Você está na página 1de 19

A Geometrical Theory for Spiral Waves in Excitable Media Author(s): James P.

Keener Source: SIAM Journal on Applied Mathematics, Vol. 46, No. 6 (Dec., 1986), pp. 1039-1056 Published by: Society for Industrial and Applied Mathematics Stable URL: http://www.jstor.org/stable/2101658 Accessed: 21/05/2009 05:03
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=siam. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that promotes the discovery and use of these resources. For more information about JSTOR, please contact support@jstor.org.

Society for Industrial and Applied Mathematics is collaborating with JSTOR to digitize, preserve and extend access to SIAM Journal on Applied Mathematics.

http://www.jstor.org

SIAM J. APPL. MATH. Vol. 46, No. 6, December 1986

1986 Society for Industrial and Applied Mathematics 007

A GEOMETRICAL THEORY FOR SPIRAL WAVES IN EXCITABLE MEDIA*


JAMES P. KEENERt Abstract.In this paper,we developa geometrical theoryfor wavesin excitablereactingmedia.Using singularperturbation arguments dispersionof travelingplane wave trains,we derivean approximate and which has strongresemblance the geometricaldiffraction to theory of theoryof wave front propagation highfrequency wavesin hyperbolic systems,governedby the eikonalequation.Usingthis theory,we study the effectof curvature waves in excitablemedia,specifically, on rotatingspiralpatternsin planarregions. for Fromthis theorywe are able to determine frequencyand wavelength spiralpatternsin excitable, the nonoscillatory media. Keywords.spirals,excitablemedia AMS(MOS)subjectclassifications. 35K57,92A09

1. Introduction. diffraction Geometric theoryhasbeenverysuccessfulin providing approximate(asymptotic) solutions for wave propagationproblems in hyperbolic used systems[8], [15], [16]. Thesetechniques,also knownas raytracing,are currently in a wide varietyof applicationsin geophysicsand electromagnetic theory. The advantage of geometricaloptics is that one can get direct approximate informationabout wave propagationin a varietyof media in the asymptoticlimit of high frequencywaves from equationsthat are substantiallyeasier to solve than the full system of equations.The loss, of course, is that as in any approximatetheory, there are situationswhere the approximate solutions fail to produceaccurateresults. In recentyears,it has been shown that systemsof nonlinearparabolicequations can have solitarywaves and periodicwave trains.Typicalexamplesdescribepropagation of electricalactivityin neuraltissue [6], [7], [12], [25], [33] and wavesof chemical activity in the Belousov-Zhabotinskiireaction [32]. Equations modeling these solutionsarevirtually impossphenomenaarenecessarily complicated. Exactanalytical ible to find and numericalsolutionsof stiffequationsin higherdimensionalmediaare difficultand costly to obtain.The value of an approximate analyticaltheoryis evident. In this paper we develop a geometrical theory for waves in excitablemedia.The theoryis asymptoticsince it applies in the limit that the excitabledynamicsevidence multiple time scales, and wave fronts occur as sharp transitions(boundarylayers) betweenslowly varyingregions.This is in directanalogywiththe weak shock approximationof geometricaloptics in hyperbolicproblems. In the next sectionwe reviewsingularperturbation theoryand dispersionof waves in one spatial dimension.In ? 3 we extend this theory to waves in two dimensions. Finally in ? 4 we apply this theoryto rotatingpatternsin two dimensionsand show how wavelengthand propagationspeed are determined.Further of this theoryto use otherproblemswill appearin forthcoming papers.
* Receivedby the editorsApril15, 1985,and in revisedformJanuary 1986.Portionsof this research 21, wereperformed the Centrefor Mathematical supported the Scienceand by at Biology,OxfordUniversity, ResearchBranchof Engineering ResearchCouncilof GreatBritainGR/C/63595, and at the Mathematics the National Instituteof Arthritis,Diabetes and Digestive and Kidney Diseases, National Institutesof Health. this grantMCS83-01881. Additionally, workwaspartially supported NationalScienceFoundation by of of t Department Mathematics, University Utah, Salt LakeCity, Utah 84112. 1039

1040

JAMES

P. KEENER

2. Wave propagationand dispersion.As a model problem, we consider the equations (2.1)


au=

=2DV2U +f(u, v),

= g(u,v)

where ? <<1. The parameter ? is a dimensionless number which is the ratio of the rates

of reactionof the quantitiesu and v. The diffusioncoefficientis scaled by &2 because we take x on the doubly infinitereal line, so space can be scaled freely. The actualdetailsof f and g are relativelyunimportant exceptthat they represent "typical"dynamicsof excitable media such as the FitzHugh-Nagumoequation [6], [7], [21], [29]. In particular, requirethe null clinef(u, v) =0 to be "cubic-shaped" we withf(u, v) < 0 for all v sufficiently large. The behaviorof g is of little consequence to the derivationof the theory. However,to keep ideas specific, we assumethat the curve g(u, v) = 0 is monotone having one intersectionwith the curvef(u, v) = 0 and that g(u, v) > 0 for u sufficiently large.This informationis depictedin Fig. 1. One importantconsequenceof these assumptionson f(u, v) is that the null cline
f(u, v) = 0 has three solution branches, denoted u = U?(v) or u = Uo(v) where U_(v) _

U0(v) U+(v) wherevercomparisonis appropriate.In fact, U_(v) exists only for v > v_ and U+(v) exists only for v < v+; it is essentialthat v+ > v-. The system(2.1) for one-dimensional spatialdomainscan be treatedusingsingular perturbation arguments [5], [13]. Thesearguments [3], revealthata waveformconsists

INES ~~~~~NULLCL

f (u,v

)0

(u,V

)0

FIG. 1. Typical nullclinesf(u, v) = 0 and g(u, v) = 0 for the excitable system (2.1).

SPIRALS

IN EXCITABLE

MEDIA

1041

of two typical regions. In the slowly varyingregion, diffusionhas little effect, and u and v must satisfy,to lowest orderin E,
(2.2) d at = g(u, v), f(u, v) = 0.

Thisis obtainedby settingE= 0 in (2.1). For reasonsof stability,equation(2.2) implies


that either u = U+(v) or u = U_(v) and that on these solution branches, the slow

dynamicsare governedby (2.3) at = g( U?(v), v)


G?(v).

The fact that u can lie on eitherthe upper branchor lower branchof f(u, v) =0 admitsthe possibilityof spatial discontinuitiesin u. At such discontinuitiesdiffusion becomes important, and regions correspondingto different solution branches of f(u, v) =0 are patched togetherby a moving boundarylayer. The boundarylayer is
found by making the change of variables x = E{ + y(r), t = r, and seeking translation

invariantsolutions of the resultingequations(i.e., set a/ar= 0). To lowest orderin E, we obtain


Du"+ cu'+f(u, vo) = 0, lim f(u, vo)= 0, vo constant, u= u= . c = y'(-).

boundary layer also has u = U+(v).

In addition, the boundarylayer must have the correctorientation,so that, for example, lime,. u(f) = U+(vo)wheneverthe region immediatelyto the right of the

Although much more can be said about the use and interpretationof these excitablemediais capturednicely equations,the behaviorof wavesin one-dimensional by the consistentapplicationof equations(2.3) and (2.4). The main consequenceof this analysisfor our theoryis the specificequation(2.4). If we solve the nonlineareigenvalueproblem
Du"+ u'+f(u, vo) =0, (2.5) lim u(W)= U+(VO),
lim
{e 3-0

u( ) = U_ (vo),

for D fixed, then the speed of propagationof the movingboundarylayer is c = IL(vo) if the boundaryconditions have the same orientation,or c = -,(vo) if the opposite orientationapplies. For most problemsthe solution of the eigenvalueproblem(2.5) must be carried There are two cases where exact analyticalsolutions are known. In out numerically. the special case thatf is piecewise linear [21], [29], [38] f(u, v) = H(u)-u-v (2.6) where H(u) is the Heavisidestep function,one can calculatedirectlythat
(2.7)
-2vo) A ( o) - O.I-2(1

In the cubic case (2.8) f(u, v) = -A(u- UO)(UU)(


- U2)

1042

JAMES

P. KEENER

wherethe roots satisfy uO< ul < u2 and depend on vo, one can again calculatethat
(2.9)
/L(vO) =

2A (uo-2u + U2).

This follows if one makes the guess u' = a(u - UO)(U- u2) and substitutesinto (2.5). the It is useful to note that the sign of ,u is determinedby integrating equation (2.5) to yield
(2.10)
(vo)

00

rU+(vO)

u'2() df=

U ( uO)

f(y, vo) dy.

_00

In general,the speed of propagation the movingboundarylayersis not constant of since in the medium ahead of the jump, the value of vo is changing in time. This complicationmakessolution of the full dynamicpropagation problemdifficultindeed. Thereare two special cases where wave speeds are constant.If the mediumahead of thejumpis at rest,thenthe leadingfronttravelsat the speed determined the constant by value of v in the medium ahead. If there is a periodic wave train of up and down jumps,we also have constantpropagationspeeds, althoughthis can occur if and only if the speed and period are chosen consistently.Suppose we seek solutions traveling to with fronts corresponding v value vo. Their speed is, naturally,c =,u(vo), and the period must be consistent with the outer dynamics. Let v1 be a v-value for which of the null cline f(u, v) =0. We take the down jump to have v value vl. Then from (2.2), the temporalperiod of the wave train must be T = T(vo) where
VI 1 1
))

A (v,) = -, (vo), or if no such value exists take v1= v+ (phase waves), the local maximum

(2.11)

T(vo)= JO g(u+(v), v) g(uv),

dv

to leading order in E. The relationshipbetween c(vo) and T(vo) is, to first order in E, the dispersion relation for the diffusion reaction system (2.1). Typically the c versus T curve is monotone increasing and bounded as T -) 00, although by varying f and g one can easily arrangea wide assortmentof dispersioncurves,and associatedsolitarypulses and/or single transitions[5], [13]. For the specific example of piecewise linear dynamics(2.6) with
(2.12) g(u, v)= u + a, 0<

a< 2

we find that (2.13)


T(v0)=In (
_vO)2

- a2

of which, coupled with (2.7), gives a parametric representation the dispersioncurve. This representation the dispersioncurvebecomes invalid as voapproachesv*, of where Iu+(vo) (u v*) du= 0. (The value v* correspondsto zero speed.) For E$0 fixed, the dispersion curve is known to have two branches,the one just calculated, correspondingto stable wave propagation,and a second branch of slow unstable periodic waves. These two branchescoalesce at a "knee" whose location depends
upon E.

SPIRALS

IN EXCITABLE

MEDIA

1043

Althoughwe shall not use the slow branchof the dispersioncurve,it is important to know wherethe knee occurs.To determinethis branch,we introducethe stretched travelingcoordinatef = (x - Ct)!? for which equations(2.1) become (2.14) Du"+ cu'+f(u, v) = 0,
cv'+
Eg(u, V) =

0.

that the validityof the fast branchfails Ournext step is motivatedby the observations in when c andthe variation v becomesmall.Thus,we take c = e 1/2kU and (2.14)becomes (2.15) Du'+f(u,v) = -e
1/2jlu,

Av= _61/2

g(u

v)

for which we seek periodic solutions. The obvious thing to do is to seek periodic solutions which are power series in 1/2 v1 + * 1/2, u+?U1+ Vo.+ V= -, and , = ,io+ e12 +*-. With this ansatz,we obtain a hierarchyof equations
(2.16) Du' +f(uo, vo) =0, ,uOv =0,

and d (u vo )u= 3u
(7ovl = -g(uo, vo).

(uO, (lzou?+ d-V V0)V1).

We The solutionof (2.16)is straightforward. takpvo= constantandthen finda periodic solution of Dug'+f(uo, vo)=0. Since this is a Hamiltoniansystem which possesses a continuousfamily of periodic solutions for each fixed vo, at this point we have too many solutions. Proceedingto the next order,we require
0Vl = -g(uo, vo).

Naturallyv, must be periodic so we require


P

(2.18)

g(uo(x),

vo)

dx = 0,

whereP is the periodof the chosen solution uo(x).This conditionenablesus to select the correctperiodic solution from among the family of periodic solutions of Dug'+ f(uo, vo)= 0. We denote the period of this correctlychosen solution as P(v0). With this periodic solution uo(x) selected, the solution of (2.17) becomes possible only if condition the orthogonality
(2.19)

|0VA +(UO

A0,+-(u0,VVjid=

v vu10 dx = 0

is satisfied. the Substituting known solution for vl, we have p p Iaf i2 2 uo dx=o
(2.20) -LO
uoj

(uo, o)g([

vo)d7

x dx

Re-expressingthis informationin the original which uniquely determines kLo(vo). coordinatesystem, we find that the wavelength,period and speed of the traveling

1044

JAMES

P. KEENER

wavetrain are, to leading order in E, A = eP(vo), (2.21)

T=
c=

(V)

12Av)

parametrized by vo. The form of the expressions (2.21) will be important to us later. As a specific example, we can calculate this branch of the dispersion curve explicitly for the piecewise linear FitzHugh-Nagumo dynamics (2.6), (2.12). With these choices off and g, equation (2.20) becomes
rP rP

<uLo udx= J
o

uodx
o

which can be evaluated directly using hyperbolic trigonometric functions. In Fig. 2, we plot the two branches of the dispersion curve for the piecewise linear FitzHughNagumo equations in the speed-frequency coordinate system, where frequency is w = 21T/ T, for a = .1, E = .1 and E = .05. These approximate results agree quite well with the exact results published in [29].

3 w
~~DISPERSION CURVE

1.5

0
0 10 20
witha = .1, e = .1 linearFitzHugh-Nagumo FIG. 2. Dispersion curvespeedvs. wfor piecewise equation and e=.05.

SPIRALS

IN EXCITABLE

MEDIA

1045

3. Derivation of the eikonal equation. We want to extend the theory of moving boundary layers in one space dimension to waves in media with higher spatial dimension. To find a boundary layer in R2, we introduce a stretched traveling coordinate system y = Y(--6 7q,r), t = Xr where 6 and 7qare locally orthogonal coordinates. The reason for this particular choice is that we seek translation invariant solutions with as the wavefront coordinate. That is, we seek solutions for which the curves e = constant are level curves of u, and the wavefront motion is depicted entirely by the movement of the coordinate system. In this new coordinate system, partial derivatives take the form (3.1)
',

x = X(--g, 7, r),

du_ au=

1
Yl)(X2+ Y7)

1 auy Y
(X2 Y3

at

J(Xc+

(23

-Y2X3)-au+ (X3Yl -Y3Xi)

yxdld a }a

(3.2)

1 V2u- 1 2x~ ~2
__X

y2a\1
___

aa
2

u+-K2a

+f__

a a/X C( au a'Y7 _i+y, a'Y


-)

-K1-a aj

ul

where Xi, Yi represent partial derivatives with respect to their ith argument, and K1, K2 are curvatures

(33)

Ki=1X2+

Y2)3/2'

The orientation for the f-7q coordinate axes is the same as the usual x-y axes, that is, the positive e axis is rotated 90 degrees in the clockwise direction from the positive '1 axis. Without loss of generality, we take X2+ Y2 = 1 and seek solutions with au/a77 = au/ar =0. With these assumptions and restrictions, we find that (2.1) becomes i) D a2 + (N2+ sDK2)

au

av sN1 av+Eav+ = + ii) N2 a (X2 + y2 +?g(u, v) ar 2(X+Y)1/2 aq where Ni = (X3Yi- Y3Xi)/(Xi+ Yi)1/2, i 1, 2, is the normal velocity of the coordinate
-

(3.4)

-e+f(u, v) = 0,

lines. From equation (3.4ii) we learn that v= vo+ 0(s). If N2 and K2 were independent of e, we could identify equation (3.4i) with (2.5) and require

(3.5)

N2+ eDK2= ,u(VO).

Fortunately, the s variation does not affect this approximation significantly. To see this, we represent the coordinate system locally by (3.6)

~~~ X(,
Y('

77, T) = XO(n, T) + (X2

+ y02 )/2 +

?(

(e2)

')=

Y?'

)+x2

+ y2o )1/2+?{)

Then, by a direct, but tedious, calculation, it follows that N2 = =N2_+ o(e2). Thus, to order _2, N2+ EDK2 is independent of e, justifying the identification (3.5).

1046 As a result

JAMES

P. KEENER

N2+?DK2= g (vo)+ 0(?). One might argue that keeping the curvature correction term ?DK2 makes no difference to this order of approximation. However, as we shall see, K2 need not be independent of ? and in fact, must be retained for spirals since there K2= O(E-2/3). The extra work needed to get the curvature correction in equation (3.5) is further justified by the fact that the easier approximation N = /.k(vo) fails miserably when applied to spiral structures. In general, the value of vo may change with the spatial and temporal position of the wave front unless we restrict our attention to solitary or periodic waves. In this case we have (3.7) N
=

1 (T)

sDK

where ,.u(T) denotes the dispersion relation for wave speed as a function of period T= T(vo), specified by (2.11), and where N-=xlyn
(x2

- YlXn
+ y2V)/2

K =Xn - YnXwn Yn /
(x2 + Y2

are the normal velocity and curvature, respectively, of the wave front at f =0. The equation (3.7) has a very nice geometrical interpretation. Since ,.u(T) is the velocity of plane waves of period T, (3.6) implies that ,.u(T) is the velocity of plane waves of period T, and the normal velocity is.decreased by curvature away from the direction of forward propagation and enhanced by curvature toward the direction of forward propagation with an influence having the same order of magnitude as the diffusion coefficient ED. Note also that the term ?DK has dimensional units of velocity. These observations make good physical sense. For example, a point just ahead of a front with curvaturein the direction of motion will be more quickly excited to threshold than if it were approached by a plane wave, since curvature tends to focus the stimuli necessary to excite the point. The equation (3.7) has been used in studies of flame front propagation [19] and crystal growth [31]. The fact that curvature influences the velocity of waves in excitable media was previously noted in [40], [41]. The equation (3.7) is related to the eikonal equation of geometrical optics. In optics, the eikonal equation is N = g, where /.k depends only on space, and not on the period of the waves. Furthermore, there is no curvature effect on the speed of wave fronts. The approximation N = lk was used in [9] for excitable media problems. In optics there are two standard ways to derive the eikonal equation. One may assume that waves have high frequency, or that solutions are discontinuous along the wavefront [34]. The second of these assumptions is analogous to our problem since we have assumed that transitions between the two possible states of the medium are sharp. The equation (3.7) is the basis of our geometrical theory. We can express this equation in a way that suggests propagation along "rays". Suppose the wavefront at and satisfies XTXI + YTY, =0 (i.e., the y=Y(q,r) e=0 is given by x=X(q,r), r-tangent vector and n-tangent vector are orthogonal), then (3.7) can be rearrangedinto (8XT (3.8) ~~X = = {X( T)- ED (X2n +
-

Xn 1 1

11

y2)3/2
V/2
+

(X2q + Y2

Yl- .kT)Exnyn-X2

Ynxn 232

xn

SPIRALS IN EXCITABLE MEDIA

1047

In this form, we see that the "eikonal" equation is parabolic and that neighboring "rays" interact through an order ? diffusion term. These equations represent a noticeable simplification of (2.1). In (3.8) we have a coupled system of nonlinear parabolic equations in the one space variable q and two state variables X, Y, whereas (2.1) is a system of parabolic equations in 2 space variables x and y and the two state variables u and v with stiff, relaxation dynamics. 4. Rotating spiral waves. Spiral patterns in excitable media occur in a wide variety of situations. For example, rotating waves have been observed in BZ reagent [1], [36], [39], in heart muscle [2], [24], [4], in rat cortex [30], and in the retina of the eye [20]. Much of the analytical work on spiral patterns has. been restricted to self-oscillatory media, especially of A - w type. In this section we discuss the application of our geometrical theory to excitable problems in polar geometry. We begin by looking for k-armed spiral wavefronts of the form
Xj=rcos 0(r)-wt+

)j

j=1,2,

.,k

(4.1) Yj=rsin 0(r)-(t+ k)

where for geometrical reasons we expect 0'w > 0. (Here the wavefront coordinate q is the radial variable r.) Our goal is to find solutions of (3.7) on an annulus ro r l, with the aim of letting ro approach zero. As boundary conditions, we require 0'(ro) = 0 and 0'(rl) = 0 which imply that the spiral arms are orthogonal to the circular boundaries, equivalent to no-flux conditions at the boundaries. If r, = ao, we want 0(r) to be asymptotically linear so that (4.1) is asymptotically Archimedean. If we specify the periodicity T of the solutions, then for consistency we must have (4.2) 2ir
)kT

From (4.1) we directly calculate the normal velocity and curvature as (4.3) (4-3)
N=

~(1 +

~~ ~~~wr
o2)1/2

Kr=4 K (l+

q/2)3/2

r(l +

q 2)1/2,

where i = r0'(r). Before solving (3.7), it is instructive to solve the zeroth order eikonal equation N = .t(T). By simple quadrature we find that (4.4) 0(r)
=

)1

-tan-'(rO

)1

C0

With the change of variables s = ((r2/ r2) -1)1/2 X(s)


=

we obtain wavefronts of the form

rocos s+ros sins,

Y(s) = rosin s - ros cos s which is the parametric representation of the involute of a circle of radius ro. In other words, the involute of a circle solves N = .. The geometrical similarity of spirals to involutes is noted in [36]. There are some significant difficulties with the involute solution. First, for consistency, we must choose ro= ,u(T) T/2ir, but for small ro this may not be possible since the dispersion curve ,u(T) has a knee. Even if there is no knee, an involute

1048

JAMES

P. KEENER

becomes degenerate as ro approaches zero. Thus, this solution fails if the central core is too small. Second, for any value of ro, near the circle of radius ro, the curvature of the involute is arbitrarily large so that curvature effects must be important, and the involute solution cannot be valid there. Finally, the boundary condition at r, cannot be satisfied unless ri = 00. All of these defects are corrected by the higher order theory (3.7). To solve (3.7), we require

(4.6)

= r-dr (1 + .2) dr

(-(1

+ .2)1/2

2 wr

(Without loss of generality, we now take D= 1.) A typical phase portrait for (4.6) is shown in Fig. 3a. The only critical point is a saddle point at r = f = 0. For values of large, the equation is quite stiff with a unique, strongly attracting trajectory. ,/? Trajectories for three different values of t/ e with ,U/ ? = 3 and r, = 9 fixed are shown in Fig. 3b. We view the equation (4.6) as an eigenvalue problem, and wish to find the parameter values ,U/ ? and w/ e for which trajectories connect the points if = 0 at r = rO with qi = 0 at r = rl. Notice that the vector field of (4.6) is monotone increasing in ,u/ ? and decreasing in w/ E.Therefore for any fixed ro, r, (including rO= 0), there are values of of w,,u for which trajectories (4.6) connect qi= 0, r = rOwith qi= 0, r = rl. We can

.5
PHASE PLOT '1 VS R

.25

0 0
FIG. 3a. Phase plot for equation (4.6) with i./ e = 1,

R2

/le = 1. Trajectoriesshownfor r, = 1.4, 1.5 and 1.6.

SPIRALS

IN EXCITABLE

MEDIA

1049

I5

PHASE PLOT

1 VS R

7.5

0 0
representthis relationshipas
(4.7) -= (ro,r,)

10

FIG. 3b. Trajectories for (4.6) with r, = 9, a/ e = 3 and wl e = 2.8, 9m* and 3.2.

where Q is monotone increasingin A, and monotone decreasingin ro and rl. The functionQlis invariant undera changeof scale.The changeof variablesr- ar, ,t - u/a and w - wl/a2 leaves the equation(4.6) invariant,so for any scalar a (4.8) or, if we pick a = 1/rl, (4.9)
w=-f
r

w= Ea 2(

aro,ar,

I-, ro?

r,1

r,

wherethe third argumentis equal to one and no longer noted. Becausethe equation (4.6) is stiff, (4.7) is independentof r, for r, large. we Using perturbation arguments, can show that

1050 and that (4.11)

JAMES

P. KEENER

331 _ 5 + O(x7). ?) Q~(x, = 2 +5x X3+ 2 560 3449600

Numerically, we have determined that for x ' 10, Ql is accurately represented by (4.12)

x,y) = m*x 2_ ax4y2+o(y4)

where m*=0.330958, a=.097. The solution of our problem can be achieved graphically by finding the intersection of the two curves (4.13) 9)
r2

!P

In Fig. 4, we plot these two curves in the ,u - w parameter space for ro= 0. For purposes of illustration we use the specific dispersion curve (2.7), (2.13) for the piecewise linear dynamics (2.6), (2.12) with a = .1 and s = .1 and e = .05. The intersection of these curves specifies the frequency w and asymptotic speed ,.u for a spiral with inner radius ro. Since this intersection depends in general on the inner core size ro, in Fig. 5 is plotted the inner core size ro as a function of w as the intersection moves along the dispersion curve and in Fig. 6 is plotted the asymptotic speed ,u as as a function of

3
W

a0 \

DISPERSIONCURVE
Y~~~~~S.

CRITICAL CURVE

1.5

0
0 10
for r1= 10,
ro
=

20
0 with a = .1, e = .1, .05.

FIG. 4. Plot of dispersion curve and critical curve (4.9)

SPIRALS

IN EXCITABLE

MEDIA

1051

SPEED
INNER RADIUS Ro

8=

1.. 0~~~~~~~~~0

0
FIG. 5. Plot ofinner radius rO as afunction of cw along the dispersioncurve with a of (4.6) (solid surve) andfor an involute (dashed curve) for e =.1, .05.
=

Ro

10

.1, rl =10for solution

inner core size romoving along the dispersion curve. The dashed curve in these figures corresponds to taking ro= u(T)/w which one obtains using involutes. The important thing to notice is that w and ,u are now determined for any core size, even ro= 0, whereas the involute fails for small ro. In these plots we have taken r, = 10, although the results are insensitive to rl. We can give analytical estimates of w and , for small - for ro small. For ro=0, the parameter curve (4.19) is approximated by
2 *

(4.14)

,m

independent of rl. We approximate the dispersion curve , = ,u(2X/ kw) by (4.15) where C c'(v*)G+(v*)G-(v*)
IGr (v*)*Gr(v*)* 1

1052

JAMES

P. KEENER

10
0

INNER RADIUS R

VS.

FREQUENCY

FIG. 6. Plot of wavespeedas a functionof innerradiusrO alongthe dispersion curvewithof for solution (4.6) (solidlcurve),andfor an involute(dashedcurve)for E =.1, .05. of

=1,

r, = lo

c(v) is the eigenvalue of equation (2.4), and c(v*) = 0. This relationship approximates the dispersion curve for large frequencies cl provided cl is not so large as to be near the knee of the dispersion curve. For the piecewise linear FitzHugh-Nagumo dynamics (2.6), (2.12)

(4.16)

C = 2Xw(4-a2).

Using these estimates, we find that the intersection of the curves (4.14) and (4.15) occurs at
\C2m*

1/3 \C

1/3

(4.17)

' =

k2E

'

= Vkm*

T'he intersection of the curves (4.14) and (4.15) does not occur near the knee of the dispersion curve. To see this, note that according to (2.21) the knee of the dispersion curve has frequencyof order E-1/2 . while here we find that cl) is of order E-113. It follows that for small enough E the frequency of the spiral is not determined by the location of the knee of the dispersion curve. The knee of the dispersion curve predicts frequencies which are too large and speeds which are too small. The spiral travectories we find here are Archimedean spirals in the limit r ner. They differ from involutes only in the neighborhood of the origin. The shape of all

SPIRALS

IN EXCITABLE

MEDIA

1053

spirals with ro= 0 and r, large is the same. For r, large and ro=0 we can write (4.7) as
d-(1

+ 4',2)(p(1 + q,2)l/2..

m*p2- if),

dip (4.18) ds dO

ds
where the radial variable is r = (?/,)p(s). form Xj(s, t)=-p(s) A Thus, the spiral wavefronts (4.1) take the

cos (O(s)-wt+)

k, ~~~~~~j=1,2,,,,k

Yj(s,t)=-8 p(s) sin 0(s)-

t+

where p(s), 0(s) are independent of all parameters of the problem. Of course, the scale factor E/l and rotation speed w are determined by the dispersion curve and the critical curve t = m* eu2/E. In Fig. 7 we compare the solution of (4.18) (solid curve) with an involute with the same asymptotic wavelength.

FIG. 7. Plot of spiral wavefrontfor (4.6) (solid curve) and for involute (dashed curve) at the intersection of dispersioncurve with critical curve (see Fig. 4) wl =6.7172, ,ie = 4.511 with r1= 10, e = .1.

1054

JAMES

P. KEENER

Although the core size ro for a fixed frequency is uniquely determined by (4.13), the reverse is not necessarily true. If the dispersion curve ,u = ,( T) is not monotone, then the inner core size ro need not be a monotone function of W.In this situation an experiment which varies core size and determines the frequency w would observe a hysteresis. A numerical simulation [27] on a modification of the FitzHugh-Nagumo equations produced such a hysteresis. This approximate theory predicts a unique wavelength and period for a rotating spiral. However, because of the approximate analysis, the wave fronts cannot be expected to be accurate very close to the center of the spiral. This is because the boundary layer wavefront has a "thickness" of order e, but near the center there is not enough room for such a boundary layer. One expects this approximation to become invalid for inner core size of order e. Even though the wavefronts are not valid near the center, the wavespeed and frequency of the spiral do not change much for small core size ro of order E.This can be seen by estimating the behavior of the solution as a function of inner core size ro. Using (4.12) and (4.14), we find that for ro<<E2/3,
km* + 3m*( 1 _~ m*k2)r+ ,

(4.19) a, = -

k2)

ro+

which is quadratic in ro. The quadratic dependence of ,u and w on ro can also be seen visually in Figs. 5 and 6. 5. Discussion. A geometrical theory for wave fronts in excitable media has been derived using singular perturbation arguments. This theory is similar to geometrical optics in that wave fronts are shock like, and the wave front equation can be interpreted as a generalization of the eikonal equation. Using this theory we have found spiral patterns on a disc. Other important geometries can also be studied. One limitation of this theory is that there are excitable systems which do not have relaxation type dynamics. However, for a wide range of physically important systems, such as the BZ reagent and myocardium, relaxation dynamics provide a reliable approximation to the truth. This theory can and should be compared with numerical and experimental results. In this paper we have applied our techniques to the specific example of piecewise linear FitzHugh-Nagumo dynamics (2.6), (2.11) to get not only qualitative but quantitative predictions on frequency and wavelength of spirals. Unfortunately, accurate numerical solutions of equations (2.1) with ? small are quite difficult to obtain. In a forthcoming paper [14], the equivalent results for the Oregonator equations will be compared with experimental results for BZ reagent. A possible extension of this theory is to use (3.7) in a kinematic way, with the correct dispersion relation, without reference to the smallness of e. How valid such an approach will be is not yet known. However, using kinematics in one spatial dimension, Rinzel and coworkers [23], [28] have been quite successful in studying the dynamics of more complicated periodic structures without assuming relaxation type dynamics. An alternate geometric approach for spiral solutions is suggested in [10], and partially exploited in [17], [22]. This theory applies to systems of diffusion reaction equations with equal diffusion coefficients with a known dispersion relation governing the speed and frequency of periodic traveling waves. The theory does not require

SPIRALS

IN EXCITABLE

MEDIA

1055

different time scales, is valid in the long wave limit and requiresa completeknowledge of the dispersionrelation.The resultingphase plane trajectories spiralsdepend in for a stronglynonlinearway on the dispersionrelation,which complicatessignificantly full analysisof the problem. In [22] an attemptis made to allow unequaldiffusioncoefficientsand to exploit the smallnessof e. However a significantproblemresultsfrom the unorthodoxway that the dispersionrelationshipis defined and used. The phase plane analysis that resultsis, in principle,the same as ours, but is more difficultto decipher.With the results of [22], analytical solutions for the piecewise-linear FitzHugh-Nagumo equationsare possible only in a restricted parameter range. Acknowledgments. author gratefully acknowledges the hospitality of the The Centrefor Mathematical Biology at OxfordUniversityand its director,J. D. Murray wherethis workwas initiated,and discussionswith J. J. Tyson,J. Rinzeland C. Peskin which were invaluable.
REFERENCES [1] K. 1. AGLADZE AND V. I. KRINSKY, Multi-armed vortices in an active chemical medium, Nature, 296 (1982), pp. 424-426. [2] M. A. ALLESSIE, F. I. M. BONKE AND F. J. G. SCHOPMAN, Circus movement in rabbit atrial muscle as a mechanism of tachycardia, Circ. Res., 41 (1977), pp. 9-18. [3] R. G. CASTEN, H. COHEN AND P. A. LAGERSTROM, Perturbationanalysis of an approximationto the Hodgkin-Huxley theory, Quart. Appl. Math., 32 (1975), pp. 365-402. [4] N. EL-SHERIF, R. MEHRA, W. B. GOUGH AND R. H. ZEILER, Ventricularactivation patterns of spontaneous and induced ventricularrhythmsin canine one-day-old myocardial infarction,Circ. Res., 51 (1982), pp. 152-166. [5] P. C. FIFE, Singular perturbationand wave front techniques in reaction-diffusionproblems, Proc. AMSSIAM Symp. on Asymptotic Methods and Singular Perturbation, New York, 1976. [6] R. FITZHUGH, Impulse and physiological states in models of nerve membrane,Biophysics J., 1 (1961), pp. 445-466. [7] , Mathematical models of excitation and propagation in nerve, in Biological Engineering, H. P. Schwan, ed., McGraw-Hill, New York, 1969, pp. 1-85. [8] K. 0. FRIEDRICHS AND J. B. KELLER, Geometrical acoustics II: Diffraction, reflectionand refraction of a weak sphericalor cylindricalshock at a plane interface,J. Appl. Phys., 26, 8 (1955), pp. 961-966. [9] B. C. GOODWIN AND M. H. COHEN, A phase-shift modelfor the spatial and temporalorganization of developing systems, J. Theoret. Biol., 25 (1969), pp. 49-107. [ 10] J. M. GREENBERG, Periodic solutions to reaction-diffusionequation,this Journal, 30 (1976), pp. 199-205. [11] J. M. GREENBERG AND S. P. HASTINGS, Spatial patterns for discrete models of diffusion in excitable media, this Journal, 34 (1978), pp. 515-523. [12] A. L. HODGKIN AND A. F. HUXLEY, A quantitative descriptionof membranecurrentand its application to conduction and excitation in nerve, J. Physics (London), 117 (1952), pp. 500-544. [13] J. P. KEENER, Waves in excitable media, this Journal, 39 (1980), pp. 528-548. [14] J. P. KEENER AND J. J. TYSON, Spiral waves in the Belousov-Zhabotinskii reaction, to appear. [15] J. B. KELLER,Geometricalacoustics I, Thetheoryofweakshocks, J. Appl. Phys., 25 (1954), pp. 1938-1947. [16] , A geometrical theory of diffraction,Proc. Sympos. Appl. Math., 8 (1958), pp. 27-52. [17] S. Koga, Rotating spiral waves in reaction diffusion systems, to appear. [18] V. I. KRINSKY,Mathematical models of cardiac arrhythmias,Pharmac. Ther. B., 3 (1978), pp. 539-555. [19] G. H. MARKSTEIN, Experimentaland theoreticalstudies offlame front stability, J. Aero. Sci., 18 (1951), pp. 199-209.
[20] H. MARTINS-FERREIRA, G. DE OLIVEIRO CASTRO, C. J. STRUCHINEA AND P. S. RODRIGUES,

Circling spreading depression in isolated chick retina, J. Neurophysics, 37 (1974), pp. 773-784. [21] H. P. MCKEAN, Nagumo's equation, Adv. in Math., 4 (1975), pp. 209-223. [22] A. S. MIKHAILOV AND V. I. KRINSKY, Rotating spiral waves in excitable media: the analytical results, Physica D, 9 (1983), pp. 346-371. [23] R. MILLER AND J. RINZEL, The dependenceof impulsepropagation speed onflringfrequency, dispersion for the Hodgkin-Huxley models, Biophysics J., 34 (1981), pp. 227-259.

1056

JAMES P. KEENER

[24] G. R. MINES, On circulating excitation on heart muscles and their possible relation to tachycardia and fibrillation, Trans. Roy. Soc. Canada, 4 (1914), pp. 43-53. [25] R. M. MIURA, Nonlinear waves in neuronal cortical structures,in Nonlinear Phenomena in Physics and Biology, R. H. Enns, B. L. Jones, R. M. Miura and S. S. Rangnekar, eds., Plenum, New York, 1981, pp. 369-400. [26] G. K. MOE, W. C. RHEINBOLDT AND J. A. ABILDSKOV, A computermodel of atrial fibrillation, Am. Heart J., 67 (1964), pp. 200-220. [27] A. M. PERTSOV, E. A. ERMAKOVA AND A. V. PANFILOV, Rotating spiral waves in a modified FitzHugh-Nagumo equation model, Physica 14 D, (1984) pp. 117-124. [28] J. RINZEL AND K. MAGINU, Kinematic analysis of spatio-temporalwave patterns, Bordeaux, 1984, to appear. [29] J. RINZEL AND J. B. KELLER, Traveling wave solutions of a nerve conduction equation, Biophysics J., 13 (1973), pp. 1313-1337. [30] M. SHIBATA AND J. BURES, Reverberation cortical of spreading depression alongclosedlooppathways in rat cerebral cortex, J. Neurophysics, 35 (1972), pp. 381-388. [31] D. TURNBULL, Phase changes, in Solid State Physics, 3, F. Seitz and D. Turnbull, eds., Academic Press, New York, 1956. [32] J. J. TYSONAND P. C. FIFE, Targetpatterns in a realistic model of the Belousov-Zhabotinskii reaction, J. Chem. Phys., 73 (5) (1980), pp. 2224-2237. [33] F. J. L. VAN CAPELLE AND D. DURRER, Computersimulation of arrhythmiasin a networkof coupled excitable elements, Circ. Res., 47 (1980), pp. 454-466. [34] G. B. WHITHAM,Linear and Nonlinear Waves, Wiley Interscience, New York, 1974. Themathematical [35] N. WEINER AND A. ROSENBLUETH, formulation theproblem conduction of of of impulses in a network of connected excitable elements, specifically in cardiac muscle, Arch. Inst. Cardiol. Mex., 16 (1946), pp. 205-265. [36] A. T. WINFREE, Spiral waves of chemical activity, Science, 175 (1972), pp. 634-636. , The Geometry of Biological Time, Springer, Berlin, New York, 1980. [37] , Rotating solutions to reaction diffusion equations in simply-connectedmedia, American Mathe[38] matical Society-Society for Industrial and Applied Mathematics Proc., 8 (1974), pp. 13-31. [39] A. M. ZHABOTINSKY AND A. N. ZHAIKIN, Autowave processes in a distributed chemical system, J. Theoret. Biol., 40 (1973), pp. 45-61. [40] V. S. ZYKOV AND A. A. PETROV, Role of the inhomogeneity of an excitable medium in the mechanism of self-sustained activity, Biophysics, 22 (1977), pp. 307-314. [41] V. S. ZYKOV,Analytical evaluation of the dependenceof the speed of excitation wave in a two-dimensional excitable medium on the curvatureof its front, Biophysics, 25 (1980), pp. 906-911.

Você também pode gostar