’-b’,v,Minf,’-r’); ylabel(’minf’);
xlabel(’v’);
subplot(2,3,4); plot(v,mtau,’-b’,v,Mtau,’-r’); ylabel(’mtau’);
subplot(2,3,2); plot(v,hinf,’-b’,v,Hinf,’-r’); ylabel(’hinf’);
subplot(2,3,5); plot(v,htau,’-b’,v,Htau,’-r’); ylabel(’htau’);
subplot(2,3,3); plot(v,ninf,’-b’,v,Ninf,’-r’); ylabel(’ninf’);
subplot(2,3,6); plot(v,ntau,’-b’,v,Ntau,’-r’); ylabel(’ntau’);

Fig. 4.6 MATLAB script file: HHtypeFncs 1.m which compares the functional forms between
original HH equations (2.3) and its modification (4.1)

4.2 Bifurcations of the Hodgkin–Huxley Neurons

In the rest of this chapter, we consider the slightly simplified HH equations (4.1)
only (we simply call these equations as the HH equations or the HH model un-
less any confusions occur) and consider the sensitivity of the HH model on the
parameter changes.
Neurons receive inputs from other neurons. Thus, the external current Iext of
(4.1) is the most important parameter. Figure 4.10a shows the bifurcation diagram
of the HH equations (4.1) similar to the diagram (Fig. 2.15) of the original HH
equations (2.3). The abscissa denotes the bifurcation parameter Iext and the ordi-
nate denotes the membrane potential v where the maximum value of v is plotted for
a periodic (oscillatory) solution. Solid and dotted curves denote stable and unstable
equilibrium points, respectively. The filled (open) circles denote stable (unstable,
resp.) periodic solutions. The parameter values of the HH equations (4.1) in Fig. 4.10
are the same as those of Fig. 4.2.
106 4 Whole System Analysis of Hodgkin–Huxley Systems

%HHtypeSimu_1.m
clear all;
global Iext TTm TTh TTn
global Gna Gk Gl Vna Vk Vl
global sm vm tm sh vh th sn vn tn
global Sm Vm Tm Sh Vh Th Sn Vn1 Vn2 Tn
Gna = 120; Gk = 36; Gl = 0.3;
Vna = 115; Vk = -12; Vl = 10.599;
Sm=0.1; Vm=24.0; Tm=0.5;
Sh=-0.13; Vh=-2.0; Th=8.5;
Sn=0.055; Vn1=10.0; Vn2=-12.0; Tn=5.7;
Iext=15; TTm=1; TTh=1; TTn=1;
TIME1=100; TIME2=TIME1;
[t1,x]= ode23(’HH_1’,[0 TIME1],[0, 0, 0, 0]);
[t2,y]= ode23(’HHtype2’,[0 TIME1],[0, 0, 0, 0]);
clf;
subplot(3,2,1); plot(t1,x(:,1),’-b’); xlabel(’t’);ylabel(’v’);
axis([0 TIME1 -20 120]);
subplot(3,2,3); plot(t2,y(:,1),’-r’); xlabel(’t’);ylabel(’v’);
axis([0 TIME2 -20 120]);

Fig. 4.7 MATLAB script file: HHtypeSimu 1.m which compares the solutions between original
HH equations (2.3) and its modification (4.1)

%HH_1.m
function [xdot,xinit,option] = HH_1(t, x)
global Iext TTm TTn TTh
global Gna Gk Gl Vna Vk Vl
C=1;
v=x(1); m=x(2); h=x(3); n=x(4);
am = 0.1*(v-25)/(1-exp(-(v-25)/10));
bm = 4*exp(-v/18);
ah = 0.07*exp(-v/20);
bh = 1/(1+exp(-(v-30)/10));
an = 0.01*(v-10)/(1-exp(-(v-10)/10));
bn = 0.125*exp(-v/80);
%
xdot = [
(Gna*m^3*h*(Vna-v) + Gk*n^4*(Vk-v) + Gl*(Vl-v) + Iext)/C;
(am*(1-m)-bm*m)/TTm; (ah*(1-h)-bh*h)/TTh; (an*(1-n)-bn*n)/TTn
];

Fig. 4.8 MATLAB script file: HH 1.m which describes the original HH equations (2.3)

At the point HB1 (Iext D 1:93), unstable periodic solutions are bifurcated from
the equilibrium point by the (sub-critical or unstable) Hopf bifurcation. Stable pe-
riodic solutions are bifurcated by the (super-critical or stable) Hopf bifurcation at
the point HB2 (Iext D 282:92). At the point DC, the double-cycle bifurcation or
4.2 Bifurcations of the Hodgkin–Huxley Neurons 107

%HHtype2.m
function [xdot,xinit,option] = HHtype2(t, x)
global Iext TTm TTn TTh
global Gna Gk Gl Vna Vk Vl
global Sm Vm Tm Sh Vh Th Sn Vn1 Vn2 Tn
C=1;
v=x(1); m=x(2); h=x(3); n=x(4);
%
Minf = 1/(1+exp(-Sm*(v-Vm))); Mtau = Tm/cosh(Sm*(v-Vm)/2);
Hinf = 1/(1+exp(-Sh*(v-Vh))); Htau = Th/cosh(Sh*(v-Vh)/2);
Ninf = 1/(1+exp(-Sn*(v-Vn1))); Ntau = Tn/cosh(Sn*(v-Vn2)/2);
%
xdot = [
(Gna*m^3*h*(Vna-v) + Gk*n^4*(Vk-v) + Gl*(Vl-v) + Iext)/C;
(Minf-m)/(Mtau*TTm); (Hinf-h)/(Htau*TTh); (Ninf-n)/(Ntau*TTn)
];

Fig. 4.9 MATLAB script file: HHtype2.m which describes the modified HH equations (4.1)

a b
125. 125.
PD PD
100. 100.

75. 75.

50. 50.

25. DC 25. DC
HB2
0. 0.
HB1
HB1
–25. –25.
–50. 0. 50. 100. 150. 200. 250. 300. 0. 1. 2. 3. 4. 5.

Fig. 4.10 One-parameter bifurcation diagram of the HH equations (4.1) with respect to the bifur-
cation parameter Iext . Other parameter values are the same as those of Fig. 4.2. Hopf bifurcations
occur at HB1 (Iext D 1:934, eigenvalues: 0:000003132 ˙ 0:436584i , 0:0941944, 4:65870)
and at HB2 (Iext D 282:916, eigenvalues: 0:000000220614 ˙ 0:969227i , 0:181518, 14:7220).
(a) Whole diagram. (b) Magnification of (a). In both panels, abscissa and ordinate denote the
bifurcation parameter Iext and the membrane potential v, respectively

saddle-node bifurcation of periodic solution occurs and a pair of unstable periodic

solutions are generated. At the point PD, the period-doubling bifurcation occurs
and the stability of periodic solution changes. Panel (b) is the magnification of (a)
near HB1. Differently from the bifurcation diagram (Fig. 2.15) of the original HH
equations, there is no region where both a stable equilibrium point and a periodic
orbit co-exist. Despite such a difference, whole nature of both Figs. 2.15 and 4.10
are similar. Therefore, both solutions shown in Fig. 4.2 of the HH equations (4.1)
and the original HH equations (2.3) much resembled each other.
108 4 Whole System Analysis of Hodgkin–Huxley Systems

a b
50. 50.

40. 40.

30. 30.
PD PD
DC DC
20. 20.

10. HB2
10.
HB1 HB1
0. 0.
–50. 0. 50. 100. 150. 200. 250. 300. 0. 1. 2. 3. 4. 5.

Fig. 4.11 Period of periodic solutions shown in (a) Fig. 4.10a and (b) Fig. 4.10b. In both panels,
abscissa and ordinate denote Iext and the period, respectively

a b
125. 125.

100. 100.

PD PD
75. 75.

50. 50.

25. DC 25. DC
HB2
0. HB1 0.
HB1
–25. –25.
–50. 50. 150. 250. 350. 40. 45. 50. 55. 60. 65. 70. 75. 80.
0. 100. 200. 300.

Fig. 4.12 One-parameter bifurcation diagram of the HH equations (4.1). Only the value of vN n;1 has
been changed from 10.0 to 5.0 from Fig. 4.10. (a) Whole diagram. (b) Magnification of (a). In both
panels, abscissa and ordinate denote the bifurcation parameter Iext and the membrane potential v,
respectively

Figure 4.11 shows the period of periodic solutions shown in Fig. 4.10. We can
see that, also with respect to the period of periodic solution, the HH equations (4.1)
and the original HH equations (2.3) resemble each other (compare this figure with
Fig. 2.16). At the Hopf bifurcation point HB1, the equilibrium point of (4.1) pos-
sesses approximately pure imaginary eigenvalues: 0:000003132 ˙ 0:436584i . The
period of (unstable) periodic orbit which is close to HB1 is determined by the imag-
inary part as 2=0:436584  14:39. Please check that this period coincides with
that of Fig. 4.11.
Figures 4.12 and 4.13 are the similar bifurcation diagrams to Figs. 4.10 and 4.11,
where only the value of vN n;1 has been changed from 10.0 to 5.0. In spite of this very
small difference, the whole nature of both bifurcation diagrams Figs. 4.10 and 4.12
are completely different each other. Particularly, the position of HB1 has moved
much in right direction, whereas that of HB2 has moved slightly. Also, the range
between HB1 and PD has been expanded significantly. The amplitudes of periodic
solutions (repetitive firing) have been decreased much, while the periods of periodic
solutions are not so different between Figs. 4.11 and 4.13. This significant change
of bifurcation structure is the reason of the unexpectedly different firing shown in
4.2 Bifurcations of the Hodgkin–Huxley Neurons 109

a b
50. 50.

40. 40.

30. 30.

20. 20.
PD PD
DC DC
10. 10.
HB1 HB2
0. 0.
–50. 50. 150. 250. 350. 40. 45. 50. 55. 60. 65. 70. 75. 80.
0. 100. 200. 300.

Fig. 4.13 Period of periodic solutions shown in (a) Fig. 4.12a and (b) Fig. 4.12b. In both panels,
abscissa and ordinate denote Iext and the period, respectively

125.

100.

75.

50.

25.

0.

–25.
–100. 0. 100. 200. 300. 400. 500.

Fig. 4.14 One-parameter bifurcation diagram of the HH equations (4.1) with respect to the bi-
furcation parameter Iext . Other parameter values are the same as those of Fig. 4.4. Abscissa and
ordinate denote the bifurcation parameter Iext and the membrane potential v, respectively

Fig. 4.3e. (Note that the value Iext D 46:87 of Fig. 4.3e locates near HB1 in Fig. 4.12
where only unstable solutions exist. Strictly speaking, the solution in Fig. 4.3e is the
result of more complicated bifurcations taken place near HB1.)
Next, consider the case that the parameter values of the HH equations (4.1) are
chosen so that the functions x 1 .v/ and
x .v/ (x D m; n; h) of (4.1) could become
close to those of the original HH equations (2.3). Figure 4.14 shows the bifurcation
diagram where the parameter values are the same as those of Fig. 4.4 where no
firing occurs. We note again that both HH equations of (4.1) and (2.3) resemble
each other, as shown in Fig. 4.4. However, as shown in the bifurcation diagram of
Fig. 4.14, the HH equations (4.1) have neither Hopf bifurcation nor firing even when
we increase the value of Iext , whereas the original HH equations (2.3) show rich
firing phenomena.
Figures 4.15 and 4.16 respectively show the bifurcation diagram and the periods
of periodic solutions when only the value of vN n;1 is increased from 5 to 10.0. Hopf
bifurcations and repetitive firing (periodic solutions) have been recovered. Note,
however, that the range of the external current Iext where periodic solutions exist
has been shrunk much in contrast to the previous bifurcation diagrams shown in
Figs. 4.10 and 4.12. Again, we would like to emphasize that the difference of the
110 4 Whole System Analysis of Hodgkin–Huxley Systems

a b
125. 125.
100. 100.
PD PD
75. 75.
50. DC 50. DC

25. 25.
HB1
0. HB2 0.
HB1
–25. –25.
–50. 0. 50. 100. 150. 200. 250. 300. 0. 2. 4. 6. 8. 10.
1. 3. 5. 7. 9.

Fig. 4.15 One-parameter bifurcation diagram of the HH equations (4.1) with respect to the
bifurcation parameter Iext . Only the value of vN n;1 is increased to 10.0 from Fig. 4.14: All parameter
values except for Iext are the same as those of Fig. 4.5. (a) Whole diagram. (b) Magnification of
(a). In both panels, abscissa and ordinate denote the bifurcation parameter Iext and the membrane
potential v, respectively

a b
50. 50.

40. 40.

30. DC 30. DC

PD
20. 20.
PD
10. 10.
HB1 HB2 HB1
0. 0.
– 50. 0. 50. 100. 150. 200. 250. 300. 0. 2. 4. 6. 8. 10.
1. 3. 5. 7. 9.

Fig. 4.16 Period of periodic solutions shown in (a) Fig. 4.15a and (b) Fig. 4.15b. In both panels,
abscissa and ordinate denote Iext and the period, respectively

parameter value between Figs. 4.14 and 4.15 are very small, although the total be-
havior of both bifurcation diagrams is completely different: The solutions of the HH
equations (4.1) are very sensitive to the parameter vN n;1 .
In contrast to the above examples, next example will show us the insensitiv-
ity (robustness) and will make us very confusing! In the bifurcation diagram of
Fig. 4.17, only the value of vN n;2 has been increased to 40:0 from 12:0 of Fig. 4.15.
Comparing Fig. 4.17 with Fig. 4.15, we can see that the whole feature of the bi-
furcation diagrams are very similar each other irrespective of the big difference
of the vN n;2 values: 40 and 12. Of course, details of the diagrams have changed:
The amplitudes of periodic solutions decreased by the increase of vN n;2 . The loci
of Hopf bifurcation points HB1 and HB2 moved rightward slightly. The locus
of period-doubling bifurcation point PD moved rightward much. The periods of
periodic solutions decreased slightly, particularly in the small Iext region (Fig. 4.18).
Note that the parameter vN n;2 affects only the function
n .v/ in (4.1) and does not af-
fect the position of the equilibrium points of (4.1) since
n .v/ does not affect the
solution of dv=dt D dm=dt D dn=dt D dh=dt D 0 at all (please examine the differ-
4.3 Two-Parameter Bifurcation Analysis of the HH Equations 111

a b
125. 125.

100. 100.

75. 75.
PD
50. 50. PD
25. DC 25. DC
0. HB2 0.
HB1 HB1
–25. – 25.
–50. 0. 50. 100. 150. 200. 250. 300. 0. 5. 10. 15. 20. 25. 30.

Fig. 4.17 One-parameter bifurcation diagram of the HH equations (4.1) with respect to the bifur-
cation parameter Iext . Only the value of vN n;2 is increased to 40.0 from Fig. 4.15. (a) Whole diagram.
(b) Magnification of (a). In both panels, abscissa and ordinate denote the bifurcation parameter Iext
and the membrane potential v, respectively

a b
50. 50.

40. 40.

30. 30.

20. 20.
DC DC PD
PD
10. 10.
HB1 HB2 HB1
0. 0.
–50. 0. 50. 100. 150. 200. 250. 300. 0. 5. 10. 15. 20. 25. 30.

Fig. 4.18 Period of periodic solutions shown in (a) Fig. 4.17a and (b) Fig. 4.17b. In both panels,
abscissa and ordinate denote Iext and the period, respectively

ence of solid and broken curves between Figs. 4.17 and 4.15). However,
n .v/ does
affect the stability of equilibrium points. In fact, the loci of Hopf bifurcation points
have been changed. Of course,
n .v/ affects the amplitudes and periods of periodic
solutions. Anyway, the big change of
n .v/ value does not change the whole bifurca-
tion structure much: the HH equations (4.1) are not sensitive to
n .v/ and are robust.
So far, using several examples, we have explored and shown that the HH
equations (4.1) are sensitive to some parameters and insensitive to other parameters.
What makes this difference of sensitivity? The two-parameter bifurcation analysis
in the following section will clarify the reason of this difference.

4.3 Two-Parameter Bifurcation Analysis of the HH Equations

Figure 4.19 shows several two-parameter bifurcation diagrams of the HH

equations (4.1). The abscissa is the bifurcation parameter Iext similarly to the pre-
vious one-parameter bifurcation diagrams, while the ordinate denotes also another
112 4 Whole System Analysis of Hodgkin–Huxley Systems

a b
75. 75. 5
4
50. HB 50.
5 4
25. 25.
6
Vn1

Vn2
2
0. 0.
1 3
–25. –25. 6

–50. –50.
–50. 0. 50. 100. 150. 200. 250. –50. 0. 50. 100. 150. 200. 250.
c d
75. 75.
3
50. 50.
4
25. 25. 6 5
Vh1

Vh2
0. 0. 1
2
–25. –25.

–50. –50.
–50. 0. 50. 100. 150. 200. 250. –50. 0. 50. 100. 150. 200. 250.
e 75.
f 75.

50. 50.
4
3
25. 2 25.
Vm1

Vm2

0. 0.

–25. –25.

–50. –50.
–50. 0. 50. 100. 150. 200. 250. –50. 0. 50. 100. 150. 200. 250.

Fig. 4.19 Two-parameter bifurcation diagram of the HH equations (4.1). Abscissae are the
bifurcation parameter Iext for all panels. Ordinates are also bifurcation parameters: (a) vN n;1 (b) vNn;2
(c) vN h;1 (d) vN h;2 (e) vN m;1 (f) vN m;2 . Parameter values other than two bifurcation parameters for each
panel are the same as those of Fig. 4.5

bifurcation parameter. All parameters of (4.1) other than the two bifurcation
parameters in each panel are the same as those of Fig. 4.15 (Fig. 4.5). In the
panel (a), the ordinate is vN n;1 . The solid curve called a bifurcation curve denotes the
loci of Hopf bifurcations; if the values of parameters .Iext ; vN n;1 / are chosen on this
curve, a Hopf bifurcation occurs. The horizontal broken line shows the line with
vN n;1 D 10:0. If we change the parameter Iext along this line, the one-parameter
bifurcation diagram shown in Fig. 4.15 is obtained. We can confirm that the Iext
values of two intersections of the broken line and the bifurcation curve are 6.9 and
82.0 which are exactly the same as those of the two Hopf bifurcation points (HB1
and HB2) in the one-parameter bifurcation diagram (Fig. 4.15).
The horizontal dotted line shows the line with vN n;1 D 5:0. Since this line never
intersect the Hopf bifurcation curve, we cannot see any Hopf bifurcations when
Iext is changed along this line. The one-parameter bifurcation diagram along the
4.3 Two-Parameter Bifurcation Analysis of the HH Equations 113

dotted line is Fig. 4.14. In fact, no bifurcation occurs in Fig. 4.14. This is the reason
why two bifurcation diagrams in Figs. 4.14 and 4.15 (also two solutions in Figs. 4.4
and 4.5) were much different each other, in spite of the small difference of the vN n;1
values. The difference of the relative positions of the broken line (Nvn;1 D 10) and
the dotted line (Nvn;1 D 5) compared to the Hopf bifurcation curve is the cause of the
sensitivity of the HH equations (4.1) to the vN n;1 value.
Figure 4.19b is a similar bifurcation diagram to panel (a), but the ordinate is the
other bifurcation parameter vN n;2 . The broken line is the line with vN n;2 D 12. Since
this value of vN n;2 is the same as that of Fig. 4.14, the one-parameter bifurcation
diagram along this line is also Fig. 4.14. Confirm that the loci (Iext values) of the
two intersections of the broken line and the solid bifurcation curve coincide with
those of the Hopf bifurcations (HB1 and HB2) of Fig. 4.14 (and also coincide with
those of Fig. 4.19a). The dotted line is the line with vN n;2 D 40 which is the same
as that of Fig. 4.17. Thus, the one-parameter bifurcation diagram along this line is
Fig. 4.17. In Fig. 4.19b, the loci of two intersections of the dotted line and the HB
bifurcation curve are very similar to the loci of two intersections of the broken line
and the HB bifurcation curve. This is the reason why the two bifurcation diagrams
shown in Figs. 4.15 and 4.17 much resemble each other, in spite of the big difference
of the vN n;2 value. As seen from the way that the broken line intersects with the HB
bifurcation curve, the solutions and Hopf bifurcations of (4.1) are not sensitive but
are robust to the change of the parameter vN n;2 .
Other panels (c)–(f) are similar two-parameter bifurcation diagrams and their or-
dinates are vN h;1 , vN h;2 , vN m;1 , vN m;2 , respectively. On the broken line of each panel,
the value of the ordinate (the second bifurcation parameter) is the same as that
of Fig. 4.14, the loci (Iext values) of intersections of the broken line and the solid
bifurcation curve are the same in all panels. From these bifurcation diagrams, we
can find that the HH equations (4.1) are very sensitive to both vN h;1 and vN m;1 , but not
sensitive to both vN h;2 and vN m;2 . These results are consistent with the case of vN n;1 and
vN n;2 . Thus, the two-parameter bifurcation diagrams shown in Fig. 4.19 lead to an
important implication that the HH equations (4.1) are very sensitive to the functions
x 1 .v/, (x D m; n; h), but not sensitive to
x .v/, (x D m; n; h).
As explained in the above section, the period of (infinitesimally small) periodic
orbits bifurcated from an equilibrium point through a Hopf bifurcation can be de-
termined by the imaginary parts of complex eigenvalues at the equilibrium point.
Figure 4.20 shows such periods along the Hopf bifurcation curves in Fig. 4.19. The
abscissa is the bifurcation parameter Iext similarly to Fig. 4.19 while the ordinate
shows the period. For example, at the point labeled as 6 of Fig. 4.19a, the Hopf
bifurcation curve terminates. From Fig. 4.20a, we can see that the period grow up
rapidly (to infinity possibly) near the termination point (the point labeled as 6 is
outside the area of Fig. 4.20a). In Fig. 4.19b, the Hopf bifurcation curve is a closed
curve, while the curve in Fig. 4.20b is not. This means that the periods of periodic
orbits along the Hopf bifurcation curve in Fig. 4.19b do not depend on the value of
vN n;2 but do depend on only the Iext value of the abscissa.
114 4 Whole System Analysis of Hodgkin–Huxley Systems

a b
50. 50.
Vn1 Vn2
40. 40.

30. 30.
Period

20. 20.
2
10. 3 1 5 6
2 1 10. 3

0. 4
0.
–50. 0. 50. 100. 150. 200. 250. –50. 0. 50. 100. 150. 200. 250.
c 50.
d 50.
Vh1 Vh2
40. 40.
3 2
30. 30.
4
20. 20.
1
2
10. 10. 6
3 5

4
0. 0.
–50. 0. 50. 100. 150. 200. 250. –50. 0. 50. 100. 150. 200. 250.
e f
50. 50.
Vm1 Vm2
40. 40.

30. 30.

20. 20. 5
1 11
10. 10. 6 12 8 9

0. 0.
–50. 0. 50. 100. 150. 200. 250. –50. 0. 50. 100. 150. 200. 250.

Fig. 4.20 Periods at the Hopf bifurcations of the two-parameter bifurcation diagram Fig. 4.19.
Abscissae are the bifurcation parameter Iext and ordinates are the period. Each panel corresponds
to that of Fig. 4.19

4.4 Numerical Bifurcation Analysis by XPPAUT

In this section, we briefly explain how to execute the two-parameter bifurcation

analyses shown in the above section, using the XPPAUT which was also briefly
explained in Sect. 1.5. Figure 4.21 shows the HHtype.ode file to analyze the
HH equations (4.1). In order to obtain the two-parameter bifurcation diagram of
Fig. 4.19b, for example, perform the following steps:
1. Start up XPPAUT and open (or make) the HHtype.ode file shown in Fig. 4.21.
2. Run the commands Initialconds and (G)o, then Initialconds and
(L)ast several times to obtain a solution sufficiently converged to a stable
equilibrium point.
3. Use File and Auto commands to open the AUTO window.
4.4 Numerical Bifurcation Analysis by XPPAUT 115

# modified Hodgkin-Huxley equation HHtype.ode

init v=-0.015124 m=0.083057 h=0.45546 n=0.35414

Minf = 1/( 1+exp(-sm*(v-vm1)) )

Mtau = tm/cosh( sm*(v-vm2)/2 )
Hinf = 1/( 1+exp(-sh*(v-vh1)) )
Htau = th/cosh( sh*(v-vh2)/2 )
Ninf = 1/( 1+exp(-sn*(v-vn1)) )
Ntau = tn/cosh( sn*(v-vn2)/2 )

v’ = (Gna*m^3*h*(Vna-v) + Gk*n^4*(Vk-v) + Gl*(Vl-v) + Iext)/C

m’ = (Minf-m)/(Mtau)
h’ = (Hinf-h)/(Htau)
n’ = (Ninf-n)/(Ntau)

par Iext=0 C=1

par Gna=120, Gk=36, Gl=0.3
par Vna=115, Vk=-12, Vl=10.599

par sm=0.1, vm1=24.0,vm2=24.0, tm=0.5

#par sh=-0.09, vh1=-2.0,vh2=-2.0, th=12.0

par sh=-0.13 vh1=-2.0 vh2=-2.0 th=8.5

#par sn=0.06, vn1=10.0, vn2=10.0, tn=5.0

par sn=0.055 vn1=10.0 vn2=-12.0 tn=5.7

@ xplot=t,yplot=v
@ total=100,dt=.03,xlo=0,xhi=100,ylo=-20,yhi=120
@ meth=runge-kutta
@ bound=200
@ ntst=50 nmax=1000 npr=500 parmin=0 parmax=100 dsmax=2
done

Fig. 4.21 The ode file of XPPAUT: HHtype.ode

4. Use Parameter command to open the Parameters window where we change

the second parameter *Par2 (which corresponds to the ordinate of Fig. 4.19b)
from C to vn2.
5. Use Numerics to change several parameter values for numerical bifurcation
analysis: Change Nmax (maximum number of computational steps) from 1,000
to 250 (this may be sufficient), Ds (usual step size of changing the bifurcation
parameter) from 0.02 to 0.1 (this is small enough), both Par Min and Par Max
(the range of changing the main (first) bifurcation parameter (Iext )) from “0 and
100” to “50 and 250” (the same scale as Fig. 4.19b). Then the window will
become as Fig. 4.22. Next, click on Ok.
116 4 Whole System Analysis of Hodgkin–Huxley Systems

Fig. 4.22 AUTO Numerics Window

6. Use Axes and hI-lo to change the range of both abscissa and ordinate of the
graph: Xmin:-50, Ymin:-20, Xmax:250, Ymax:120.
7. Run and Steady state to obtain a one-parameter bifurcation diagram of
equilibrium points.
8. Use Grab and tab (or !, keys on a keyboard) to move the cursor on the
Hopf bifurcation point (usually labeled by 2) of the bifurcation diagram. Then
hit return or enter key to grab the point.
9. Use Run and Two Param to obtain a two-parameter bifurcation diagram.
Finally, use Axes and Two par to change *Y-axis from V to vn2, Ymin
from 20 to 50, Ymax from 120 to 75. Then, reDraw gives us the two-
parameter bifurcation diagram shown in Fig. 4.23.
10. Note that, if you failed in some of the above steps, close the AUTO window
and return to Step2. Then, pay attention to the fact that the values of the pa-
rameters as well as those of the variables v, m, h, n have been changed by
bifurcation analysis: Reset the values to initial or desired values. Also note that
when we return to the AUTO window, previous results still remain: Reset all re-
sults of the previous bifurcation analysis by File and Reset diagram (and/or
Clear grab) in the AUTO window. If you still fail, return to Step1: quit XP-
PAUT and restart it!

Exercise 4.2. Using the XPPAUT and the HHtype.ode file shown in Fig. 4.21, try
the following exercises:
 Draw the one-parameter bifurcation diagram (bifurcation parameter is Iext ) such
as those shown in Fig. 4.15.
 Changing minimum and maximum values of axes, magnify the one-parameter
bifurcation diagram near the Hopf bifurcation.
 Return to the XPP main window, then obtain several solutions of the HH
equations with changing the value of Iext (draw not only “v vs. t” plot but also
4.4 Numerical Bifurcation Analysis by XPPAUT 117

Fig. 4.23 Two-parameter bifurcation diagram made by XPPAUT similar to Fig. 4.19b

“m vs. v” plot and “h vs. v” plot, etc.). When you make these simulations, consult
the (magnified) one-parameter bifurcation diagram.
 Changing the value of parameters (for example vn1), repeat above steps.
 Try to make the two-parameter bifurcation diagram of Fig. 4.23, then try to make
other two-parameter bifurcation diagrams shown in Fig. 4.19. 
Chapter 5
Hodgkin–Huxley-Type Models of Cardiac
Muscle Cells

Following the HH formalism introduced in Chap. 2, various kinds of HH-type

models of neurons and other excitable cells are proposed (Canavier et al. 1991;
Chay and Keizer 1983; Cronin 1987; Gerber and Jakobsson 1993; Hayashi and
Ishizuka 1992; Keener and Sneyd 1998; Noble 1995; Rinzel 1990; Traub et al.
1991), and are analyzed (Alexander and Cai 1991; Av-Ron 1994; Bertram 1994;
Bertram et al. 1995; Butera 1998; Canavier et al. 1993; Chay and Rinzel 1985;
Doi and Kumagai 2005; Guckenheimer et al. 1993; Maeda et al. 1998; Rush and
Rinzel 1994; Schweighofer et al. 1999; Terman 1991; Tsumoto et al. 2003, 2006;
Yoshinaga et al. 1999). The HH-type equations include many variables depending
on the number of different ionic currents and their gating variables considered in the
equations, whereas the original HH equations possess only four variables (a mem-
brane voltage, activation and inactivation variables of NaC current and an activation
variable of KC current). Among the diverse family of HH-type equations, this chap-
ter explores the dynamics and the bifurcation structure of the HH-type equations of
heart muscle cells (cardiac myocytes).

5.1 Action Potentials in a Heart

A heart repeats contraction and relaxation, and these motions are controlled by
electrical signals (action potentials). Periodic electrical signals are generated in
sino-atrial node (cardiac pacemaker) and they are propagated to the whole heart.
The regular generation of the periodic electrical signals and their regular propaga-
tions (conductions) are essential for the pumping function of the heart. Figure 5.1
shows various action potentials of different parts of heart. The action potential wave-
forms differ part by part in the heart. Therefore, the waveform itself is important for
the normal function of the heart (Noble 1975).
Figure 5.2 shows a typical action potential waveform in ventricular myocardial
cell with five “phases.” Phase 0 is called the rapid depolarization phase. In this
phase, NaC currents flow into the cell, and the membrane potential is increased to
positive value rapidly. KC currents outflow from the cell in phase 1 called the over-
shoot and early repolarization phase. Phase 2 is kept by a balance between inward

S. Doi et al., Computational Electrophysiology, 119

DOI 10.1007/978-4-431-53862-2 5,  c Springer 2010
120 5 Hodgkin–Huxley-Type Models of Cardiac Muscle Cells

sinoatrial
node

atrial

atrioventricular
node

Purkinje

ventricular

Fig. 5.1 Schematic illustrations of the heart and various action potentials of different parts of the
heart

Na+ Ca2+ K+ K+
out
cell membrane
in
40
V (mV)

phase 4 phase 4
–100
phase 0 phase 1 phase 2 phase 3

0 time (msec) 500

Fig. 5.2 Schematic illustration of an action potential waveform in a ventricular myocardial cell.
In each phase of the action potential, specific ions move between the inside and outside of the cell
membrane

Ca2C current and outward KC current, and it is called the plateau phase. Phase 3 and
phase 4 are due to outward KC current, and they are called the repolarization phase
and the resting potential phase, respectively. The period from phase 0 to phase 3 is
called a refractory period. During this period, it is difficult or impossible for the cell
to respond to external stimuli.
5.2 Pacemaker Cell Model 121

5.2 Pacemaker Cell Model

5.2.1 The YNI Model of Cardiac Pacemaker Cell

At first, let us consider a Hodgkin–Huxley-type model of cardiac pacemaker cells

(sinoatrial-node cells). The Yanagihara–Noma–Irisawa (YNI) model (Yanagihara
et al.1980) of sinoatrial-node cells is a typical (and rather classical) Hodgkin–
Huxley-type model of cardiac cells and is described as follows:

dV 1
D  .INa C Is C Ih C IK C Il /; (5.1a)
dt C
dx
D ˛x .V /.1  x/  ˇx .V /x; .x D m; h; d; f; q; p/; (5.1b)
dt

where V is the membrane potential and x (D m; h; d; f; q; p) are the gating vari-

ables. The YNI model is described by differential equations with seven variables.
In (5.1a), there are five ionic currents: INa is the sodium current, Is is the slow inward
current, Ih is the hyperpolarization-activated current, IK is the potassium current and
Il is the leak current. (Notice that an inward current through ionic channels is de-
noted as a “negative current” in the YNI model. See the minus sign of the r.h.s. of
(5.1a).) Figure 5.3 illustrates these ionic currents or ionic channels. Note that, differ-
ently from the original HH model (2.3) of squid giant axon, there are five kinds of
different ionic currents considered. Equations which describe the five ionic currents
are as follows:

INa D cNa GNa m3 h.V  30/; GNa D 0:5I

   
V  30
Is D cs Gs .0:95d C 0:05/.0:95f C 0:05/ exp 1 ; Gs D 12:5I
15
Ih D ch Gh q.V C 45/; Gh D 0:4I

INa IK

Fig. 5.3 Schematic diagram

of the Yanagihara–Noma–
Irisawa (YNI) model (5.1)
I1 Is Ih
which has five ionic channels
122 5 Hodgkin–Huxley-Type Models of Cardiac Muscle Cells

exp .0:0277.V C 90//  1

IK D cK GK p ; GK D 0:6I
exp .0:0277.V C 40//
  
V C 60
Il D cl Gl 1  exp  ; Gl D 0:8I
20

where the parameters cNa , cs , ch , cK , cl are the coefficients which change the con-
ductances of ionic currents.
The functions ˛x .V / and ˇx .V / (x D m; h; d; f; q; p) in (5.1b) are as follows:
 
V C 37 V C 62
˛m .V / D  V C37  ; ˇm .V / D 40 exp  I
1  exp  10 17:8
 
3 V C 20 1
˛h .V / D 1:209  10 exp  ; ˇh .V / D  I
6:534 1 C exp  V C30
10
1:045  102 .V C 35/ 3:125  102 V
˛d .V / D  V C35  C  V I
1  exp  2:5 1  exp  4:8
4:21  103 .V  5/
ˇd .V / D  5  I
1  exp V2:5
3:55  104 .V C 20/ 9:44  104 .V C 60/
˛f .V / D  C20  ; ˇf .V / D  C29:5  I
1  exp V5:633 1 C exp  V 4:16
3:4  104 .V C 100/
˛q .V / D  V C100  C 4:95  105 I
exp 4:4  1
5  104 .V C 40/
ˇq .V / D   C 8:45  105 I
1  exp  V C40
6
9  103 2:25  104 .V C 40/
˛p .V / D  C3:8  C 6  104 ; ˇp .V / D  C40  :
1 C exp  V9:71 1  exp V13:3

Let us examine the dynamics and bifurcation structure by XPPAUT. The ode file
of the YNI model (5.1) is shown in Fig. 5.4. When we start XPPAUT and choose
the YNI.ode file, the main window will appear. If we click on the Initialconds
button and then click on the (G)o button, a waveform of “V vs T” will be drawn as
Fig. 5.5. Notice that, since the range where the variable T should be changed, is set
to a large value of 2,000 as a default value (the statement “total=2000” in the ode
file), a message “storage full” may appear. If we click on Yes (possibly several
times), then we can go on the computation (or if we increase the value of dt, the
messages may disappear or the number of messages may decrease). As shown in
Fig. 5.5, the YNI model shows repetitive spiking (periodic solution) and its period
is about 380 ms. In the following, let us examine how such a pacemaker activity of a
heart pacemaker cell appears by using bifurcation analysis. Pacemaker activity con-
trols the rhythmic motion of a heart. Thus, the period of the pacemaker activity is
essential for the normal function of the heart and for our life. In the following, we
also examine how the period varies under the changes of parameter values.
5.2 Pacemaker Cell Model 123

# YNI model YNI.ode

#
dV/dt=-1.0*(INa+IK+Il+Is+Ih-Iext)/C
dm/dt=alpha_m*(1.0-m)-beta_m*m
dh/dt=alpha_h*(1.0-h)-beta_h*h
dp/dt=alpha_p*(1.0-p)-beta_p*p
dd/dt=alpha_d*(1.0-d)-beta_d*d
df/dt=alpha_f*(1.0-f)-beta_f*f
dq/dt=alpha_q*(1.0-q)-beta_q*q
#
INa=cNa*GNa*m*m*m*h*(V-30)
IK=cK*GK*p*(exp(0.0277*(V+90))-1)/(exp(0.0277*(V+40)))
Il=cl*Gl*(1-exp((V+60)/(-20)))
Is=cs*Gs*(0.95*d+0.05)*(0.95*f+0.05)*(exp((V-10)/15)-1)
Ih=ch*Gh*q*(V+45)
#
alpha_m=(V+37)/(1-exp((V+37)/(-10)))
beta_m=40*exp((V+62)/(-17.8))
alpha_h=1.209*0.001*exp((V+20)/(-6.534))
beta_h=1/(1+exp((V+30)/(-10)))
alpha_p=9*0.001/(1+exp((V+3.8)/(-9.71)))+6*0.0001
beta_p=(-2.25*0.0001*(V+40))/(1-exp((V+40)/13.3))
alpha_d=1.045*0.01*(V+35)/(1-exp((V+35)/(-2.5)))
+3.125*0.01*V/(1-exp(V/(-4.8)))
beta_d=(-4.21*0.001*(V-5))/(1-exp((V-5)/2.5))
alpha_f=-3.55*0.0001*(V+20)/(1-exp((V+20)/5.633))
beta_f=(9.44*0.0001*(V+60))/(1+exp((V+29.5)/(-4.16)))
alpha_q=3.4*0.0001*(V+100)/(-1+exp((V+100)/4.4))+4.95*0.00001
beta_q=5*0.0001*(V+40)/(1-exp((V+40)/(-6)))+8.45*0.00001
#
param cNa=1, cK=1.0, cl=1.0, cs=1.0, ch=1.0
param C=1, Iext=0
param GNa=0.5, GK=0.7, Gl=0.8, Gs=12.5, Gh=0.4
V(0)=-50.0, m(0)=0.0, h(0)= 0.0
p(0)=0.0, d(0)=0.0, f(0)=0.0, q(0)= 0.0
#
@ total=2000,bound=2000,dt=0.05,dtmin=1e-6,xhi=2000
@ yhi=50,ylo=-100,meth=runge-kutta
@ autoxmin=-2,autoxmax=5,autoymin=-100,autoymax=50
@ ntst=50 nmax=250 npr=200 parmin=-2 parmax=5 dsmax=0.5 ds=0.1
#
done

Fig. 5.4 The YNI.ode file

124 5 Hodgkin–Huxley-Type Models of Cardiac Muscle Cells

Fig. 5.5 Repetitive action potentials of the YNI model

5.2.2 Bifurcation Analysis of the YNI Model

If we want to make a one-parameter bifurcation diagram, we need to change the

parameter values, since at the present parameter values, the YNI model does not
possess any stable equilibrium points but have a stable periodic solution. Let us
click on the Param button in the top row, then a new window will appear as Fig. 5.6.
We change the value of cNa to 2:0. Click on the Initialconds button and then
click on the (G)o button. After it finished, we click on the Initialconds button
and then click on the (L)ast button several times in order to obtain a sufficient
convergence to a stable equilibrium point. Finally, we click on the File button and
then click on the Auto button, a AUTO window will appear.
If we click on Run and Steady state buttons in the AUTO window, a one-
parameter bifurcation diagram will be drawn as Fig. 5.7 where the bifurcation
parameter on the horizontal axis is cNa . Next, click on Grab button and a cross (cur-
sor) will appear on the one-parameter bifurcation diagram. We use Tab and then
enter (return) keys to move to and grab the Hopf bifurcation (HB) point. The
information about the point will be displayed below the one-parameter bifurcation
diagram. Here we select the HB point whose label is 5.
Next, let us explain how to label specific solutions (for example, periodic
solutions which have a specific period) in a bifurcation diagram. We click on
Usr period button and select 4 button to input four user-specified conditions.
A new window of AutoPer will appear and we set the values of Uzr1, . . . , Uzr4,
as cNa D 1.0, T D 300, T D 500, T D 700, as shown in Fig. 5.8. Finally, if we click
5.2 Pacemaker Cell Model 125

Fig. 5.6 The Parameters

window

Fig. 5.7 The one-parameter bifurcation diagram of equilibrium points

on Run and Periodic buttons, a bifurcation diagram of both equilibrium points

and periodic orbits with several labels will appear as Fig. 5.9. The abscissa is the
bifurcation parameter cNa and the ordinate is the membrane potential V . The points
whose labels are 10, 11, 12 and 13 are the points we have specified; at the point 11,
the values of cNa is 1.0, and the periods of periodic orbits at the points 10, 12 and
13, are 300, 500 and 700, respectively. Also, we can see several information about
the solutions (periods and values of parameters) below the bifurcation diagram if
we move to the points by Grab button and Tab key.
126 5 Hodgkin–Huxley-Type Models of Cardiac Muscle Cells

Fig. 5.8 The AutoPer window

Fig. 5.9 The one-parameter bifurcation diagram of equilibrium points and periodic orbits with
user-specified labels

In the bifurcation diagram of Fig. 5.9, there are two Hopf bifurcation points
(labeled 2 and 5). There are also two saddle-node bifurcation points labeled 3 and 4
which are denoted as LP (Limit Point) in the AUTO. The points with labels 7 and 8
are the double-cycle bifurcation points (or saddle-node bifurcation points of peri-
odic orbits) which are also denoted as LP in the AUTO. The pacemaker activity
(repetitive spiking) bifurcates from an equilibrium points through the right Hopf bi-
furcation labeled 5. The periods of these periodic solutions increase as the parameter
value of cNa decreases. The pacemaker activity disappears near the left Hopf bifur-
cation point and the saddle-node bifurcation point 3. Therefore, pacemaker cells
cannot show regular pacemaker activity if cNa takes a large value or a negative value
5.2 Pacemaker Cell Model 127

(note that the negative value is physiologically impossible because this denotes a
conductance). Next, we examine how the Hopf bifurcation points move if we change
the values of other parameters; we make two-parameter bifurcation diagrams.

5.2.3 Two-Parameter Bifurcation Analysis of the YNI Model

In order to make a two-parameter bifurcation diagram of the HB points, we Grab the

HB point labeled as 5. Then we click on Axes and Two par buttons. A new window
as shown in Fig. 5.10 will appear, and we set as follows: Ymin=-0.5, Ymax=2.5.
If we click on Run and Two Param buttons, we can obtain a part of a two-parameter
bifurcation diagram as shown in Fig. 5.11. Since the Hopf bifurcation curve imme-
diately escapes the area of the diagram, only a small portion of bifurcation curve
appears. Note that the second parameter on the vertical axis is cK (which is preset as
the default second parameter in the YNI.ode file). If we want to change this param-
eter, click on the Parameter button and then change the value of *Par2 from cK
to any other parameters we like. In order to obtain a whole Hopf bifurcation curve,
we reverse the tracing direction of the Hopf bifurcation curve. To do so, we click on
Numerics button to change Ds to -0.1. Then, we Grab the HB point whose label is
5 again. Clicking on Run and Two Param buttons, a full part of the two-parameter
bifurcation diagram is obtained as Fig. 5.12.
Next, let us calculate a two-parameter bifurcation diagram of another HB point la-
beled as 2 in Fig. 5.7 or Fig. 5.9. To do so, click on Grab button, then following the
information below the two-parameter bifurcation diagram, we grab the HB point 2.
Next, we set back the value of Ds to 0.1 using Numerics button and then click on
Run and Two Param buttons. Again, change the value of Ds to -0.1, then clicking
on Run and Two Param buttons will give us the two parameter bifurcation diagram

Fig. 5.10 AutoPlot window

128 5 Hodgkin–Huxley-Type Models of Cardiac Muscle Cells

Fig. 5.11 Partial two-parameter bifurcation diagram

Fig. 5.12 Whole two-parameter bifurcation diagram

shown in Fig. 5.13. Note that in this bifurcation diagram the previous bifurcation
curve traced from the HB point 5 is also superimposed (remained). In the parameter
region between the two Hopf bifurcation curves, the YNI model shows repetitive
spiking or pacemaker activity. Next, we examine how the period of pacemaker ac-
tivity changes in such a parameter region.
5.2 Pacemaker Cell Model 129

Fig. 5.13 Two Hopf bifurcation curves started from the HB points 5 and 2

Fig. 5.14 A contour plot on periodic solutions which have a specific period of 380.1 ms

Let us draw a contour plot on the period of periodic solutions. Grab the point
whose label is 11 (the point of cNa D 1.0) following the information below the bi-
furcation diagram, as explained before. Then, clicking on Run and Fixed Period
buttons, a part of the contour plot will be drawn as Fig. 5.14. Along this curve
130 5 Hodgkin–Huxley-Type Models of Cardiac Muscle Cells

Fig. 5.15 Reverse continuation of the contour plot of Fig. 5.14

between the two Hopf bifurcation curves, all periodic solutions have the same
period (380.1 ms) as that of the point labeled by 11 in Fig. 5.9. Grab the point
labeled as 11 again. Also, we reverse the sign of Ds using Numerics button. Click-
ing on Run and Fixed Period buttons gives us the full part of the contour plot as
shown in Fig. 5.15. We can see that the contour plot extends and terminates near the
lower Hopf bifurcation curve.
For the other points which we have labeled as 10, 12 and 13 by using
Usr period button in the above section, we can obtain the contour plots simi-
larly. The final result is shown in Fig. 5.16 where the four contour lines between the
two Hopf bifurcation curves correspond to the four specific periods: 300, 380, 500
and 700 ms (leftward). We can see that the distance of two contour lines becomes
narrower near the upper Hopf bifurcation curve. This leads an important implication
that the period of pacemaker activity is very sensitive to a parameter variation near
the (upper) Hopf bifurcation curve.
Exercise 5.1. Following the explanation of this section, try to reproduce the two-
parameter bifurcation diagram of Fig. 5.16 (please change the specification of the
period such as 300, 500, 700 ms in Fig. 5.16 to other values whatever you like).
Also, changing the bifurcation parameters on the abscissa and the ordinate, try to
compute other two-parameter bifurcation diagrams. 

5.2.4 Bifurcation and Parameter Sensitivity

Figure 5.17 is the more detailed bifurcation diagram of Fig. 5.16 where the
bifurcation parameters are cNa and cK same as Fig. 5.16. This bifurcation diagram
5.2 Pacemaker Cell Model 131

Fig. 5.16 Contour plot on periodic solutions which have specific periods: 300, 380, 500 and
700 ms

was computed by AUTO instead of XPPAUT, and more bifurcation curves are
included than Fig. 5.16. The labels HB, SN, DC, PD and HC mean the bifurcation
points of Hopf, saddle-node, double-cycle, period-doubling and homoclinic bifur-
cations, respectively. The curve labeled with “normal” denotes the contour curve
of period 380 ms and other numbers denote the periods of other periodic orbits.
The special point labeled as BT denotes the Bogdanov–Takens bifurcation point
where three bifurcation curves of Hopf, saddle-node and homoclinic bifurcation
meet together.
The bifurcation curves of HB1 and HB2 separate Fig. 5.17 into three areas.
In area 2, various periodic solutions exist. When .cNa ; cK / takes the values near
(1.0, 0.0), periodic solutions with long period exist, and these solutions correspond
to sinus bradycardia. The period becomes small when cNa is increased, and it be-
comes large when cK is increased. Panels (c) and (e) show two abnormal waveforms
of membrane potential when cNa takes a small and a large value (cK is fixed to 1.0),
respectively. If we want to get the normal period 380 ms in such abnormal cases of
cNa , the value of cK should be adjusted as shown in the panels (d) and (f ). Panels (a)
and (b) show the typical waveforms in area 1 and area 3, respectively. Both of the
membrane potentials converge to an equilibrium point eventually and cannot show
repetitive action potential (thus, abnormal case).
If cK is fixed to 1.0 and cNa is varied (this corresponds to a horizontal line with
cK D 1 in Fig. 5.17), the “one-parameter” bifurcation diagram on the parameter cNa
can be obtained as shown in Fig. 5.18, in which the value of V in the steady state was
plotted for each value of cNa . The solid and broken curves show stable and unstable
132 5 Hodgkin–Huxley-Type Models of Cardiac Muscle Cells

c
cNa = 0.50, cK = 1.00
50
25

V (mV)
a b 0
–25
cNa = 0.00, cK = 2.50 cNa =4.00, cK =0.25 –50
50 50
25 25 –75
–100

V (mV)
V (mV)

0 0
– 25 – 25 0 500 1000 1500 2000
– 50 – 50 t (msec)
– 75 – 75 d cNa = 0.50, cK = 0.62
– 100 – 100 50
25

V (mV)
0 500 1000 1500 2000 0 500 1000 1500 2000
0
t (msec) t (msec) –25
–50
–75
2.5 –100
700 500 0 500 1000 1500 2000
PD3 area 2 t (msec)
2.0
al

area 1 e cNa = 2.50, cK = 1.00

rm

HB2 DC1 50
no

1.5 25

V (mV)
HC1,2 PD2
350 300 0
cK

1.0 SN1 –25

–50
0.5 DC2,PD1 –75
BT –100
HB1 0 500 1000 1500 2000
0.0 DC3 t (msec)
area 3 f
SN2 cNa = 2.50, cK = 2.10
– 0.5 50
– 2.0 – 1.0 0.0 1.0 2.0 3.0 4.0 5.0 25
V (mV)

cNa 0
–25
HB SN DC PD HC –50
–75
–100
0 500 1000 1500 2000
t (msec)

Fig. 5.17 Two-parameter bifurcation diagram of the YNI model as for the two bifurcation
parameters cNa and cK . (a–f) Examples of membrane potential waveforms

equilibrium points, respectively. The symbols  and ı show the maximum values
of V of stable and unstable periodic solutions, respectively. Periods of periodic
solutions are also shown in the diagram. In normal condition (cNa D 1:0), a sta-
ble periodic solution whose period is about 380 ms exists, and panel (c) shows the
corresponding waveform of membrane potential V .
For each value of cNa between HB1 and HB2, a periodic solution (stable or unsta-
ble) exists. The period of periodic solution varies with cNa . When cNa is increased,
the period decreases, and thus the heart rate increases. In general, a very high
heart rate (>325 beats/min) corresponds to sinus tachycardia, and a very low heart
rate (<130 beats/min) corresponds to sinus bradycardia (note that the YNI model
is the cardiac pacemaker cell model of a rabbit). Figure 5.18b,d show two typical
waveforms of membrane potential, whose periods are large and small, respectively.
For the treatment of arrhythmia, it is important to consider the drug sensitivity of ion
channels. In Fig. 5.18, the variation of period is small when cNa is increased from
5.2 Pacemaker Cell Model 133

a cNa = 0.0 b cNa = 0.5 c cNa = 1.0

50 50 50
25 25 25

V (mV)

V (mV)
V (mV)

0 0 0
– 25 –25 – 25
– 50 –50 – 50
– 75 –75 – 75
– 100 – 100 – 100
0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000
t (msec) t (msec) t (msec)
d cNa = 2.5
50 50
380(normal) 25

V (mV)
500 260 PD2 0
25 700 350 300 – 25
600 PD1
800 DC2 – 50
0 – 75
HC1 400 DC1
V (mV)

– 100
– 25 SN1 PD3 0 500 1000 1500 2000
t (msec)
SN2 HB1 e cNa = 5.0
– 50 50
HC2 HB2 25

V (mV)
– 75 0
– 25
– 100 – 50
– 1.0 0.0 1.0 2.0 3.0 4.0 5.0 – 75
cNa – 100
0 500 1000 1500 2000
t (msec)

Fig. 5.18 One-parameter bifurcation diagram as for the bifurcation parameter cNa obtained by
AUTO. (a–e) Examples of membrane potential waveforms

1.0 (normal value), and it is large when cNa is decreased from 1.0. In particular,
when cNa takes a value near 0.25, the period changes drastically. These results show
that the drug sensitivity in the case of a small value of cNa is stronger than that in
the case of a large value of cNa .
In both the left side of HB2 and the right side of HB1, only equilibrium points
exist. Because of the abnormality of NaC channel there (cNa is too small or too
large), pacemaker cells cannot generate action potentials periodically and continu-
ously. The typical waveforms of membrane potential in the two cases are shown in
panels (a) and (e), respectively. Both of the membrane potentials converge to the
equilibrium points, but the values of membrane potential are different in the two
cases (one is low [repolarized] and the other is high [depolarized]).
Since only unstable periodic solutions and unstable equilibrium points were de-
tected by AUTO for the values of cNa between PD2 and DC1 in Fig. 5.18, we
also computed the one-parameter bifurcation diagram by numerical simulations
(Fig. 5.19) for the parameter values of cNa between 3.6 and 4.0. In this diagram,
both the local maximum and minimum values of V for each value of cNa were plot-
ted. There are many bifurcations and possibly chaotic solutions. The waveforms
of membrane potentials when cNa D 3:7 and 3.8 are shown in panels (a) and (b),
respectively. In both cases, the amplitude of membrane potential varies with time,
which shows serious abnormalities in the action potential generation.
134 5 Hodgkin–Huxley-Type Models of Cardiac Muscle Cells

a c Na =3.7
50
50 25

V (mV)
0
25 PD2 –25
–50
DC1 –75
0
– 100
V (mV)

6000 8000 10000

– 25 t (msec)
b cNa = 3.8
– 50 50
25

V (mV)
– 75 0
– 25
– 100 – 50
3.6 3.7 3.8 3.9 4.0 – 75
– 100
cNa 6000 8000 10000
t (msec)

Fig. 5.19 One-parameter bifurcation diagram obtained by numerical simulations. (a, b) Examples
of membrane potential waveforms

5.3 Ventricular Cell Model

5.3.1 The LRd Model of Ventricular Cell

The YNI model is a relatively simple and old HH-type model of cardiac pacemaker
cells. In this section, we briefly explain the bifurcation structure of more detailed
cardiac cell model. The Luo–Rudy dynamic (LRd) model (Luo and Rudy 1994)
is one of such detailed cardiac cell models and is a model of cardiac ventricular
cell. “Ingredients” of the LRd model are schematically illustrated in Fig. 5.20 and
Table 5.1. Comparing with the YNI model of a pacemaker cell in Fig. 5.3, we can
see that many factors are taken into account in the LRd model. For example, there
are different types of Ca2C channels such as ICa.L/ , ICa.T/ and ICa;b , which flow
the same ion Ca2C but have different characteristics. Other than ionic channels,
the LRd model includes “pump” and “exchanger.” INaCa denotes the electric cur-
rent of the Na–Ca exchanger which simultaneously moves 3NaC ions into the cell
(down its electrochemical gradient) and 1Ca2C ions out of the cell (up its electro-
chemical gradient); as a whole, INaCa is an inward current in its normal mode of
operation. Energy of the movement of Ca2C against its electrochemical gradient is
provided by the movement of NaC ions down its electrochemical gradient. INaK de-
notes the Na–K pump which moves NaC out of the cell and KC into the cell. Both
ions move up their electrochemical gradients and energy for this movement comes
from the hydrolysis of ATP.
The flow of KC out of the cell and the flow of NaC and Ca2C into the cell down
their electrochemical gradients through ion channels gradually change the intracel-
lular ionic concentrations and destroy their concentration gradients across the cell
membrane. Both Na–Ca exchanger and Na–K pump maintain the concentration gra-
dients. (The changes of intracellular ionic concentrations are very slow. Therefore,
5.3 Ventricular Cell Model 135

INaCa Ip(Ca)
INa,b ICa(L) ICa(T)
ICa,b
INa

Irel Sarcoplasmic Reticulm

JSR NSR
Troponin Itr
Iup
Ileak
Calsequestrin
Calmodulin

Ins(Ca)
IKr IK(ATP)
IKs IK1 IKp IK(Na)
Ito INaK

Fig. 5.20 Schematic diagram of the LRd model. This is a modification from Luo and Rudy (1994),
and details are provided in the reference. This model includes many ionic channels, ionic pumps,
exchangers, and Ca2C buffers

in the YNI model, Na–Ca exchanger and Na–K pump are not taken into account,
and the ionic concentrations are treated as constants.) In the LRd model, all ionic
concentrations are considered not as constants but variables. In particular, the intra-
cellular concentration of Ca2C is modulated not only by ionic channels and Na–Ca
exchangers but also by the “Ca2C buffers” such as troponin, calmodulin and calse-
qeustrin. Taking all ingredients in Fig. 5.20 into account, the LRd model is described
by the nonlinear ordinary differential equations with 21 variables. In the same way
as other HH-type models, the temporal variation of the membrane potential V .mV/
is described by
dV 1
D  .Itotal  Iext /; (5.2)
dt C
where C . F cm2 / is the membrane capacitance, Itotal . A cm2 / the sum of all
ionic currents across the membrane, and Iext . A cm2 / the external stimulus cur-
rent. The term Itotal includes many ionic currents via not only ion channels but also
ion exchangers and pumps, and thus the LRd model becomes 21-dimensional. Since
the equations of LRd model is very complicated and lengthy, they are omitted in this
book and all details of the equations can be found in Luo and Rudy (1994).
Figure 5.21 shows an action potential waveform of the LRd model, when a pe-
riodic pulse is applied. (Note that, differently from the cardiac pacemaker cell, the
ventricular cell does not produce any action potentials without external stimulus.)
Since it is rather difficult to analyze the bifurcation structure of the LRd model un-
der periodic pulsatile stimulation, we use the constant direct current as an external
stimulus Iext in the following.
136 5 Hodgkin–Huxley-Type Models of Cardiac Muscle Cells

Table 5.1 Ingredients of the LRd model

Notation Explanation
INa Fast NaC current
INa;b Background NaC current
ICa.L/ L-type Ca2C current
INaCa NaC Ca2C exchange current
Ip.Ca/ Ca2C pump current
ICa.T/ T-type Ca2C current
ICa;b Background Ca2C current
IKr Rapid delayed rectifier KC current
IKs Slow delayed rectifier KC current
IK1 Time-independent KC current
IKp Plateau KC current
Ito Transient outward current
INaK NaC KC pump current
IK.Na/ NaC activated KC current
IK.ATP/ ATP activated KC current
Ins.Ca/ Non-specific Ca2C activated current
NSR Network sarcoplasmic reticulum
JSR Junctional sarcoplasmic reticulum
Iup Ca2C uptake from myoplasm to NSR
Itr Ca2C transfer from NSR to JSR
Ileak Ca2C leak from NSR to myoplasm
Irel Ca2C release from JSR to myoplasm
Troponin
Calmodulin Calcium buffers
Calsequestrin

Fig. 5.21 The action

Iext(μA/cm2)

potential waveform simulated 80

by the LRd model and the
periodic stimulus current Iext 0
(upper panel). The stimulus 80
current is applied to the cell 60
every 1,000 ms 40
20
V (mV)

0
–20
–40
–60
–80
– 100
0 500 1000 1500 2000 2500 3000
t (msec)

5.3.2 Bifurcation Structure of the LRd Model

Figure 5.22 is a one-parameter bifurcation diagram when G Na is 600 and the other
parameters are set in their normal values. The bifurcation parameter in the abscissa
5.3 Ventricular Cell Model 137

Fig. 5.22 One-parameter 75

bifurcation diagram as for the HC1 DC
50
bifurcation parameter Iext
(G Na D 600) 25 DC DC
DC

V (mV)
0 DC
SN1
–25
HB2 HB1
–50 HC3
HC2 SN2
–75 HB3
–100
–1 0 1 2 3 4 5
Iext (μA/cm2)

Iext = –1.0, GNa = 16.0 Iext = 0.8, GNa = 16.0

80 80
40 40
V (mV)
V (mV)

0 0
– 40 –40 Iext = 3.5, GNa = 16.0
–80 –80
80
0 1000 0 1000 40

V (mV)
t (msec) t (msec)
0
–40
800 HC3 –80
SN
600 HC2 HB3 HB1 HB 0 1000
HC t (msec)
400
GNa (mS/μF)

HB2
200 HC1
SN1 Iext = 2.5, GNa = 16.0
0
area 2 80
– 200
SN2 40
V (mV)

– 400 area 1 area 3 area 4

0
– 600 –40
–80
– 800 0 1000
–1 0 1 2 3 4 t (msec)
Iext (μA/cm2)

Fig. 5.23 Two-parameter bifurcation diagram on the two parameters Iext and G Na . This figure also
shows examples of a typical waveform in each area which is separated by the bifurcation curves

is the constant stimulus current Iext . There are two saddle-node bifurcation points
(SN1, SN2), three Hopf-bifurcation points (HB1–HB3), several double-cycle bifur-
cation points (DC), and three homoclinic bifurcation points (HC1–HC3).
Figure 5.23 is the two-parameter bifurcation diagram where the abscissa denotes
the constant stimulus current Iext and the ordinate is the maximum conductance
G Na of the INa current. In areas 1 and 2 where the stimulus current Iext is small, the
LRd model does not generate any action potentials, since ventricular cells cannot
generate any action potentials without sufficiently large stimulus current. Area 1 in
Fig. 5.23 corresponds to the left side of SN1 in Fig. 5.22 and area 2 to the region
between SN1 and SN2 in Fig. 5.22. In both regions, the LRd model possesses a
unique stable equilibrium point.
138 5 Hodgkin–Huxley-Type Models of Cardiac Muscle Cells

If the stimulus becomes large enough, the LRd model generates a train of action
potentials. Area 3 corresponds to the region between SN2 and HB1 in Fig. 5.22, and
one stable periodic solution (repetitive excitation) exists. If the stimulus becomes
too large, the LRd model cannot produce any action potentials again. Area 4 in
Fig. 5.23 corresponds to the right side of HB1 in Fig. 5.22 where the LRd model
possesses a unique stable equilibrium point (the potential is depolarized, differently
from the region leftward SN2).
The two-parameter bifurcation diagram shows that a moderate change of G Na
near the normal value (G Na  16.0) does not significantly change the relative posi-
tion between two saddle-node bifurcation points (SN1, SN2) and a Hopf bifurcation
point (HB1). If we increase G Na greatly from its normal value, four new bifurcation
points (HB2, HB3, HC2, HC3) are born. These results show that the LRd model is
insensitive to the variation of G Na .
Next, let us consider the sensitivity to the so-called slow delayed rectifier
current IKs :

IKs D cKs G Ks xs1 xs2 .V  EKs /;

where xs1 is the activation variable, xs2 is the inactivation variable, G Ks is the
maximum conductance of this channel, and EKs is the reversal potential of IKs .
The coefficient cKs is the parameter for the conductance change of this cur-
rent. Figure 5.24 is the two-parameter bifurcation diagram on the coefficient cKs
and Iext . Differently from Fig. 5.23, we can see that several bifurcation curves
are intertwined; the small change of cKs significantly affects the dynamics of
the LRd model. Therefore, the LRd model is very sensitive to the IKs current.
We investigated the sensitivities to other ionic currents systematically. Consult
Yamaguchi et al. (2007) for detail.

5.4 Other HH-Type Models of Cardiac Cells

In this chapter, we have presented only the two examples of cardiac cell models
(YNI and LRd models) and analyzed their dynamics and bifurcation structures.
There are, however, diverse HH-type models on the electrophysiology of vari-
ous cardiac cells. HH-type models and the computational physiology on cardiac
cells has started with the Noble model (Noble 1962), which is a slight modification
of the original HH equations of a squid and is a model of excitable membrane of
Purkinje fibres. In the late 1970s, McAllister et al. formulated a model of action po-
tentials in Purkinje fibres (McAllister et al. 1975), and Beeler and Reuter (1977)
formulated a model of mammalian ventricular myocardial cells (Beeler–Reuter
model). In 1980s, the development of single channel recording technology allowed
quantitative measurements of various ionic channels. DiFrancesco and Noble (1985)
developed the model of Purkinje fibres. Luo and Rudy (1991) released a mam-
malian ventricular myocardial cell model (LRI model: Luo–Rudy I or Luo–Rudy
5.4 Other HH-Type Models of Cardiac Cells 139

Iext = 0.0, cKs = 0.5 Iext = 1.0, cKs = 0.5 Iext = 2.7, cKs = 0.5
80 80 80
40 40 40
V (mV)

V (mV)

V (mV)
0 0 0
– 40 –40 – 40
–80 –80 –80
0 1000 0 1000 0 1000
t (msec) t (msec) t (msec)

1
SN Iext = 3.0, cKs = 0.5
HB area 2
0.5 HC 80
SN1
area 1 area 4 40

V (mV)
0
c Ks

area 5
0 – 40
area 3 – 80
0 1000
– 0.5 area 6 t (msec)
HB1 HC1 SN2

–1
–5 –4 –3 –2 –1 0 1 2 3
Iext(μA/cm2)

Iext = 1.8, cKs = 0.5 Iext = 1.8, cKs = 0.5

Iext = –1.0, cKs = 0.5 Vo = 0.0 Vo = –90.0
80 80 80
40 40
V (mV)

V (mV)

40
V (mV)

0 0 0
– 40 – 40 – 40
–80 –80 –80
0 1000 0 1000 2000 0 1000 2000
t (msec) t (msec) t (msec)

Fig. 5.24 Two-parameter bifurcation diagram on the two parameters Iext and G Ks . This figure also
shows examples of a typical waveform in each area which is separated by the bifurcation curves

phase I model) that modified the Beeler–Reuter model. The LRI model contains
six ionic currents and eight variables. They also proposed the Luo–Rudy dynamic
(LRd) model (Luo and Rudy 1994). The LRd model is based on recent experimen-
tal data and takes into account not only ionic channels but also ionic pumps and
exchangers, and contains 21 variables as stated in the above section. Ten Tusscher
et al. (2004) made the first version of human ventricular cell model, and they im-
proved their model with more realistic description of intracellular calcium dynamics
(ten Tusscher and Panfilov 2006a). Furthermore, they presented the reduced version
of their human ventricular cell model (ten Tusscher and Panfilov 2006b). The list of
these models which we have mentioned here is the very partial and incomplete one.
In fact, there are a tremendous number of HH-type models of cardiac cells. Also,
there is a huge repertoire of HH-type models of other excitable cells. For example,
consult the cellML web page: http://models.cellml.org. Figure 5.25 shows a
few examples of HH-type cardiac cell models other than the YNI and LRd models.
140 5 Hodgkin–Huxley-Type Models of Cardiac Muscle Cells

a Noble and Noble b Zhang et al.

If INa Ib,Na Isi Ib,Ca Ist INa ICa,L ICa,T INaCa Ib,NaIb,Ca If

IK IK1 Ito Ip INaCa Ip IK,r IK,s Ib,k Ito Isus

c Sarai et al. d Difrancesco and Noble

ICa,b ICaL ICaT INaCa INa IbNSC
If
INa Ib,Na Isi Ib,Ca

Cytoplasm
Sacroplasmic Reticulum
ISR,T

ISR,U ISR,L IRyR

IK,ATP IK1 IKr IK,ACHINaK IKpl Iha Ist

IK IK1 Ito Ip INaCa

Sinoatrial cell Perkinje fibre

e Hilgeman and Noble f Ramirez et al.

INa INa,b ICa ICa,b INa Ib,NaINaCaICa Ib,Ca Ip,CaICl,Ca

Sacroplasmic Reticulum
ATP
JSR NSR
Irel Itr I Iup
leak
ADP+Pi

Ip IK1 IK INaCa IKr IKs IK1 IKur,d Ito INaK

Atrial cell

Fig. 5.25 Schematic illustrations of some HH-type cardiac cell models: (a) Noble and Noble
(1984), (b) Zhang et al. (2000), (c) Sarai et al. (2003), (d) DiFrancesco and Noble (1985),
(e) Hilgemann and Noble (1987), (f) Ramirez and Nattel (2000).
5.4 Other HH-Type Models of Cardiac Cells 141

g h
Beeler and Reuter Priebe and Beuckelmann
ICa,b ICa INaCa INa,b INa
INa Is

Irel JSR NSR

CMDN Itr I Iup
CSQN leak

IX1 IK1
Ito IKr IK1 IKs INaK
i Sarai et al.
ICa,b ICaL ICaT INaCa INa IbNSC

Cytoplasm
Sacroplasmic Reticulum
ISR,T

ISR,U ISR,L IRyR

IK,ATP IK1 IKr IKs INaK Ito IKpl

Ventricular cell

Fig. 5.25 (continued) (g) Beeler and Reuter (1977), (h) Priebe and Beuckelmann (1998), (i) Sarai
et al. (2003). Some of these illustrations are modified with consulting the cellML web page: http://
models.cellml.org
References

Abbott LF, Kepler TB (1990) Model neurons: from Hodgkin–Huxley to Hopfield. In: Garrido L
(ed) Statistical mechanics of neural networks. Springer, Berlin
Adams P (1982) Voltage-dependent conductances of vertebrate neurones. Trends Neurosci
5:116–119
Aihara K, Matsumoto G, Ikegaya Y (1984) Periodic and non-periodic responses of a periodically
forced Hodgkin–Huxley oscillator. J Theor Biol 109:249–269
Aihara K, Takabe T, Toyoda M (1990) Chaotic neural networks. Phys Lett A 144:333–340
Alexander JC, Cai DY (1991) On the dynamics of bursting systems. J Math Biol 29:405–423
Alexander JC, Doedel EJ, Othmer JC (1990) On the resonance structure in a forced excitable
system. SIAM J Appl Math 50:1373–1418
Arnold L (1995) Random dynamical systems. In: Johnson R (ed) Dynamical systems. Lecture
Notes in Mathematics, vol 1609. Springer, Berlin, pp 1–43
Arnold L (1998) Random dynamical systems. Springer, Berlin
Av-Ron E (1994) The role of a transient potassium current in a bursting neuron model. J Math Biol
33:71–87
Bedrov YA, Akoev GN, Dick OE (1992) Partition of the Hodgkin–Huxley type model parameter
space into the regions of qualitatively different solutions. Biol Cybern 66:413–418
Beeler GW, Reuter H (1977) Reconstruction of the action potential of ventricular myocardial fibres.
J Physiol (London) 268:177–210
Bertram R (1994) Reduced-system analysis of the effects of serotonin on a molluscan burster
neuron. Biol Cybern 70:359–368
Bertram R, Butte MJ, Kiemel T, Sherman A (1995) Topological and phenomenological classifica-
tion of bursting oscillations. Bull Math Biol 57:413–439
Braaksma B (1993) Critical phenomena in dynamical systems of van der Pol type. Thesis,
Rijksuniversiteit Utrecht, Utrecht
Braaksma B (1998) Singular Hopf bifurcation in systems with fast and slow variables. J Nonlinear
Sci 8:457–490
Bulsara AR, Elston TC, Doering CR, Lowen SB, Lindenberg K (1996) Cooperative behavior in pe-
riodically driven noisy integrate-fire models of neuronal dynamics. Phys Rev E 53:3958–3969
Buonocore A, Nobile AG, Ricciardi LM (1987) A new integral equation for the evaluation of first-
passage-time probability densities. Adv Appl Prob 19:784–800
Butera RJ Jr (1998) Multirhythmic bursting. Chaos 8:274–284
Caianiello ER (1961) Outline of a theory of thought-processes and thinking machines. J Theor Biol
2:204–235
Canavier CC, Clark JW, Byrne JH (1991) Simulation of the bursting activity of neuron R15
in Aplisia: role of ionic currents, calcium balance, and modulatory transmitters. J Neurophysiol
66:2107–2124
Canavier CC, Baxter DA, Clark JW, Byrne JH (1993) Nonlinear dynamics in a model neuron
provide a novel mechanism for transient synaptic inputs to produce long-term alterations of
postsynaptic activity. J Neurophysiol 69:2252–2257

143
144 References

Carpenter GA (1977) A geometric approach to singular perturbation problems with applications to

nerve impulse equations. J Diff Eqns 23:335–367
Chay TR, Keizer J (1983) Minimal model for membrane oscillations in the pancreatic ˇ-cell.
Biophys J 42:181–190
Chay TR, Rinzel J (1985) Bursting, beating, and chaos in an excitable membrane model. Biophys
J 47:357–366
Chhikara RS, Folks JL (1988) The inverse Gaussian distribution: theory, methodology, and appli-
cations. M. Dekker, New York
Clay JR (1976) A stochastic analysis of the graded excitatory response of nerve membrane. J Theor
Biol 59:141–158
Clay JR (1998) Excitability of the squid giant axon revisited. J Neurophysiol 80:903–913
Connor JA, Walter D, McKown R (1977) Neural repetitive firing: modifications of the Hodgkin–
Huxley axon suggested by experimental results from crustacean axons. Biophys J 18:81–102
Coombes S, Bressloff PC (eds) (2005) Bursting: the genesis of rhythm in the nervous system.
World Scientific, Singapore
Crill WE, Schwindt PC (1983) Active currents in mammalian central neurons. Trends Neurosci
6:236–240
Cronin J (1987) Mathematical aspects of Hodgkin–Huxley neural theory. Cambridge University
Press, Cambridge
DiFrancesco D, Noble D (1985) A model of cardiac electrical activity incorporating ionic pumps
and concentration changes. Philos Trans R Soc Lond [Biol] 307:353–398
Doedel E, Wang X, Fairgrieve T (1995) AUTO94 – software for continuation and bifurcation
problems in ordinary differential equations. CRPC-95-2, California Institute of Technology
Doi S (1993) On periodic orbits of trapezoid maps. Adv Appl Math 14:184–199
Doi S, Kumagai S (2005) Generation of very slow neuronal rhythms and chaos near the Hopf
bifurcation in single neuron models. J Comp Neurosci 19:325–356
Doi S, Sato S (1995) The global bifurcation structure of the BVP neuronal model driven by periodic
pulse trains. Math Biosci 125:229–250
Doi S, Inoue J, Kumagai S (1998) Spectral analysis of stochastic phase lockings and stochastic
bifurcations in the sinusoidally-forced van der Pol Oscillator with additive noise. J Stat Phys
90:1107–1127
Doi S, Inoue J, Sato S, Smith CE (1999) Bifurcation analysis of neuronal excitability and oscil-
lations. In: Poznanski R (ed) Modeling in the neurosciences: from ionic channels to neural
networks, chap 16. Harwood, Newark, NJ, pp 443–473
Fenichel N (1979) Geometric singular perturbation theory for ordinary differential equations. J Diff
Eqns 31:53–98
FitzHugh R (1960) Thresholds and plateaus in the Hodgkin–Huxley nerve equations. J Gen Physiol
43:867–896
FitzHugh R (1961) Impulses and physiological states in theoretical models of nerve membrane.
Biophy J 1:445–466
Fukai H, Doi S, Nomura T, Sato S (2000a) Hopf bifurcations in multiple parameter space of the
Hodgkin–Huxley equations. I. Global organization of bistable periodic solutions. Biol Cybern
82:215–222
Fukai H, Nomura T, Doi S, Sato S (2000b) Hopf bifurcations in multiple parameter space of the
Hodgkin–Huxley equations. II. Singularity theoretic approach and highly degenerate bifurca-
tions. Biol Cybern 82:223–229
Gardiner CW (1983) Handbook of stochastic methods for physics, chemistry and the natural
sciences. Springer, Berlin
Gerber B, Jakobsson E (1993) Functional significance of the A-current. Biol Cybern 70:109–114
Gerstein GL, Mandelbrot B (1964) Random walk models for the spike activity of a single neuron.
Biophys J 4:41–68
Glass L, Mackey MC (1979) A simple model for phase locking of biological oscillators. J Math
Biol 7:339–352
References 145

Glass L, Mackey MC (1988) From clocks to chaos, the rhythms of life. Princeton University Press,
Princeton
Glass L, Sun J (1994) Periodic forcing of a limit-cycle oscillator: fixed points, Arnold tongues, and
the global organization of bifurcations. Phys Rev E 50:5077–5084
Golomb D, Guckenheimer J, Gueron S (1993) Reduction of a channel-based model for a stomato-
gastric ganglion LP neuron. Biol Cybern 69:129–137
Grasman J, Jansen MJW (1979) Mutually synchronized relaxation oscillators as prototypes of
oscillating systems in biology. J Math Biol 7:171–197
Guckenheimer J (1975) Isochrons and phaseless sets. J Math Biol 1:259–273
Guckenheimer J (1986) Multiple bifurcation problems for chemical reactors. Physica D 20:1–20
Guckenheimer J (1996) Towards a global theory of singularly perturbed dynamical systems. Prog
Nonlinear Diff Eqns Appl 19:213–225
Guckenheimer J, Holmes P (1983) Nonlinear oscillations, dynamical systems, and bifurcation of
vector fields. Springer, Berlin
Guckenheimer J, Labouriau IS (1993) Bifurcation of the Hodgkin and Huxley equations: a new
twist. Bull Math Biol 55:937–952
Guckenheimer J, Gueron S, Harris-Warrick RM (1993) Mapping the dynamics of a bursting neu-
ron. Phil Trans R Soc Lond B 341:345–359
Guevara MR, Glass L (1982) Phase locking, period-doubling bifurcations and chaos in a math-
ematical model of a periodically driven oscillator: a theory for the entrainment of biological
oscillators and the generation of cardiac dysrhythmias. J Math Biol 14:1–23
Guttman R, Lewis S, Rinzel J (1980) Control of repetitive firing in squid axon membrane as model
for neuroneoscillator. J Physiol 305:377–395
Hadeler KP, an der Heiden U, Schumacher K (1976) Generation of the nervous impulse and
periodic oscillations. Biol Cybern 23:211–218
Hassard B (1978) Bifurcation of periodic solutions of the Hodgkin–Huxley model for the squid
giant axon. J Theor Biol 71:401–420
Hassard BD, Shiau LJ (1989) Isolated periodic solutions of the Hodgkin–Huxley equations.
J Theor Biol 136:267–280
Hata M (1982) Dynamics of Caianiello’s equation. J Math Kyoto Univ 22(1):155–173
Hayashi H, Ishizuka S (1992) Chaotic nature of bursting discharges in the Onchidium pacemaker
neuron. J Theor Biol 156:269–291
Hilgemann DW, Noble D (1987) Excitation-contraction coupling and extracellular calcium tran-
sients in rabbit atrium: reconstruction of the basic cellular mechanisms. Proc R Soc Lond B
Biol Sci 230:163–205
Hille B (1992) Ionic channels of excitable membranes, 2nd edn. Sinauer, Sunderland, MA
Hodgkin AL, Huxley AF (1952) A quantitative description of membrane current and its applica-
tions to conduction and excitation in nerve. J Physiol 117:500–544
Hoppensteadt FC, Keener JP (1982) Phase locking of biological clocks. J Math Biol 15:339–349
Horikawa Y (1994) Period-doubling bifurcations and chaos in the decremental propagation of a
spike train in excitable media. Phys Rev E 50:1708–1710
Inoue J, Doi S (2007) Sensitive dependence of the coefficient of variation of interspike intervals
on the lower boundary of membrane potential for the leaky integrate-and-fire neuron model.
Biosystems 87:49–57
Izhikevich EM (2006) Dynamical systems in neuroscience: the geometry of excitability and burst-
ing. MIT Press, Cambridge, MA
Jones C (1996) Geometric singular perturbation theory. In: Johnson R (ed) Dynamical systems.
Lecture Notes in Mathematics, vol. 1609. Springer, Berlin
Jones C, Kopell N (1994) Tracking invariant manifolds with differential forms. J Diff Eqns
108:64–88
Kawato M (1981) Transient and steady phase response curves of limit cycle oscillators. J Math
Biol 12:13–30
Kawato M, Suzuki R (1978) Biological oscillators can be stopped. Topological study of a phase
response curve. Biol Cybern 30:241–248
146 References

Keener JP, Glass L (1984) Global bifurcations of a periodically forced nonlinear oscillator. J Math
Biol 21:175–190
Keener JP, Sneyd J (1998) Mathematical physiology. Springer, Berlin
Keener JP, Hoppensteadt FC, Rinzel J (1981) Integrate-and-fire models of nerve membrane re-
sponse to oscillatory input. SIAM J Appl Math 41:503–517
Kepler TB, Marder E (1993) Spike initiation and propagation on axons with slow inward currents.
Biol Cybern 68:209–214
Kepler TB, Marder E, Abbott LF (1990) The effect of electrical coupling on the frequency of model
neuronal oscillators. Science 248:83–85
Kepler TB, Abbott LF, Marder E (1992) Reduction of conductance-based neuron models. Biol
Cybern 66:381–387
Kokoz YuM, Krinskii VI (1973) Analysis of the equations of excitable membranes. II. Method of
analysing the electrophysiological characteristics of the Hodgkin–Huxley membrane from the
graphs of the zero-isoclines of a second order system. Biofizika 18:878–885
Konig P, Engel AK, Singer W (1996) Integrator or coincidence detector? The role of the cortical
neuron revisited. Trends Neurosci 19:130–137
Koper MTM (1995) Bifurcation of mixed-mode oscillations in a three-variable autonomous Van
der Pol–Duffing model with a cross-shaped phase diagram. Physica D 80:72–94
Krinskii VI, Kokoz YuM (1973) Analysis of the equations of excitable membranes. I. Reduction
of the Hodgkin–Huxley equations to a second order system. Biofizika 18:506–511
Labouriau IS (1985) Degenerate Hopf bifurcation and nerve impulse. SIAM J Math Anal
16:1121–1133
Labouriau IS (1989) Degenerate Hopf bifurcation and nerve impulse. Part II. SIAM J Math Anal
20:1–12
Labouriau IS, Ruas MAS (1996) Singularities of equations of Hodgkin–Huxley type. Dyn Stab
Syst 11:91–108
Lasota A, Mackey MC (1994) Chaos, fractals, and noise: stochastic aspects of dynamics. Springer,
Berlin
Leonov NN (1959) Map of the line on to itself. Radiofisica 2:942–956
Llinas RR (1988) The intrinsic electrophysiological properties of mammalian neurons: insights
into central nervous system function. Science 242:1654–1664
Luo CH, Rudy Y (1991) A model of the ventricular cardiac action potential. Depolarization, repo-
larization, and their interaction. Circ Res 68:1501–1526
Luo CH, Rudy Y (1994) A dynamic model of the ventricular cardiac ventricular action potential.
I. Simulations of ionic currents and concentration changes. Circ Res 74:1071–1096
Maeda Y, Pakdaman K, Nomura T, Doi S, Sato S (1998) Reduction of a model for an Onchidium
pacemaker neuron. Biol Cybern 78:265–276
Matsumoto G, Aihara K, Ichikawa M, Tasaki A (1984) Periodic and nonperiodic responses
of membrane potentials in squid giant axons during sinusoidal current stimulation. J Theor
Neurobiol 3:1–14
McAllister RE, Noble D, Tsien RW (1975) Reconstruction of the electrical activity of cardiac
Purkinje fibres. J Physiol (London) 251:1–59
Mcdonald SW, Grebogi C, Ott E, et al (1985) Fractal basin boundaries. Physica D 17(2):125–153
Meunier C (1992) Two and three-dimensional reductions of the Hodgkin–Huxley system: separa-
tion of time scales and bifurcation schemes. Biol Cybern 67:461–468
Mira C (1987) Chaotic dynamics. World Scientific, Singapore
Mirollo RE, Strogatz SH (1990) Synchronization of pulse-coupled biological oscillators. SIAM J
Appl Math 50:1645–1662
Nagumo J, Sato S (1972) On a response characteristic of a mathematical neuron model. Kybernetik
10:155–164
Nagumo J, Arimoto S, Yoshizawa S (1962) An active pulse transmission line stimulating nerve
axon. Proc Inst Radio Eng 50:2061–2070
Nakano H, Saito T (2002) Basic dynamics from a pulse-coupled network of autonomous integrate-
and-fire chaotic circuits. IEEE Trans Neural Netw 13:92–100
References 147

Noble D (1962) Modification of Hodgkin–Huxley equations applicable to purkinje fibre action and
pace-maker potentials. J Physiol (London) 160:317–352
Noble D (1975) The initiation of the heartbeat. Oxford University Press, Oxford
Noble D (1995) The development of mathematical models of the heart. Chaos Solitons Fractals
5:321–333
Noble D, Noble SJ (1984) A model of sino-atrial node electrical activity based on a modification
of the DiFrancesco–Noble (1984) equations. Proc R Soc Lond B Biol Sci 222:295–304
Nomura T, Sato S, Doi S, Segundo JP, Stiber MD (1994a) Global bifurcation structure of a Bon-
hoeffer van der Pol oscillator driven by periodic pulse trains. Comparison with data from an
inhibitory synapse. Biol Cybern 72:55–67
Nomura T, Sato S, Doi S, Segundo JP, Stiber MD (1994b) A modified radial isochron clock with
slow and fast dynamics as a model of pacemaker neurons. Global bifurcation structure when
driven by periodic pulse trains. Biol Cybern 72:93–101
Okuda M (1981) A new method of nonlinear analysis for threshold and shaping actions in transient
state. Prog Theor Phys 66:90–100
Pakdaman K (2001) Periodically forced leaky integrate-and-fire model. Phys Rev E 63:041907
Plant RE (1976) The geometry of the Hodgkin–Huxley model. Comp Prog Biomed 6:85–91
Poznanski RR (1998) Electrophysiology of a leaky cable model for coupled neurons. J Austral
Math Soc B 40:59–71
Priebe L, Beuckelmann DJ (1998) Simulation study of cellular electric properties in heart gailure.
Circ Res 82:1206–1223
Ramirez RJ, Nattel S (2000) Courtemanche M: Mathematical analysis of canine atrial action po-
tentials: rate, regional factors, and electrical remodeling. Am J Physiol Heart Circ Physiol
279:H1767–H1785
Rescigno R, Stein RB, Purple RL, Poppele RE (1970) A neuronal model for the discharge patterns
produced by cyclic inputs. Bull Math Biophys 32:337–353
Ricciardi LM (1977) Diffusion processes and related topics in biology. Springer, Berlin
Ricciardi LM, Sato S (1988) First-passage-time density and moments of the Ornstein–Uhlenbeck
process. J Appl Prob 25:43–57
Rinzel J (1978) On repetitive activity in nerve. Fed Proc 37:2793–2802
Rinzel J (1985) Excitation dynamics: insights from simplified membrane models. Fed Proc
44:2944–2946
Rinzel J (1990) Discussion: electrical excitability of cells, theory and experiment: review of the
Hodgkin–Huxley foundation and update. Bull Math Biol 52:5–23
Rinzel J, Keener JP (1983) Hopf bifurcation to repetitive activity in nerve. SIAM J Appl Math
43:907–922
Rinzel J, Miller RN (1980) Numerical calculation of stable and unstable periodic solutions to the
Hodgkin–Huxley equations. Math Biosci 49:27–59
Rocşoreanu C, Georgescu A, Giurgiţeanu N (2000) The Fitzhugh–Nagumo model: bifurcation and
dynamics. Springer, Berlin
Rush ME, Rinzel J (1994) Analysis of bursting in a thalamic neuron model. Biol Cybern
71:281–291
Rush ME, Rinzel J (1995) The potassium A-current, low firing rates and rebound excitation in
Hodgkin–Huxley models. Bull Math Biol 57:899–929
Sarai N, Matsuoka S, Kuratomi S, Ono K, Noma A (2003) Role of individual ionic current systems
in the SA node hypothesized by a model study. Jpn J Physiol 53:125–134
Scharstein H (1979) Input–output relationship of the leaky-integrator neuron model. J Math Biol
8:403–420
Schweighofer N, Doya K, Kawato M (1999) Electrophysiological properties of inferior olive neu-
rons: a compartmental model. J Neurophysiol 82:804–817
Shadlen MN, Newsome WT (1994) Noise, neural codes and cortical organization. Curr Opin Neu-
robiol 4:569–579
Shadlen MN, Newsome WT (1998) The variable discharge of cortical neurons: implications for
connectivity, computations, and information coding. J Neurosci 18:3870–3896
148 References

Shiau LJ, Hassard BD (1991) Degenerate Hopf bifurcation and isolated periodic solutions of the
Hodgkin–Huxley model with varying sodium ion concentration. J Theor Biol 148:157–173
Softky WR, Koch C (1993) The highly irregular firing of cortical cells is inconsistent with temporal
integration of random EPSPs. J Neurosci 13:334–350
Stein RB, French AS, Holden AV (1972) The frequency response, coherence, and information
capacity of two neuronal models. Biophys J 12:295–322
Strassberg AF, DeFelice LJ (1993) Limitations of the Hodgkin–Huxley formalism: effects of single
channel kinetics upon transmembrane voltage dynamics. Neural Comput 5:843–855
Takahashi N, Hanyu Y, Musha T, Kubo R, Matsumoto G (1990) Global bifurcation structure in
periodically stimulated giant axons of squid. Physica D 43:318–334
Tateno T, Doi S, Sato S, Ricciardi LM (1995) Stochastic phase-lockings in a relaxation oscilla-
tor forced by a periodic input with additive noise: a first-passage-time approach. J Stat Phys
78:917–935
ten Tusscher KHW, Panfilov AV (2006a) Alternans and spiral breakup in a human ventricular tissue
model. Am J Physiol Heart Circ Physiol 291:H1088–H1100
ten Tusscher KHW, Panfilov AV (2006b) Cell model for efficient simulation of wave propaga-
tion in human ventricular tissue under normal and pathological conditions. Phys Med Biol
51:6141–6156
ten Tusscher KHW, Noble D, Noble PJ, Panfilov AV (2004) A model for human ventricular tissue.
Am J Physiol Heart Circ Physiol 286:H1573–H1589
Terman D (1991) Chaotic spikes arising from a model of bursting in excitable membranes. SIAM
J Appl Math 51:1418–1450
Torikai H, Saito T (1999) Return map quantization from an integrate-and-fire model with two
periodic inputs. IEICE Trans Fundam E82-A:1336–1343
Traub RD, Wong RKS, Miles R, Michelson H (1991) A model of a CA3 hippocampal pyra-
midal neuron incorporating voltage-clamp data on intrinsic conductances. J Neurophysiol
66:635–650
Troy WC (1978) The bifurcation of periodic solutions in the Hodgkin–Huxley equations. Q Appl
Math 36:73–83
Tsumoto K, Yoshinaga T, Aihara K, Kawakami H (2003) Bifurcations in synaptically coupled
Hodgkin–Huxley neurons with a periodic input. Int J Bifurcat Chaos 13:653–666
Tsumoto K, Kitajima H, Yoshinaga T, Aihara K, Kawakami H (2006) Bifurcations in Morris–Lecar
neuron model. Neurocomputing 69:293–316
Tuckwell HC (1988) Introduction to theoretical neurobiology. Cambridge University Press, Cam-
bridge
Tuckwell HC (1989) Stochastic processes in the neurosciences. SIAM, Philadelphia
van der Pol B (1926) On “relaxation-oscillations”. Phil Mag 2:978–992
Winfree AT (1980) The geometry of biological time. Springer, New York
Xu J-X, Jiang J (1996) The global bifurcation characteristics of the forced van der Pol oscillator.
Chaos Solitons Fractals 7:3–19
Yamaguchi R, Doi S, Kumagai S (2007) Bifurcation analysis of a detailed cardiac cell model and
drug sensitivity of ionic channels. In: Proc. 15th IEEE international workshop on Nonlinear
Dynamics of Electronic Systems 2007, pp 205–208
Yanagida E (1985) Stability of fast traveling pulse solutions of the FitzHugh–Nagumo equations.
J Math Biol 22:81–104
Yanagida E (1987) Branching of double pulse solutions from single pulse solutions in nerve axon
equations. J Diff Eqns 66:243–262
Yanagida E (1989) Stability of double-pulse solutions in nerve axon equations. SIAM J Appl Math
49:1158–1173
Yanagihara K, Noma A, Irisawa H (1980) Reconstruction of sinoatrial node pacemaker potential
based on the voltage clamp experiments. Jpn J Physiol 30:841–857
Yellin E, Rabinovitch A (2003) Properties and features of asymmetric partial devil’s staircases
deduced from piecewise linear maps. Phys Rev E 67:016202
References 149

Yoshinaga T, Sano Y, Kawakami H (1999) A method to calculate bifurcations in synaptically cou-

pled Hodgkin–Huxley equations. Int J Bifurcat Chaos 9:1451–1458
Yu X, Lewis ER (1989) Studies with spike initiators: linearization by noise allows continuous
signal modulation in neural networks. IEEE Trans Biomed Eng 36:36–43
Zhang H, Holden AV, Kodama I, Honjo H, Lei M, Varghese T, Boyett MR (2000) Mathematical
models of action potential in the periphery and center of the rabbit sinoatrial node. Am J Physiol
279:H397–H421
Index

A Devil’s staircase, 77
Action potential, 37 Difference approximation, 6
Activation variable, 43 1D map, 13
All-or-none, 38 Double-cycle bifurcation, 54, 62
Asymptotic stability, 14
Attractor, 25
E
Axon, 37 Eigenvalue, 4
Eigenvector, 4
Equilibrium point, 2, 6, 18
B Equilibrium potential, 41
Basin, 25 Exchanger, 134
Basin of attraction, 25 Excitation, 37
Bifurcation, 13 Exponential of a matrix, 6
curve, 112
diagram, 17
parameter, 17 F
Bogdanov–Takens bifurcation, 68 Firing, 37
Bonhoeffer–van der Pol (BVP) model, 55 First-passage time (FPT), 89
Brownian motion, 89 FitzHugh–Nagumo (FHN) model, 55
BVP equations, 56 Fixed point, 2, 6, 14
Flip bifurcation, 16
Focus, 3, 9
C Fold bifurcation, 16, 17, 19
Caianiello’s neuron model, 76 Fractal, 25
Cantor function, 77
Cell membrane, 39
Center, 3, 9 G
Chaos neuron model, 78 Gate, 39
Chaotic solution, 17 Gate variable, 41
Circle map, 14 Gaussian white noise, 94
Coefficient of variation (CV), 88
Coordinate transformation, 12 H
Cusp, 68 HH equation, 42
HH formalism, 37, 40
HH-type equations, 37
D Hodgkin–Huxley equations, 40
Degree, 14, 75 Homoclinic bifurcation, 131
Dendrite, 37 Hopf bifurcation, 21
Depolarized state, 46 Hopf bifurcation curve, 65

151
152 Index

I P
Inactivation variable, 43 Pacemaker activity, 122
Initial state, 1, 6 Period, 5, 10, 14
Integrate-and-fire (IF) neuronal model, 69 Period-doubling bifurcation, 16, 17
Interspike interval (ISI), 88 Periodic orbit, 5, 10, 14
Inverse Gaussian Distribution (IGD), 89 Periodic point, 14
Ion channel, 39 Phase, 25
Isochrone, 74 Phase locking, 70, 84
Phase space, 2
Phase transition curve (PTC), 71
J Poincaré oscillator, 74
Jacobian matrix, 12 Pump, 134

K Q
KC channel, 40 QTP (quasi-type) separatrix, 80
KC current, 41 Quasi-periodic, 15
Quasi-periodic orbit, 5
Quasi-threshold, 56
L Quiescent state, 44
Leak current, 41
Leaky integrate-and-fire (LIF) neuron, 72
Lift, 14 R
Limit cycle, 21 Radial isochron clock (RIC), 74
Limit Point, 126 Refractory period, 38
Linearization, 12 Repetitive firing, 52
Linearized system, 12, 20 Resting potential, 41
Luo–Rudy dynamic (LRd) model, 134 Rotation number, 14

M S
Markov operator, 95 Saddle, 3, 9, 20
Maximum conductance, 41 Saddle-node bifurcation, 16, 19, 20
Mean firing rate (MFR), 77 Saddle-node bifurcation of periodics, 54
Membrane, 38 Sine-circle map, 15
Membrane potential, 39, 41 Singular limit, 82
Miniature end-plate potential, 88 Soma, 37
Space-clamp, 42
Spike, 37
N Stability, 11
NaC channel, 39 Stable, 2, 8, 11
NaC current, 41 Stable manifold, 26
Nernst potential, 41 Standard circle map, 15
Neuron, 37 State, 1
Noble model, 138 State point, 1
Node, 3, 9, 20 State space, 2, 6
Nullcline, 44, 58 Steady state, 6
Stochastic bifurcation, 98
Stochastic neuron model, 88
O Sub-critical Hopf bifurcation, 22
One-dimensional map, 13 Super-critical Hopf bifurcation, 21
Orbit, 2, 6, 8 Super stable, 14
Ornstein–Uhlenbeck (OU) process, 90 Synapse, 37
Index 153

T V
Threshold, 38 Vector field, 6, 8
Trajectory, 2, 6 v–m subsystem, 44
Two-parameter bifurcation diagram, 65
Type-I neuronal model, 45
W
Wiener process, 89
U
Unit circle, 14
Unstable fixed point, 2, 10 Y
Unstable manifold, 26 Yanagihara–Noma–Irisawa (YNI) model, 121