Path-integral evolution of chaos embedded in noise: Dufng neocortical analog

Lester Ingber
Lester Ingber Research, P.O. Box 857, McLean, Virginia 22101 U.S.A.
[email protected]
Ramesh Srinivasan
Department of Psychology and Institute of Cognitive and Decision Sciences,
and Electrical Geodesics, Inc., Eugene, OR 97403, U.S.A.
[email protected]
Paul L. Nunez
Department of Biomedical Engineering, Tulane University, New Orleans, Louisiana 70118 U.S.A.
[email protected]
AbstractA two dimensional time-dependent Dufng oscillator model of macroscopic neocortex
exhibits chaos for some ranges of parameters. We embed this model in moderate noise, typical of the
context presented in real neocortex, using PATHINT, a non-Monte-Carlo path-integral algorithm that is
particularly adept in handling nonlinear Fokker-Planck systems. This approach shows promise to
investigate whether chaos in neocortex, as predicted by such models, can survive in noisy contexts.
Keywords: chaos, path integral, Fokker Planck, Dufng equation, neocortex
Path-integral of Dufng model - 2 - Ingber, Srinivasan and Nunez
1. INTRODUCTION
A global theory of neocortical dynamic function at macroscopic scales was developed to explain
salient features of electroencephalographic(EEG) data recorded from human scalp [1]. In some limiting
cases, the theory predicts EEG standing wav es composed of linear combinations of spatial modes. These
modes (which may be called order parameters in the parlance of modern dynamical methods) are
governed by linear or quasilinear ordinary differential equations [1-3]. These data and theoretical works
suggest several connections to reports of chaotic attractors estimated for EEG data [1]. In order to
illuminate possible relations between sub-macroscopic chaos (for example in neocortical columns) and
large scale data observed at the scalp, we suggested a simple metaphoric system, the linear stretched
string with nonlinear attached springs [4-6]. We hav e studied a forced partial differential equation
describing the string-spring system, where the forcing term might represent a steady-state evoked
potential in the neocortical analogy. In the limiting case when string tension is zero, the forced Dufng
equation is obtained. The Dufng equation exhibits a rich dynamic behavior of periodic and chaotic
motion, for different parameters (associated with different wav e numbers in the string-spring system) [7].
Another series of papers has outlined a statistical mechanics of neocortical interactions (SMNI),
demonstrating that the statistics of synaptic and neuronal interactions develops a stochastic background of
moderate noise that survives up to the macroscopic scales that are typically measured by scalp
EEG [4,8-12].
This study is concerned with the relation between chaos that may occur in particular spatial modes
(the order parameters) and EEG signals consisting of combinations of modes and noise. In this context,
we may consider the noise to be either instrumental or biologic, e.g., as detailed in the SMNI papers.
That is, can we reasonably expect correlation dimension estimates of EEG data to provide information
that can be used to construct physiologically-based theories of neocortical dynamics.
Section 2 describes the Dufng metaphor of neocortex that exhibits chaos. Section 3 describes the
embedding of the deterministic model into a stochastic medium. Section 4 describes the PATHINT code
used for the stochastic model. Section 5 gives the results of the PATHINT calculations. Section 6 is a
brief conclusion.
Path-integral of Dufng model - 3 - Ingber, Srinivasan and Nunez
2. DUFFING ANALOG OF NEOCORTEX
At this stage of our understanding, choosing a specic t to a macroscopic EEG model is
necessarily limited to the specic data and would necessarily be controversial. However, a relatively
simple physical system in which the general issues of spatial scale interaction and connection to
experimental EEG can be discussed with minimal complication is a linear stretched string with attached
nonlinear springs [4].
The ends of the string may be considered xed or the string may form a closed loop to make the
system more analogous to the closed neocortical/corticocortical ber surface. A closed loop of
transmission line with nonlinear dielectric conductance is the equivalent electric system. In the proposed
metaphorical link with the brain, the springs represent local circuit dynamics at columnar scales whereas
the string is believed to be somewhat more analogous to neocortical interactions along the corticocortical
bers [5,10].
String displacement, considered to be somewhat analogous to the modulation of cortical surface
potential is governed by

t
2
c
2

2

x
2
+

t
+ [
2
0
+ f ()] F(x, t) , (1)
In the case of xed-end supports, the forcing function takes the form
F(x, t) B(x) cos t . (2)
Here c
2
is the usual string tension parameter (or wav e velocity parameter),
2
0
is proportional to the linear
spring constant, and is the damping parameter. In the neocortical metaphor, F(x, t) might represent
ongoing random subcortical input (e.g., from the thalamus), or sensory information designed to study
linear/nonlinear resonance. Since the string/springs is simply a metaphor for the brain, the nonlinearity is
arbitrary, so we choose a cubic nonlinearity and sinusoidal driving so that string displacement is governed
by the partial differential equation

t
2
c
2

2

x
2
+

t
+
2
0
+
3
B(x) cos t . (3)
The driving has been included in order to connect the system to the well studied forced Dufngs equation
as well as to obtain chaotic dynamics. In the limiting case of a completely relaxed string (i.e., the velocity
of propagation c
2
is zero), the springs behave independently and drive the string, governed by the forced
Path-integral of Dufng model - 4 - Ingber, Srinivasan and Nunez
Dufngs equation at x x
0
,

t
2
+

t
+
2
0
+
3
B(x) cos t . (4)
The alternative way to arrive at Dufngs equation is to do a spatial mode expansion of the string equation
and then only keep the lowest wav enumber [6]. For instance, the Lorenz system is derived in this way
from the atmospheric turbulence equations [13].
This equation has been studied numerically for the case
0
0 in the parameter space (, B) using
spectral analysis and phase-plane projections [7]. A rich structure of periodic, quasi-periodic and chaotic
motion were reported. We duplicated Uedas results using a standard fth and sixth order Runge-Kutta
algorithm (IMSL routine DIVPRK). The dynamics were classied as chaotic on the basis of
characteristic broadening of spectral lines and aperiodic trajectories in phase space as well as correlation
dimension analysis. In all simulations discussed here, the damping parameter 0. 05 and B 5. 5.
Figs. 1a and 1b illustrate the evolution of the phase space for the non-chaos and chaos cases of

0
1. 0 and
0
0. 1 cases, resp., after transients have been passed (t > 250). Figs. 2a and 2b illustrate
the evolution of the phase space for the non-chaos and chaos cases of
0
1. 0 and
0
0. 1 cases, resp.,
during a transient period between t 12 and t 15. 5. The non-chaos cases are similar in appearance
except for the settling of transients.
3. EMBEDDING DETERMINISTIC DUFFING SYSTEM IN NOISE
The deterministic Dufng system described above can be written in the form
x f (x, t) ,
f x
2
0
x + B cos t , (5)
and recast as
x y ,
y f (x, t) ,
f y
2
0
x + B cos t . (6)
Note that this is equivalent to a 3-dimensional autonomous set of equations, e.g., replacing cos t by cos z,
Path-integral of Dufng model - 5 - Ingber, Srinivasan and Nunez
dening z , and taking to be some constant.
We now add independent Gaussian-Markovian (white) noise to both x and y,
j
i
, where the
variables are represented by i {x, y} and the noise terms are represented by j {1, 2},
x y + g
1
x

1
,
y f (x, t) + g
2
y

2
,
<
j
(t) >

0 ,
<
j
(t),
j
(t) >

jj
(t t) . (7)
In this study, we take moderate noise and simply set g
j
i
1. 0
j
i
.
The above denes a set of Langevin rate-equations, which can be recast into equivalent Fokker-
Planck and path-integral equations [14]. A compact derivation has been given in sev eral papers [10,15].
The equivalent short-time conditional probability distribution P, in terms of its Lagrangian L,
corresponding to these Langevin rate-equations is
P[x, y; t + t| x, y, t]
1
(2 t)( g
11
g
22
)
2
exp(Lt) ,
L
( x y)
2
2( g
11
)
2
+
( y f )
2
2( g
22
)
2
. (8)
4. PATHINT CODE
PATHINT is a non-Monte-Carlo histogram C-code developed to evolve an n-dimensional system
(subject to machine constraints), based on a generalization of an algorithm demonstrated by Wehner and
Wolfer to be extremely robust for nonlinear Lagrangians with moderate noise [16-18]. This algorithm has
been used for studies in neuroscience [19,20], nancial markets [21,22], combat analysis [23,24], and in a
study of stochastic chaos [25].
PATHINT computes the path-integral of a system in terms of its Lagrangian L,
P[q
t
|q
t
0
]dq(t)

. . .

Dq exp

min
t
t
0

dtL
_

,
((q(t
0
) q
0
)) ((q(t) q
t
)) ,
Path-integral of Dufng model - 6 - Ingber, Srinivasan and Nunez
Dq
u

lim
u+1
1

g
1/2
i

(2 t)
1/2
dq
i

,
L( q
i
, q
i
, t)
1
2
( q
i
g
i
)g
ii
( q
i
g
i
) ,
g
ii
(g
ii
)
1
,
g det(g
ii
) . (9)
If the diffusions are not constant, then there are additional terms [14].
The histogram procedure recognizes that the distribution can be numerically approximated to a high
degree of accuracy as sums of rectangles at points q
i
of height P
i
and width q
i
. For convenience, just
consider a one-dimensional system. The above path-integral representation can be rewritten, for each of
its intermediate integrals, as
P(x; t + t)

dx[g
1/2
(2 t)
1/2
exp(L t)]P(x; t)

dxG(x, x; t)P(x; t) ,
P(x; t)
N
i1

(x x
i
)P
i
(t) ,
(x x
i
)

'

1 , (x
i

1
2
x
i1
) x (x
i
+
1
2
x
i
) ,
0 , otherwise .
(10)
This yields
P
i
(t + t) T
ij
(t)P
j
(t) ,
T
ij
(t)
2
x
i1
+ x
i
x
i
+x
i
/2
x
i
x
i1
/2
dx

x
j
+x
j
/2
x
j
x
j1
/2
dxG(x, x; t) . (11)
T
ij
is a banded matrix representing the Gaussian nature of the short-time probability centered about the
(possibly time-dependent) drift.
Here, this histogram procedure s used in two dimensions, i.e., using a matrix T
ijkl
[23]. Explicit
dependence of L on time t also can be included without complications. Care must be used in developing
Path-integral of Dufng model - 7 - Ingber, Srinivasan and Nunez
the mesh in q
i
, which is strongly dependent on the diagonal elements of the diffusion matrix, e.g.,
q
i
(tg
ii
)
1/2
. (12)
Presently, this constrains the dependence of the covariance of each variable to be a (nonlinear) function of
that variable, in order to present a straightforward rectangular underlying mesh.
Since integration is inherently a smoothing process [21], tting data with the short-time probability
distribution, effectively using an integral over this epoch, permits the use of coarser meshes than the
corresponding stochastic differential equation. For example, the coarser resolution is appropriate,
typically required, for numerical solution of the time-dependent path integral. By considering the
contributions to the rst and second moments conditions on the time, variable meshes can be derived [16].
The time slice essentially is determined by t L
1
, where L is the uniform Lagrangian, respecting
ranges giving the most important contributions to the probability distribution P. Thus, t is roughly
measured by the diffusion divided by the square of the drift.
Monte Carlo algorithms for path integrals are well known to have extreme difculty in evolving
nonlinear systems with multiple optima [26], but this algorithm does very well on such systems. The
PATHINT code was tested against the test problems given in previous one-dimensional systems [16,17].
Tw o-dimensional runs were tested by using cross products of one-dimensional examples whose analytic
solutions are known.
5. PATHINT CALCULATION OF DUFFING SYSTEM
While we appreciate that a true calculation of chaos of a stochastic system is quite difcult, here
our approach is to simply perform a rigorous numerical calculation of the evolution of the Fokker-Plank
equation in its equivalent path-integral representation. Then, we can compare the chaos and non-chaos
topologies of the deterministic system, which have been seen above to (barely) be different, to the
topologies of the stochastic system. The topology of chaotic systems is an important invariant that often
can be used to identify its presence [27].
A mesh of t 0. 1 within ranges of {x, y} of t25 was reasonable to calculate the evolution of this
system for all runs. To be sure of accuracy in the calculations, off-diagonal spreads of meshes were taken
as t5 increments. This lead to an initial four-dimensional matrix of 159 159 11 11 = 3 059 001
points, which was cut down to a kernel of 2 954 961 points because the off-diagonal points did not cross
Path-integral of Dufng model - 8 - Ingber, Srinivasan and Nunez
the boundaries. Reecting Neumann boundary conditions were imposed by the method of images,
consisting of a point image plus a continuous set of images leading to an error function [28], but
calculations support the premise that they are unimportant here as the evolving distributions stayed well
within the boundaries. Since we have added noise to a set of ordinary differential equations, bringing
along additional boundary conditions, we thought it prudent to establish solutions that would be rather
independent of any Dirichlet or Neumann boundary conditions.
A Convex 120 supercomputer was used, but there were problems with its C compiler, so gcc
version 2.60 was built and used. Runs across several machines, e.g., Suns, Dec workstations, and Crays,
checked reproducibility of this compiler on this problem. It required about 12 CPU min to build the
kernel and perform the calculation of the next distribution at each t-folding. Data was accumulated after
ev ery 5 foldings. The chaos case was run up until t 33, requiring 67.5 CPU hours, and the non-chaos
case was run up until t 15. 5, requiring 32.0 CPU hours.
The above calculations took over 100 CPU-hours, and this was a self-imposed limit for this rst
examination of this problem. As demonstrated by the deterministic calculations above, further
calculations would have to be performed to enter into the time domain of t > 50 after transients have
diminished and where chaos truly is strongly exhibited. Furthermore, for t > 20, we noticed that there is
sufcient diffusion of the system towards the boundaries at {x, y} t 25, so that these ranges would have
to be increased, thereby consuming even more CPU time per t run. For the purposes of this paper, we
just look at epochs where the deterministic calculations show differences between the chaos and non-
chaos cases, and compare these to our observations of the stochastic system.
Figs. 3a, 3b and 3c illustrate the stochastic evolution of the phase space for the non-chaos case of

0
1. 0 during a transient period of t 12 15; compare to Fig. 2a. Figs. 4a, 4b and 4c illustrate the
stochastic evolution of the phase space for the chaos case of
0
0. 1 during a transient period of
t 12 15; compare to Fig. 2b. The differences in the stochastic cases are practically non-existent, and
any differences that are present are likely due to differences in drifts due to changes in the parameters.
6. CONCLUSION
We hav e presented calculations of a two-dimensional time-dependent Dufng system, that has some
support for modeling interactions in macroscopic neocortex, that possesses chaotic and non-chaotic
Path-integral of Dufng model - 9 - Ingber, Srinivasan and Nunez
regions of activity. We embedded this system in moderate noise and investigated the epochs within the
transient periods of these systems to see if there are any apparent differences between the deterministic
and stochastic systems.
Future studies taking hundreds of CPU-hours will investigate time epochs beyond the transients to
see if moderate noise suppresses chaos in such models of neocortex.
ACKNOWLEDGEMENT
RS acknowledges support for this project under NIMH National Research Service Award
1-F32-MH11004-01.
Path-integral of Dufng model - 10 - Ingber, Srinivasan and Nunez
FIGURE CAPTIONS
FIG. 1. Phase space plots of versus

. (a) The evolution of the phase-space for the non-chaos

case of
0
1. 0, after transients have been passed. (b) The evolution of the phase-space for the chaos
case of
0
0. 1, after transients have been passed.
FIG. 2. Phase space plots of versus

. (a) The evolution of the phase-space for the non-chaos

case of
0
1. 0, during the transient period of t 12 15. 5. (b) The evolution of the phase-space for the
chaos case of
0
0. 1, during the transient period of t 12 15. 5.
FIG. 3. Conditional probability distributions as a function of versus

. (a) The stochastic

ev olution of the phase space for the non-chaos case of
0
1. 0 at t 12. (b) The stochastic evolution of
the phase space for the non-chaos case of
0
1. 0 at t 13. 5. (c) The stochastic evolution of the phase
space for the non-chaos case of
0
1. 0 at t 15.
FIG. 4. Conditional probability distributions as a function of versus

. (a) The stochastic

ev olution of the phase space for the chaos case of
0
0. 1 at t 12. (b) The stochastic evolution of the
phase space for the chaos case of
0
0. 1 at t 13. 5. (c) The stochastic evolution of the phase space for
the chaos case of
0
0. 1 at t 15.
Path-integral of Dufng model - 11 - Ingber, Srinivasan and Nunez
REFERENCES
1. P.L. Nunez, Neocortical Dynamics and Human EEG Rhythms, Oxford University Press, New York,
NY, (1995).
2. A. Fuchs, J.A.S. Kelso, and H. Haken, Phase transitions in the human brain: Spatial mode
dynamics, Int. J. Bifurcation Chaos 2 (4), 917-939 (1992).
3. V.K. Jirsa, R. Friedrich, H. Haken, and J.A.S. Kelso, A theoretical model of phase transitions in the
human brain, Biol. Cybern. 71, 27-35 (1994).
4. L. Ingber and P.L. Nunez, Multiple scales of statistical physics of neocortex: Application to
electroencephalography, Mathl. Comput. Modelling 13 (7), 83-95 (1990).
5. P.L. Nunez and R. Srinivasan, Implications of recording strategy for estimates of neocortical
dynamics with electroencephalography, Chaos 3 (2), 257-266 (1993).
6. R. Srinivasan and P.L. Nunez, Neocortical dynamics, EEG standing wav es and chaos, in Nonlinear
Dynamical Analysis for the EEG, (Edited by B.H. Jansen and M. Brandt), pp. 310-355, World
Scientic, London, (1993).
7. Y. Ueda, Steady motions exhibited by Dufngs equation: A picture book of regular and chaotic
motions, in New Approaches to Nonlinear Problems in Dynamics, (Edited by P.J. Holmes), pp.
311-322, SIAM, Philadelphia, (1980).
8. L. Ingber, Statistical mechanics of neocortical interactions. I. Basic formulation, Physica D
5, 83-107 (1982).
9. L. Ingber, Statistical mechanics of neocortical interactions. Dynamics of synaptic modication,
Phys. Rev. A 28, 395-416 (1983).
10. L. Ingber, Statistical mechanics of neocortical interactions: A scaling paradigm applied to
electroencephalography, Phys. Rev. A 44 (6), 4017-4060 (1991).
11. L. Ingber, Statistical mechanics of multiple scales of neocortical interactions, in Neocortical
Dynamics and Human EEG Rhythms, (Edited by P.L. Nunez), pp. 628-681, Oxford University
Press, New York, NY, (1995).
12. L. Ingber, Statistical mechanics of neocortical interactions: Multiple scales of EEG, in Fr ontier
Science in EEG: Continuous Waveform Analysis (Electroencephal. clin. Neurophysiol. Suppl. 45),
Path-integral of Dufng model - 12 - Ingber, Srinivasan and Nunez
(Edited by R.M. Dasheiff and D.J. Vincent), pp. 79-112, Elsevier, Amsterdam, (1996).
13. M. Tabor, Chaos and Integrability in Nonlinear Dynamics, Wiley, New York, (1989).
14. F. Langouche, D. Roekaerts, and E. Tirapegui, Functional Integration and Semiclassical
Expansions, Reidel, Dordrecht, The Netherlands, (1982).
15. L. Ingber, Statistical mechanics of nonlinear nonequilibrium nancial markets: Applications to
optimized trading, Mathl. Computer Modelling 23 (7), 101-121 (1996).
16. M.F. Wehner and W.G. Wolfer, Numerical evaluation of path-integral solutions to Fokker-Planck
equations. I., Phys. Rev. A 27, 2663-2670 (1983).
17. M.F. Wehner and W.G. Wolfer, Numerical evaluation of path-integral solutions to Fokker-Planck
equations. II. Restricted stochastic processes, Phys. Rev. A 28, 3003-3011 (1983).
18. M.F. Wehner and W.G. Wolfer, Numerical evaluation of path integral solutions to Fokker-Planck
equations. III. Time and functionally dependent coefcients, Phys. Rev. A 35, 1795-1801 (1987).
19. L. Ingber, Statistical mechanics of neocortical interactions: Path-integral evolution of short-term
memory, Phys. Rev. E 49 (5B), 4652-4664 (1994).
20. L. Ingber and P.L. Nunez, Statistical mechanics of neocortical interactions: High resolution path-
integral calculation of short-term memory, Phys. Rev. E 51 (5), 5074-5083 (1995).
21. L. Ingber, Statistical mechanical aids to calculating term structure models, Phys. Rev. A
42 (12), 7057-7064 (1990).
22. L. Ingber, M.F. Wehner, G.M. Jabbour, and T.M. Barnhill, Application of statistical mechanics
methodology to term-structure bond-pricing models, Mathl. Comput. Modelling
15 (11), 77-98 (1991).
23. L. Ingber, H. Fujio, and M.F. Wehner, Mathematical comparison of combat computer models to
exercise data, Mathl. Comput. Modelling 15 (1), 65-90 (1991).
24. L. Ingber, Statistical mechanics of combat and extensions, in Toward a Science of Command,
Control, and Communications, (Edited by C. Jones), pp. 117-149, American Institute of
Aeronautics and Astronautics, Washington, D.C., (1993).
25. L. Ingber, Path-integral evolution of multivariate systems with moderate noise, Phys. Rev. E
51 (2), 1616-1619 (1995).
Path-integral of Dufng model - 13 - Ingber, Srinivasan and Nunez
26. K. Binder and D. Stauffer, A simple introduction to Monte Carlo simulations and some specialized
topics, in Applications of the Monte Carlo Method in Statistical Physics, (Edited by K. Binder), pp.
1-36, Springer-Verlag, Berlin, (1985).
27. H.D.I. Abarbanel, R. Brown, J.J. Sidorowich, and L.Sh. Tsimring, The analysis of observed chaotic
data in physical systems, Rev. Mod. Phys. 65 (4), 1331-1392 (1993).
28. N.S. Goel and N. Richter-Dyn, Stochastic Models in Biology, Academic Press, New York, NY,
(1974).
Path-integral of Dufng model Ingber, Srinivasan and Nunez
Figure 1a
Path-integral of Dufng model Ingber, Srinivasan and Nunez
Figure 1b
Path-integral of Dufng model Ingber, Srinivasan and Nunez
Figure 2a
Path-integral of Dufng model Ingber, Srinivasan and Nunez
Figure 2b
Path-integral of Dufng model Ingber, Srinivasan and Nunez
Figure 3a
Path-integral of Dufng model Ingber, Srinivasan and Nunez
Figure 3b
Path-integral of Dufng model Ingber, Srinivasan and Nunez
Figure 3c
Path-integral of Dufng model Ingber, Srinivasan and Nunez
Figure 4a
Path-integral of Dufng model Ingber, Srinivasan and Nunez
Figure 4b
Path-integral of Dufng model Ingber, Srinivasan and Nunez
Figure 4c