0810.4721/rf.tex
1: %\documentclass[aps,pre,preprint,amsfonts,amsmath,showpacs,a4paper]{revtex4}
2: \documentclass[aps,pre,twocolumn,amsfonts,amsmath,showpacs,a4paper]{revtex4}
3: \usepackage{color}
4: \usepackage{epic}
5: \usepackage{graphicx}
6: \usepackage{psfrag}
7: \usepackage{times}
8: 
9: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
10: % Keysaving macros
11: 
12: % Abbreviations 
13: \newcommand{\eg}[1]%                    % exempli gratia
14:   {{\it e.g.\/}\ifx#1.\else\expandafter#1\fi}
15: \newcommand{\eq}[1]{(\ref{#1})}         % ref to an equation
16: \def\eqtwo(#1,#2){(\ref{#1},\ref{#2})}  % ref to two eqns
17: \newcommand{\etal}[1]%                  % et alia
18:   {{\it et al.\/}\ifx#1.\else\expandafter#1\fi}
19: \newcommand{\Fig}[1]{Fig.~\ref{#1}}     % ref to a fig in start of a stce
20: \newcommand{\Figs}[1]{Figures~\ref{#1}} % ref to figs in a start of a stce
21: \newcommand{\fig}[1]{fig.~\ref{#1}}     % ref to a fig 
22: \newcommand{\ie}[1]%                    % id est
23:   {{\it i.e.\/}\ifx#1.\else\expandafter#1\fi}
24: 
25: \newcommand{\dblfigure}[3]%             % double coulumn figure
26:   {\begin{figure*}[tbp]#1\caption[]{#2}\label{#3}\end{figure*}}
27: \newcommand{\sglfigure}[3]%             % single column figure
28:   {\begin{figure}[tbp]#1\caption[]{#2}\label{#3}\end{figure}}
29: 
30: % Common notations
31: \renewcommand{\@}{\partial}             % partial differential
32:   \newcommand{\<}{\langle}              % brac and 
33: \renewcommand{\>}{\rangle}              %   ket of an inner product
34: \newcommand{\const}{\mathrm{const}}     % constant
35: \renewcommand\d{{\mathrm d}}            % ordinary differential
36: \newcommand{\e}{\mathrm{e}}             % Euler's number
37: \renewcommand{\Im}[1]{{\rm Im}\left(#1\right)}	% Imaginary part
38: \renewcommand{\i}{\mathrm{i}}           % imaginary unit
39: \newcommand{\inner}[2]%                 %  the inner product
40:   {\left\<#1\, , \,#2\right\>}
41: \newcommand{\Mx}[1]{%                   % mx and col vec const
42: \left[\begin{array}{cccccccc}#1\end{array}\right]}
43: \newcommand{\mx}[1]{\mathbf{#1}}        % mx and col vec vars
44: \renewcommand{\Re}[1]{{\rm Re}\left(#1\right)}	% Real part
45: \newcommand{\Real}{\mathbb{R}}          % real set
46: \newcommand{\T}{^{\mathrm{T}}}          % matrix transposition 
47: 
48: % Specific notations
49: \newcommand{\Ampl}{A}                   % solution amplitude for visualization
50: \newcommand{\D}{\mx{D}}                 % diffusion matrix
51: \newcommand{\dfdu}{\@_{\u}\f(\U)}       % kinetics Jacobian
52: \newcommand\dtime{\dot }                % ordinary time derivative
53: \newcommand{\F}{F}                      % perturb projections - generic
54: \newcommand{\f}{\mx{f}}                 % reaction rates
55: \newcommand{\g}{\mx{g}}                 % spiral perturbation
56: \newcommand{\h}{\mx{h}}                 % perturbation
57: \renewcommand{\L}{\mathcal{L}}          % the linearised operator
58: \newcommand{\Lp}{\L^{+}}                % the adjoint linearised operator
59: \renewcommand{\O}{\mathcal{O}}          % the approximation oder 
60: \newcommand{\para}{a}                   % FHN a parameter
61: \newcommand{\parb}{b}                   % FHN b parameter
62: \newcommand{\pareps}{\varepsilon}       % FHN epsilon parameter
63: \newcommand{\R}{{\vec R}}               % R vector (s.w. centre coords) 
64: \newcommand{\RF}[1]{\mx{W}^{(#1)}}      % left eigenvectors
65: \renewcommand{\r}{{\vec r}}             % r vector (any pt on the plane)
66: \newcommand{\Tr}[1]{\mx{V}\ifx#1.\else^{(#1)}\fi} % right eigenvectors
67: \newcommand{\U}{\mx{U}}                 % unperturbed soln (s.w.)
68: \renewcommand{\u}{\mx{u}}               % concentration field
69: \newcommand{\vF}{{\vec{\F}}}            % spatial drift velocity
70: 
71: % Numerics
72: \newcommand{\Avg}[3]%                   % angular average
73:   {\langle{#1}\rangle\ifx#2.\else^{(#2)}\fi_{#3}}
74: \newcommand{\anaTr}[1]%                  % right eigenvectors analytical
75:   {\breve{\mx{V}}\ifx#1.\else^{(#1)}\fi}
76: \newcommand{\dr}{\Delta\rho}            % radius step
77: \newcommand{\dt}{\Delta\theta}          % angle step
78: \newcommand{\intdist}[1]%               % L2 distance
79:   {\mathcal{D}_{#1}}
80: \newcommand{\maxdist}[1]%               % C0 distance
81:   {\mathcal{D}^{\prime}_{#1}}
82: \newcommand{\interr}[1]%                % L2 error
83:   {\mathcal{E}_{#1}}
84: \newcommand{\maxerr}[1]%                % C0 error
85:   {\mathcal{E}^{\prime}_{#1}}
86: \newcommand{\Nr}{N_{\rho}}              % num of steps in radius
87: \newcommand{\Nt}{N_{\theta}}            % num of steps in angle
88: \newcommand{\Ntot}{N}                   % tot num of vars of discr pblm
89: \newcommand{\numlambda}{\hat\lambda}    % num evalues
90: \newcommand{\nummu}{\hat\mu}            % num evalues
91: \newcommand{\numomega}{\hat\omega}      % freq numerical
92: \newcommand{\numRF}[1]%                 % RF numerical
93:   {\hat{\mx{W}}\ifx#1.\else^{(#1)}\fi}
94: \newcommand{\numTr}[1]%                 % GM numerical
95:   {\hat{\mx{V}}\ifx#1.\else^{(#1)}\fi}
96: \newcommand{\numU}{\hat\U}              % spiral numerical
97: \newcommand{\Ortana}{O_a}               % orthogonality analytical
98: \newcommand{\Ortnum}{O_n}               % orthogonality numerical
99: \newcommand{\rp}{\rho_{\max}}           % disk radius
100: \newcommand{\B}{{\cal{B}}}            % the whole big disk
101: \renewcommand{\S}{{\cal{S}}}            % smaller disk
102: 
103: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
104: \begin{document}
105: 
106: 
107: \title{Computation of the response functions of spiral waves in active media}
108: 
109: \author{I.V.~Biktasheva}
110: \affiliation{Department of Computer Science, University of Liverpool, 
111:      Ashton Building, Ashton Street, Liverpool L69 3BX, UK}
112: 
113: \author{D.~Barkley}
114: \affiliation{Mathematics Institute, University of Warwick, Coventry CV4 7AL, UK}
115: 
116: \author{V. N.~Biktashev}
117: \affiliation{Department of Mathematical Sciences, University of Liverpool, 
118:    Mathematics \& Oceanography Building, Peach Street, Liverpool, L69 7ZL, UK}
119: 
120: \author{G.V.~Bordyugov}
121: \affiliation{former: Department of Computer Science, University of Liverpool, 
122:      Ashton Building, Ashton Street, Liverpool L69 3BX, UK}
123: 
124: \altaffiliation{Present: The University of Potsdam, Campus Golm,Department of Physics 
125: and Astronomy (Haus 28), Karl-Liebknecht-Strasse 24/25, 14476 Potsdam, Germany}
126: 
127: \author{A.J.~Foulkes}
128: \affiliation{Department of Mathematical Sciences, University of Liverpool, 
129:    Mathematics \& Oceanography Building, Peach Street, Liverpool, L69 7ZL, UK}
130: 
131: 
132: \date{\today}
133: 
134: \begin{abstract}
135:   Rotating spiral waves are a form of self-organization observed in
136:   spatially extended systems of physical, chemical, and biological
137:   nature. A small perturbation causes gradual change in spatial
138:   location of spiral's rotation center and frequency, i.e. drift. The
139:   response functions (RFs) of a spiral wave are the eigenfunctions of
140:   the adjoint linearized operator corresponding to the critical
141:   eigenvalues $\lambda = 0, \pm i\omega$. The RFs describe the spiral's
142:   sensitivity to small perturbations in the way that a spiral is
143:   insensitive to small perturbations where its RFs are close to
144:   zero. The velocity of a spiral's drift is proportional to the 
145:   convolution of RFs
146:   with the perturbation. Here we
147:   develop a regular and generic method of computing the RFs of
148:   stationary rotating spirals in reaction-diffusion equations. We
149:   demonstrate the method on the FitzHugh-Nagumo system and 
150:   also show convergence of the method with
151:   respect to the computational parameters, i.e.  discretization steps
152:   and size of the medium.  The obtained RFs are
153:   localized at the spiral's core.
154: \end{abstract}
155: 
156: \pacs{%
157:   02.70.-c, % Computational techniques, simulations
158:   05.10.-a, % Computational methods in statistical physics and nonlinear dynamics 
159:   82.40.Bj,% Oscillations, chaos, and bifurcation
160:   82.40.Ck, % Pattern formation in reactions with diffusion
161:   87.10.-e % General theory and mathematical aspects 
162: }
163: 
164: \maketitle
165: 
166: \section{Introduction}
167: 
168: Autowave vortices, or spiral waves in two-dimensions (2D), are types 
169: of self-organization observed 
170: in dissipative media of physical \cite{%
171:   Frisch-etal-1994,%
172:   Yu-etal-1999,%
173:   Madore-Freedman-1987,%
174:   Schulman-Seiden-1986%
175: }, chemical \cite{%
176:   Zhabotinsky-Zaikin-1971,%
177:   Jakubith-etal-1990, %
178:   Agladze-Steinbock-2000%
179: }, and biological nature \cite{%
180:   Allessie-etal-1973,%
181:   Gorelova-Bures-1983,%
182:   Alcantara-Monk-1974,%
183:   Lechleiter-etal-1991,%
184:   Carey-etal-1978,%
185:   Murray-etal-1986%
186: }, where wave propagation is
187: supported by a source of energy stored in the medium.
188: The common feature of all these phenomena is that
189: they can be mathematically described, with various degrees of 
190: accuracy, by reaction-diffusion partial differential equations,
191: %
192: \begin{equation}
193: \@_t\u = \f(\u) + \D \nabla^2 \u, \quad 
194: \u,\f\in\Real^\ell,\; 
195: \D\in\Real^{\ell\times\ell},\;
196: \ell\ge2,					\label{RDS}
197: \end{equation}
198: %
199: where $\u(\r,t)=(u_1,\dots u_{\ell})\T$ 
200: is a column-vector of the reagent concentrations,
201: $\f(\u)=(f_1,\dots f_{\ell})\T$ is a column-vector of the reaction rates, 
202: $\D$ is the matrix of diffusion coefficients, and  
203: $\r\in\Real^2$ is the vector of coordinates on the plane.
204: 
205: 
206: The existence of vortices is not due to singularities in the medium but
207: is determined only by development from initial conditions.
208: %
209: A rigidly rotating spiral wave solution to the system \eq{RDS} has the form 
210: % 
211: \begin{equation}
212: \tilde \U=\U(\rho (\r-\R),\vartheta (\r-\R) + \omega t - \Phi) ,
213:                                       \label{SW}
214: \end{equation}
215: %
216: where 
217: $\rho(\r-\R),\vartheta(\r-\R)$ are polar coordinates centered at $\R$, 
218: vector $\R=(X,Y)\T$ 
219: defines the center of rotation, and $\Phi$ is the initial rotation phase.
220: For a steady, \ie\ rigidly rotating, 
221: spiral $\R$ and $\Phi$ are constants. The system of reference 
222: co-rotating with the spiral's  initial phase and angular velocity $\omega$ around the 
223: spiral's center of rotation is called the system of reference of the 
224: spiral. In this system of 
225: reference, $\R=0$, $ \Phi=0$, and the polar angle is given by $\theta =
226: \vartheta + \omega t$. In this frame the spiral 
227: wave solution $\U(\rho,\theta)$ does not depend on time and satisfies the
228: equation 
229: %
230: \begin{equation}
231: \f(\U) - \omega \U_\theta + \D \nabla^2 \U = 0 . 	\label{SW-own}
232: \end{equation}
233: %
234: In this equation, the unknowns are the field $\U(\rho,\theta)$ and the scalar $\omega$.
235: 
236: A slightly perturbed steady spiral wave solution
237: %
238: \begin{equation*}
239: \tilde \U(\rho,\theta,t)=\U(\rho,\theta)+ \epsilon \g(\rho,\theta,t), \quad 
240: 		\g\in\Real^\ell, \quad 0 < \epsilon \ll 1,
241: \end{equation*}
242: %
243: substituted in \eq{RDS}, 
244: at leading order in $\epsilon$,
245: yields the evolution equation for the perturbation $\g$,
246: %
247: \begin{equation*}
248: \@_t \g = \dfdu \g 
249:       - \omega\@_\theta \g + \D\nabla^2 \g.
250: \end{equation*}
251: %
252:  
253: Thus, the linear stability spectrum of a steady spiral  
254: % 
255: \begin{equation}
256:   \L \Tr. = \lambda \Tr.                \label{right-evp}
257: \end{equation}
258: %
259:  is defined by the linearized operator
260: \begin{equation}
261: \L = \D\nabla^2 - \omega\@_\theta + \dfdu.
262:                                         \label{L}
263: \end{equation}
264: %
265: 
266: The operator $\L$ has critical ($\Re{\lambda}=0$) eigenvalues
267: \begin{equation}
268:   \lambda_n=\i n\omega, \quad n=0,\pm1,   \label{lambdas}
269: \end{equation}
270: %
271: which correspond to eigenfunctions related to equivariance of \eq{RDS}
272: with respect to translations and rotations, \ie\ ``Goldstone modes'' (GMs)
273: \cite{%
274:   Biktashev-1989,%
275:   Biktashev-1989a,%
276:   Barkley-1992,%
277:   Biktashev-Holden-1995%
278: }
279: %
280: \begin{eqnarray}
281: \Tr{0} &=&  
282:   - \@_\theta \U(\rho,\theta), \nonumber\\
283: \Tr{\pm1} &=& 
284: % -(\@_x\pm i\@_y)\U(\rho,\theta)\nonumber\\ &=&
285:   -\frac12 \e^{\mp\i\theta} \left(
286:     \@_\rho\mp\i\rho^{-1}\@_\theta
287:   \right) \U(\rho,\theta) .  \label{Goldstone} 
288: \end{eqnarray}
289: %
290: The stability spectra of steady spiral waves was originally obtained
291: numerically by Barkley \cite{Barkley-1992}. Subsequently the spectrum was analysed for
292: infinite and large bounded domains by Sandstede and Scheel~\cite{%
293:   Sandstede-Scheel-PhysD-2000,%
294:   Sandstede-Scheel-PhysRevE-2000,%
295:   Sandstede-Scheel-PhysLett-2001%
296: } 
297: with follow-on numerical investigations by Wheeler and Barkley
298: \cite{Wheeler-Barkley-2006} confirming the large domain behavior
299: of the stability spectrum.
300: 
301: In a slightly perturbed problem
302: %
303: \begin{equation}
304: \@_t\u = \f(\u) + \D \nabla^2 \u + \epsilon \h, \quad 
305: 		\h\in\Real^\ell, \quad 0 < \epsilon \ll 1,
306:                                        \label{RDS_pert}
307: \end{equation}
308: %
309: where $\epsilon \h(\u,\r,t)$ is some small perturbation, 
310: spiral waves may drift, \ie change rotational phase and/or center location.
311: Then, the center of rotation and the initial phase are no longer constants 
312: but become functions of time, $\R=\R(t)$ and $\Phi=\Phi(t)$.
313: 
314: In linear approximation, assuming that 
315: \[
316:   \dtime \R, \; \dtime \Phi = \O (\epsilon),
317: \]
318: the drifting spiral wave solution can be represented as 
319: % 
320: \begin{equation}
321: \tilde \U=\U(\rho (\r-\R(t)),\vartheta (\r-\R(t)) + \omega t - \Phi(t)) + \epsilon \g(\r, t),
322:                                       \label{SW_pert}
323: \end{equation}
324: %
325: where $\epsilon \g(\r, t) \;$ is a small perturbation of the steady spiral wave solution $\U$.
326: 
327: Then, the solution perturbation $\g$ in the laboratory frame of reference will satisfy the linearized system % \cite{Biktashev-Holden-1995} 
328: 
329: \begin{eqnarray}
330: (\partial_t &-& \D\nabla^2 - \dfdu) \g
331: \nonumber\\&&
332: = \h(\u,\r,t) - \frac{1}{\epsilon}(\dtime \R\cdot\nabla +
333: \dtime \Phi\,\partial_\theta)\U.
334:                        \label{g_eqn}             
335: %
336: \end{eqnarray}
337: %
338: The solvalability condition for equation \eq{g_eqn} for $\g \;$, \ie\ Fredholm 
339: alternative, re-written in the spiral frame
340: of reference,  requires that the free term must be orthogonal to the kernel of the adjoint operator to $\L$ defined in \eq{L}. 
341: This leads to the following system of equations for the drift velocities 
342: 
343: %
344: \begin{equation}
345: \dtime\Phi = \epsilon \F_0(\R,t), \quad
346: \dtime\R=\epsilon\vF_1(\R,t).             \label{ptb}
347: \end{equation}
348: %
349: 
350: Thus, the drift velocities $\dtime \Phi \;$ and $\dtime \R \;$ are determined by the ``forces'' 
351: $\F_0 \;$ and $\vF_1=\left(\Re{F_1},\Im{F_1}\right)\T$ which, after sliding averaging
352: (more specifically, central moving average)
353: over the spiral wave rotation 
354: period, can be expressed \cite{Biktashev-Holden-1995} as  
355: %
356: \begin{eqnarray}
357: &&  \F_n(\R,t) =
358:   \e^{ \i n \Phi }
359:   \oint\limits_{t-\pi/\omega}^{t+\pi/\omega} 
360:   \frac{\omega\d\tau}{2\pi}   
361:   \e^{-\i n\omega\tau} 
362: \nonumber\\ &&
363:   \times \inner{
364:     \RF{n}\left(\rho (\r-\R),\vartheta (\r-\R)+\omega\tau-\Phi\right)
365:   }{
366:   \h(\r,\tau)
367:   } , 
368: \nonumber\\ &&
369:   	n=0, \pm1.
370:   					\label{forces}
371: \end{eqnarray}
372: (of course, $F_{-1}=\bar{F_1}$).
373: Here 
374: $\inner{\cdot}{\cdot}$ stands for the scalar product in functional space,
375: \[
376:   \inner{\mx{w}}{\mx{v}} = \int\limits_{\Real^2}
377:   \overline{\mx{w}(\r)}\,\T \mx{v}(\r) \,\d^2\r .
378: \]
379: The kernels $\RF{n}$ of convolution-type integrals in \eq{forces} 
380: are the spiral wave's \emph{response functions} (RFs), 
381: \ie, the critical eigenfunctions 
382: %
383: \begin{equation}
384: \Lp \RF{n} = \mu_n\RF{n}, 
385:                       \label{left-evp} 
386: \end{equation} 
387: %
388: where
389: %
390: \begin{equation}
391:   \mu_n=-\i\omega n,\quad n=0,\pm1,      \label{mus}
392: \end{equation}
393: % 
394: of the adjoint linearized operator: 
395: % 
396: \begin{equation} \Lp = \D\nabla^2 + \omega\@_\theta + \left(\dfdu\right)\T ,  
397:                                \label{Lp}
398: \end{equation}
399: %
400: chosen to be biorthogonal 
401: %
402: \begin{equation}
403: \inner{ \RF{j} }{ \Tr{k} }=\delta_{j,k} ,  		\label{norm}
404: \end{equation}
405: %
406: to the Goldstone modes \eq{Goldstone}. 
407: Note that the RFs do not depend on time, \ie\ are functions of the 
408: coordinates only, in the co-rotating system of reference.
409: 
410: The asymptotic theory just outlined reduces the description of the
411: smooth dynamics of spiral waves from the system of nonlinear
412: partial differential equations \eq{RDS} to the system of ordinary
413: differential equations \eq{ptb}, describing the movement of the
414: core of the spiral and the shift of its angular velocity.
415: %
416: Several qualitative results in the asymptotic theory of spiral and
417: scroll dynamics have been obtained without the use of response functions,
418: \eg.~\cite{%
419:   Zykov-1987,%
420:   Davydov-etal-1988,%
421:   Biktashev-1989a,%
422:   Keener-Tyson-1990,%
423:   Keener-Tyson-1991,%
424:   Davydov-etal-1991,%
425:   Biktashev-Holden-1994,%
426:   Biktashev-Holden-1995,%
427:   Biktashev-1996,%
428:   Krinsky-etal-1996,%
429:   Henry-2004%
430: }.  However, an explicit knowledge of RFs makes possible a
431: quantitative description, which obviously can be
432: much more efficient for the understanding and control of spiral
433: wave dynamics in numerous applications, \eg\ control of re-entry
434: in the heart.
435: 
436: The asymptotic properties of the RFs at large distances are crucial for convergence of 
437: the convolution integrals in \eq{forces}.
438: An early version of the asymptotic theory,
439: developed by Keener~\cite{Keener-1988} for scroll wave dynamics, 
440: considered the RFs asymptotically
441: periodic in the limit $\rho\to\infty$, in much the same way as 
442: spiral waves are, thus requiring an artifical cut-off procedure to
443: tackle the divergence of the integrals in \eq{forces} 
444: following from such an asumption.
445:   
446: Based on observations and empirical data of
447: spiral wave dynamics,
448: Biktashev~\cite{Biktashev-1989,Biktashev-etal-1994} 
449: conjectured
450: that the response functions quickly decay at large $\rho$,
451: \ie\ are effectively localized.
452: %
453: This conjecture implies that the
454: integrals in \eq{forces} converge and no cut-off procedure is
455: required.
456: 
457: 
458:       
459: To prove existence of the localized responce functions,
460: Biktasheva \etal\ \cite{Biktasheva-etal-1998} explicitly
461: computed them in the complex Ginzburg-Landau equation (CGLE) for a
462: particular set of parameters. Those computations exploited an
463: additional symmetry present in the CGLE, which permitted the reduction
464: of the 2D problem to the computation of 1D components. The
465: computations were verified by numerical convergence of the
466: method with respect to the space discretisation and the size of
467: the medium. Following this work, the computed RFs were successfully used for
468: quantitative prediction of the spiral's resonant drift and drift
469: due to media
470: inhomogeneity~\cite{Biktasheva-etal-1999,Biktasheva-2000}. By
471: explicitly computing the RFs in the CGLE for a broad range of
472: the model's parameters, Biktasheva and Biktashev~\cite{
473:   Biktasheva-Biktashev-2001,%
474:   Biktasheva-Biktashev-2003%
475: } showed that the RFs are localized for stable spiral wave
476: solutions and qualitatively change at crossing the
477: charachteristic lines in the model parameter plane.
478:  
479: Recently, there has been a significant theoretical progress in
480: mathematical treatment of the localization of the response
481: functions. Sandstede and Scheel~\cite[Corollary
482: 4.6]{Sandstede-Scheel-2004} analytically proved such
483: localization for one-dimensional wave dislocations, which may be
484: considered as analogues of a spiral wave in one spatial
485: dimension. Hopefully this can be extended to two spatial
486: dimensions, \ie\ to spiral waves.
487: 
488: For cardiac applications, dynamics of
489: spiral waves in \emph{excitable} media is more important than in 
490: \emph{oscillatory} media such as the CGLE,
491: as most cardiac tissues are
492: excitable. 
493: These models do not allow 
494: % meaningful 
495: reduction to 1D, making quantitatively accurate computation of the response functions more challenging. 
496: So far, the response functions have been computed 
497: in the Barkley \cite{Hamm-1997,Henry-Hakim-2002} and FitzHugh-Nagumo
498: \cite{Biktasheva-etal-2006} models of excitable media.
499: %
500: For the chosen sets of model parameters, the computed RFs appeared
501: effectively localized in the vicinity of the spiral wave core.  
502: Hamm~\cite{Hamm-1997} and
503: Biktasheva~\etal~\cite{Biktasheva-etal-2006} 
504: calculated RFs on
505: Cartesian grids, but the accuracy was not sufficient for quantitative
506: prediction of drift. Hakim and Henry~\cite{Henry-Hakim-2002}
507: took the advantage of a polar grid and Barkley model 
508: to compute the spiral wave solution with an
509: accuracy of
510: $10^{-8}$ and RFs with accuracy $10^{-6}$
511: (both in the sense of $l_2$-norm of the residue of the discretized equations) 
512: leading to
513: quantitative prediction of drift velocities with about 4\%
514: accuracy. 
515: % It is not obvious from the paper \cite{Henry-Hakim-2002}
516: % how to improve the method further if a higher accuracy is required.
517: 
518: Encouraging as these results are, there is a need for a more 
519: computationally efficient, accurate and robust
520: method to compute the response functions of spiral waves in a variety
521: of excitable media with required accuracy.
522: The aim of this paper is to present a
523: method which is superior to previous methods used to 
524: compute response functions 
525: and to demonstrate that it works for stationary rotating
526: spirals in FitzHugh-Nagumo system. We also
527: demonstrate convergence of the method with respect to the
528: computational parameters, \ie\ discretization steps and size of the
529: medium, and show that the method is vastly more
530: efficient than the
531: methods used before~\cite{Henry-Hakim-2002,Biktasheva-etal-2006}.
532: % The obtained response functions are 
533: % indeed localized at the spiral's core.
534: 
535: \section{Methods}
536: 
537: \subsection{Computations}
538: 
539: To compute the response functions, we use methods similar to those described in
540: \cite{Barkley-1992,Wheeler-Barkley-2006}. 
541: 
542: The nonlinear problem \eq{SW-own} is considered on a disk $\rho\le\rp$, with
543: homogeneous Neumann boundary conditions,
544: $\partial_{\rho}\U(\rp,\theta)=0$.  The fields are discretized on a
545: regular polar grid $(\rho_j,\theta_k)=(j\dr,k\dt)$ where $0<j\le\Nr$
546: and $0\le k<\Nt$ plus the center point $\rho=0$. Hence there are
547: $\Nr\Nt+1$ grid points and correspondingly $\Ntot=\ell(\Nr\Nt+1)$
548: unknowns and the same number of equations in the discretization of
549: \eq{SW-own}. For the inner points $j<\Nr$, the $\rho$-derivatives are calculated via second-order
550: central differences.  The $\theta$-derivatives are calculated using Fornberg's
551: {\ttfamily weights.f} subroutine \cite{Fornberg-1998} which uses all
552: $\Nt$ values so, in theory, provides an approximation of
553: $\theta$-derivatives of the order of $\Nt$. 
554: The discretization of the
555: Laplacian at the center point is via the difference between the
556: average around the innermost circle $\rho=\dr$ and the center point,
557: and the approximation at $j=\Nr$ takes into account the boundary conditions at
558: $\rho=\rp$.
559: 
560: The discretized nonlinear steady-state spiral problem \eq{SW-own} is
561: solved by Newton's method, starting from initial approximations obtained
562: by interpolation of results of simulations of the time-dependent problem
563: \eq{RDS} using EZSPIRAL. % The Newton's 
564: The Newton iterations involve inversion
565: of the linearized matrix which has a banded structure with the
566: bandwidth $1+2\ell\Nt$. 
567: This is achieved by the appropriate ordering of
568: the unknowns of the discretized problem within the $\Ntot$-dimensional
569: vector of unknowns, so that the index enumerating components of reagent
570: vectors from $\Real^{\ell}$ varied fastest, followed by the index
571: enumerating angular grid points $k\dt$, followed by the index
572: enumerating the radial grid points $j\dr$. 
573: 
574: The thus posed discretized nonlinear problem inherits the symmetry of
575: \eq{SW-own} with respect to rotations. To select a unique solution out
576: of a family of solutions generated by this symmetry, we impose a
577: ``pinning condition'' of the form $U_{\ell_*}(j_*\dr,k_*\dt)=u_*$, where
578: $\ell_*$, $u_*$ and $j_*$ may be selected arbitrarily and $k_*$
579: is chosen as the $\theta$-grid point in the $\rho=j_*\dr$ circle that
580: gives  the $\ell_*$-component value closest to $u_*$ in the initial approximation. Since
581: $U_{\ell_*}(j_*\dr,k_*\dt)$ is fixed, it is no longer an unknown,
582: and its place in the $\Real^{\Ntot}$-vector of unknowns is taken by
583: $\omega$, also to be found from \eq{SW-own}. In this way, the balance of
584: the unknowns and equations is preserved. As $\omega$ is present in all 
585: equations, 
586: the corresponding non-zero column of the linearization matrix 
587: destroys the bandedness of the matrix. This obstacle is 
588: overcome by employing the Sherman-Morrison formula
589: \cite{Numerical-Recipes} to find solutions of the corresponding linear
590: systems using only banded matrices.
591: % Newton's 
592: Newton iterations are performed until the residual in solution of the
593: discretized version of equation  \eq{SW-own} 
594: becomes sufficiently small.
595: 
596: The linearized problems \eq{right-evp} and \eq{left-evp} are
597: considered in the same domain with similar boundary conditions.
598: The critical eigenvalues and eigenvectors of the discretized operators $\L$ and $\Lp$ are computed
599: with the help of a complex shift and Cayley transform. 
600: 
601: For a matrix
602: $\mx{L}$, be it discretization of $\L$ or $\Lp$, the complex
603: shift is defined as
604: \[
605:   \mx{A} = \mx{L} + \i \kappa \mx{I}
606: \]
607: and the subsequent Cayley transform as 
608: \begin{equation}
609:   \mx{B} = (\xi\mx{I} +  \mx{A})^{-1} (\eta\mx{I} + \mx{A})       \label{mxB}
610: \end{equation}
611: where $\kappa$, $\xi$ and $\eta$ are real parameters and $\mx{I}$ is the
612: identity matrix. If $\lambda$, $\alpha$ and $\beta$ are eigenvalues of
613: $\mx{L}$, $\mx{A}$
614: and $\mx{B}$, respectively, this implies 
615: \[
616:   \alpha=\lambda+\i\kappa, \qquad \beta=\frac{\eta+\alpha}{\xi+\alpha}.
617: \]
618: The selected eigenvalues and eigenvectors of the thus constructed matrices
619: $\mx{B}$ are then found by the Arnoldi method, 
620: using ARPACK~\cite{ARPACK}.
621: 
622: 
623:  We have used $\xi=0$, $\eta=1$ and
624: $\kappa=0,\mp \omega$ when seeking, respectively, $\Tr{0,\pm1}$ and
625: $\RF{0,\mp1}$, where $\omega$ is the solution of the corresponding
626: nonlinear problem previously obtained. With this choice of $\xi$, $\eta$ and
627: $\kappa$, the numerical 
628: eigenvalues $\numlambda$ and $\nummu$ closest to the theoretical critical eigenvalues
629: \eq{lambdas} and \eq{mus} correspondingly, generate the largest
630: $|\beta|$. Hence, the Arnoldi method in each case is required to
631: obtain the eigenvalue with the largest absolute value.
632: 
633: To normalize the eigenvectors, we use the ``analytical'' Goldstone
634: modes $\anaTr{k}$, obtained by numerical differentiation of the
635: numerical spiral wave solution $\numU$, namely,
636: %
637: \begin{eqnarray*}
638: \anaTr{0} &=&  
639:   - \@_\theta \numU(\rho,\theta), \\
640: \anaTr{\pm1} &=& 
641:   -\frac12 \e^{\mp\i\theta} \left(
642:     \@_\rho\mp\i\rho^{-1}\@_\theta
643:   \right) \numU(\rho,\theta) ,
644: \end{eqnarray*}
645: %
646: where differentiation has been implemented using the same discretization 
647: schemes as used in calculations. 
648: 
649: First, the response functions $\numRF{k}$ computed by ARPACK are
650: normalized with respect to the ``analytical'' Goldstone
651: modes $\anaTr{k}$ so that
652: \[
653:   \inner{ \numRF{k} }{ \anaTr{k} }=1, \qquad k=0,\pm1,
654: \]
655: where numerical integration involved in $\inner{\cdot}{\cdot}$ has been
656: carried out using the trapezoidal rule. 
657: 
658: Then, the ``numerical'' Goldstone modes $\numTr{k}$ computed by ARPACK 
659: are normalized with respect to the normalized response functions so that
660: \[
661:   \inner{ \numRF{k} }{ \numTr{k} }=1, \qquad k=0,\pm1.
662: \]
663: 
664: Thus, we finally obtain 
665: \begin{itemize}
666: \item a numerical solution for the spiral wave problem \eq{SW-own} together with the angular velocity $\omega$,
667: \item ``analytical'' Goldstone modes $\anaTr{k}$, 
668: \item normalized ``numerical'' Goldstone modes $\numTr{k}$, and 
669: \item normalized response functions $\numRF{k}$.
670: \end{itemize}
671: 
672: \subsection{Analysis}
673: 
674: \label{sec:anal}
675: 
676: To validate the computed response functions, we have to
677: demonstrate convergence of the solution with respect to the numerical
678: approximation parameters such as the size of the medium $\rp$, and
679: the discretization steps $\dr$ and $\dt$.
680: 
681: First of all, we have to demonstrate convergence  of the computed eigenvalues
682: of $\numlambda_n$ and $\nummu_n$ to their theoretical values \eq{lambdas}
683: and \eq{mus}, taking for $\omega$ its numerical approximation $\numomega$ found
684: by numerical solving the discretized problem \eq{SW-own}. Since the
685: ``theoretical'' value for $\omega$ is not available, we can only check
686: convergence of $\numomega$ to some limit. 
687: 
688: The accuracy of the ``numerical'' Goldstone modes is quantified by the 
689: distance between the 
690: ``numerical'' and ``analytical'' Goldstone modes, in $L_2$ norm
691: \[
692:   \intdist{j} = \left( \int\limits_{\S}
693:     \left|\anaTr{j}(\r)-\numTr{j}(\r)\right|^2\,\d^2\r \right)^{1/2}
694: \]
695: as well as $C_0$ norm
696: \[
697:   \maxdist{j} = \max\limits_{\r\in\S} \left|\anaTr{j}(\r)-\numTr{j}(\r)\right|
698: \]
699: over a disk $\S$ of half the radius of the computational domain:
700: \[
701:   \S=\{ \r: |\r|\le\rp/2 \} .
702: \]
703: The smaller disk is used to exclude the effects of boundary
704: conditions. The issue is that the exact GM $\anaTr.$ do not satisfy Neumann
705: boundary conditions whereas $\numTr.$ do, hence there is an inevitable
706: deviation between them near $\rho=\rp$, which is an artefact of
707: restricting our problem to a finite domain, and is not indicative of
708: the accuracy of the computed $\numRF.$, which are expected to be exponentially
709: small near $\rho=\rp$. 
710:  
711: 
712: The accuracy of the computed response functions $\numRF.$ could be
713: tested directly in the same way as the accuracy of the computed
714: $\numomega$, \ie\ by the numerical convergence to some limit.
715: This is however, difficult to implement for the numerical
716: solutions obtained on different grids. Nevertheless, we are able to 
717: examine the convergence in $\dr$ where coarser grids are subgrids of the
718: finer grids by restricting the fine-grid solutions to the coarse grid, 
719: without the need for any interpolation.
720: Specifically, we calculate 
721: \[
722:   \interr{j} = \left( \int\limits_{\B}
723:     \left|\numRF{j}_{\dr}(\r)-\numRF{j}_{\dr_*}(\r)\right|^2\,\d^2\r \right)^{1/2}
724: \]
725: and
726: \[
727:   \maxerr{j} = \max\limits_{\r\in\B} \left|\numRF{j}_{\dr}(\r)-\numRF{j}_{\dr_*}(\r)\right|
728: \]
729: over the whole computational domain
730: \[
731:   \B=\{ \r: |\r|\le\rp \} ,
732: \]
733: where $\numRF{j}_{\dr}(\r)$ are the numerical response functions
734: calculated at the radius step $\dr$ which is an integer multiple of the
735: minimal radius step $\dr_*$, and the finest numerical response functions
736: $\numRF{j}_{\dr_*}(\r)$ have been restricted to the coarser grid of
737: $\numRF{j}_{\dr}(\r)$ of the solution to which they are compared, so the numerical
738: integration is done over the coarser grid. Note that in the series
739: with varying $\rp$ and fixed $\dt$ and $\dr$, the coarser grids are
740: also subgrids of the finer grids, but as the pinning point is
741: defined via $\rp$, solutions at different $\rp$ are again not
742: directly comparable to each other so this series is not used 
743: in this comparison.
744: 
745: 
746: We also assess accuracy 
747: indirectly via the bi-orthogonality between the response functions 
748: and the Goldstone modes required by \eq{norm}. 
749: Specifically, we examine
750: the orthogonality of the RFs 
751: to the ``analytical'' GMs, 
752: quantified by
753: \begin{equation}
754:   \Ortana=\sum\limits_{j=0,\pm1}\sum\limits_{k=0,\pm1}
755:   \left|\inner{ \numRF{j} }{ \anaTr{k} } - \delta_{j,k}\right|^2
756:                                                \label{Ortana}
757: \end{equation}
758: and orthogonality of the RFs to the ``numerical'' GMs quantified by
759: \[
760:   \Ortnum=\sum\limits_{j=0,\pm1}\sum\limits_{k=0,\pm1}
761:   \left|\inner{ \numRF{j} }{ \numTr{k} } - \delta_{j,k}\right|^2.
762: \]
763: Note, that by construction the diagonal elements of both the ``numerical'' and
764: ``analytical'' bi-orthogonality matrices here are all equal to 1 up to
765: round-off errors.
766: 
767: The measures $\Ortana$ and $\Ortnum$ require some discussion.
768: The bi-orthogonality should be exact for exact RFs and GMs.
769: However, what we calculate are approximations of these functions, subject to
770: discretization in $\rho$ and $\theta$ and restriction to a finite domain
771: $\rho\le\rp$. The bi-orthogonality of numerical solutions is therefore not
772: exact and its deviation from the ideal is an indication of the accuracy of
773: calculation, and its convergence in $\dr$, $\dt$ and $\rp$ is an indication,
774: albeit indirect, of the accuracy of the solutions.  
775: 
776: In more detail, if the the matrices representing discretization of
777: $\L$ and $\Lp$ were transposes of one another, then their
778: eigenvectors corresponding to different eigenvalues would be exactly
779: orthogonal in $l_2$, and so a measure of their orthogonality would
780: not depend on the spatial discretization but only on the accuracy of
781: the calculation of the eigenvectors by ARPACK. However, $\L$ and
782: $\Lp$ are conjugate with respect to the scalar product which is
783: approximated by a discrete inner product with a weight, hence the
784: matrices of $\L$ and $\Lp$ are not transposed. Moreover, because of
785: the approximation used for these operators (\eg\ high-order
786: approximation in $\dt$ vs second-order approximation in $\dr$), the
787: corresponding matrices are not adjoint of each other with respect to
788: the weighted $l_2$ either. So, $\Ortnum$ provides a measure of the
789: consistency of these matrix representations together with the
790: accuracy with which the eigenvectors are computed with ARPACK.
791: 
792: Moreover, apart from the question of accuracy of finding the eigenvectors
793: of the discretized operators and accuracy of finding the eigenfunctions 
794: of the original continuous operators, there remains a question of 
795: whether the found eigenvectors and eigenfunctions are the ones
796: that we need, that correspond to $0$ and $\pm i\omega$, 
797: rather than eigenfunctions corresponding to eigenvalues which happened to be
798: close to $0$ and $\pm i\omega$~\footnote{
799:   Close neighbours of the translational eigenmodes are 
800:   always a possibility in a large enough disk, 
801:   see \cite{Wheeler-Barkley-2006}.
802: }. For the GMs, the answer to this question is ensured by
803: checking the distance $\intdist{j}$; however, this answer is not absolute as the
804:  comparison
805: is made only over part of the disk, for reasons discussed above. 
806: We note, however, that the $\Lp$ eigenfunctions corresponding to the eigenvalues close to but different from $0,\pm i\omega$, are orthogonal to the GMs and for them $\Ortana$ would be not small~\footnote{
807:   Equation \eq{Ortana} gives $\Ortana=3$ if all nine scalar products vanish;
808:   however in reality the scalar products of 
809:   respective GMs and RFs are used for normalization, so in the case
810:   of wrong RFs, all scalar products would be divided by small
811:   numbers which may result in rather large values of $\Ortana$. 
812: }.
813: Since $\Ortana$ is defined in terms of scalar products with the mode determined
814: directly from the underlying spiral wave, its smallness
815: provides the additional assurance that the adjoint eigenfunctions are indeed the RFs
816: that we are after, not just some adjoint eigenfunctions. 
817: 
818: 
819: \section{Results}
820: 
821: \subsection{General}
822: 
823: We have tested our method for computing the response functions in the case of
824: the FitzHugh-Nagumo model, $\ell=2$,
825: % 
826: \begin{eqnarray*}
827: f_1 & = & \pareps^{-1}(u_1-u_1^3/3-u_2), \nonumber \\
828: f_2 & = & \pareps(u_1-\para u_2 + \parb),
829: \end{eqnarray*}
830: % 
831: $\D=\Mx{1&0\\0&0}$, with parameters $\para=0.5$,
832: $\parb=0.68$, $\pareps=0.3$. 
833: For pinning, we have used $\ell_*=2$, $u_*=0.1$ and $j_*=\Nr/2$. 
834: % Newton's
835: Newton
836: iterations have been performed until the Euclidean ($l_2$) norm of the
837: residual in the discretized nonlinear equation falls below
838: $10^{-8}$. For comparison, we have also run cases, discussed later in 
839: \fig{comp}, in which iterations continue until the
840: norm of the residual no longer decreases (typically such norms were
841: below $10^{-9}$ down to $10^{-13}$). 
842: The tolerance in ARPACK's
843: routines {\ttfamily znaupd} and {\ttfamily zneupd} has been set to the
844: default ``machine epsilon''. For the Krylov subspace dimensionality we
845: have tried 3 and 10, with no perceptible difference in either the
846: numerical results.
847: 
848: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
849: \dblfigure{
850:   \includegraphics{fig1.eps}
851: }{
852:   Solutions of the nonlinear problem \eq{SW-own} and the linearized problem
853:   \eqtwo(right-evp,L), \ie\ the Goldstone modes, at the ``best'' parameters,
854:   $\rp=25$, $\Nr=1280$, $\Nt=64$, 
855:   as density plots. 
856:   Numbers under the density plots are their amplitudes $\Ampl$: 
857:   white of the plot corresponds to the value $\Ampl$ and black corresponds
858:   to the value $-\Ampl$ of the designated field.
859:   Upper row: 1st components, lower row: 2d components.
860: }{gmpics}
861: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
862: 
863: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
864: \dblfigure{
865:   \includegraphics{fig2.eps}
866: }{
867:   Same visualization as in \fig{gmpics}, for the adjoint linearized problem
868:   \eqtwo(left-evp,Lp), \ie\ the response functions. 
869: }{rfpics}
870: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
871: 
872: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
873: \dblfigure{
874:   \includegraphics{fig3.eps}
875: }{
876:   Radial dependence of the angle-averaged solutions for the spiral
877: wave (a), Goldstone modes (b) and response functions (c).  In (c), the
878: dependence of $\Delta\omega(\rp)=\numomega(\rp)-\numomega(25)$ 
879: is shown for comparison, where
880: $\numomega(\rp)$ is the numerically
881: found spiral angular velocity in the disk of given radius $\rp$. 
882: }{decay}
883: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
884: 
885: 
886: Before discussing the performance of our numerical techniques, we briefly 
887: present typical solutions.
888: %
889: \Figs{gmpics} and \ref{rfpics} 
890: illustrate the spiral wave solution and the GMs and
891: RFs for $\rp=25$, $\Nr=1280$ and $\Nt=64$. This solution is taken as 
892: the best achievable given memory restrictions (4Gb
893: of real memory).
894: The angular velocity for it was found to be 
895: $\numomega\approx0.5819341748776017$.
896: For the GMs and RFs, we show the $n=0$ and $n=1$ modes only, since the 
897: calculated $n=-1$ modes are almost exactly the 
898: complex conjugates of the $n=1$ modes, which of course they should be.
899: 
900: One can see that the GMs $\numTr.$ are indeed proportional to
901: corresponding derivatives of the spiral wave solution $\numU$, and
902: that the RFs $\numRF.$ are localized in a small region of the spiral
903: tip and are indistinguishable from zero outside that region.
904: 
905: The character of the RFs' decay with distance is illustrated in more
906: detail in \fig{decay}. We plot the
907: angle-averaged values of the solutions, defined as
908: \[
909:   \Avg{X}{n}{i}(\rho) = \frac{1}{2\pi} \oint \hat{X}_i^{(n)}(\rho,\theta)
910:   \,\d\theta,
911: \]
912: for $X=U,V$ and $W$. 
913: Note the difference in the behavior of $\Avg{U}{n}{i}$ and 
914: $\Avg{V}{n}{i}$ on one hand and $\Avg{W}{n}{i}$ on the other hand.
915: In the
916: semilogarithmic (linear for horizontal axis, logarithmic
917: for vertical axis)
918: coordinates of \fig{decay}(c) the graphs of
919: $\Avg{W}{n}{i}(\rho)$ are straight for a large range of $\rho$,
920: not too close to 0 or $\rp=25$, and for several decades of magnitude of
921: $\Avg{W}{n}{i}$. 
922: % This is an evidence of the expected exponentialcharacter of decay.
923: This shows clearly the expected exponential localization of the RFs.
924: %
925: For comparison, we also show
926: convergence of $\numomega=\numomega(\rp)$ in a disk as a function of the
927: disk radius $\rp$.  Theory \cite{Hagan-1982, Biktashev-1989-inNW, Biktasheva-Biktashev-2001, Sandstede-private} predicts that the
928: $\Avg{W}{n}{i}(\rho)$ and $\Delta\omega(\rp)=\numomega(\rp)-\numomega(\infty)$
929: dependencies should both be decaying 
930: exponentials with the same characteristic exponent; this
931: agrees well with the numerical results shown in \fig{decay}(c).
932: 
933: Sandstede and Scheel~\cite{%
934:   Sandstede-Scheel-PhysRevE-2000,%
935:   Sandstede-Scheel-PhysLett-2001%
936: } have computed exponential decay/increase rates of eigenfunctions of
937: periodic wavetrains in one spatial dimension. A similar technique
938: should, in principle, also work for the adjoint eigenfunctions. Knowning the
939: asymptotic wavelength of the spiral wave, this can be used
940: to predict the exponential decay rates of the RFs of spiral
941: waves. As can be seen from the results of Wheeler and Barkley~\cite{Wheeler-Barkley-2006},
942: although such correspondence between 1D and 2D calculations 
943: can be established, the accuracy of decay rate estimates for two-dimensional
944: eigenfunctions achieved in this way is insufficient for a meaningful estimate of the
945: accuracy of those eigenfunctions. 
946: 
947: 
948: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
949: \dblfigure{
950:   \includegraphics{fig4.eps}
951: }{
952:   Convergence in numerical parameters: of deviation of the numerical
953:   eigenvalues from theoretical (upper row), of $L_2$ distance between numerical
954:   and theoretical eigenfunctions (second row) and of orthogonality,
955:   \ie\ Frobenius norm of the difference of the matrix
956:   of scalar products of eigenfunctions and adjoint eigenfunctions from
957:   the unity matrix (third row), all in logarithmic scales,
958:   as dependencies of disk radius (first
959:   and second columns, linear scale), radius
960:   discretization step (third column, logarithmic scale) and
961:   polar angle discretization step
962:   (fourth column, logarithmic scale).
963:   In the first column, $\rp$ is changed while
964:   the values of $\dr$ and
965:   $R\dt$ are kept constant.  
966:   In the second column, $\rp$ is changed while $\dr$
967:   and $\dt$ are kept constant. 
968: }{convergence}
969: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
970: 
971: \subsection{Convergence}
972: 
973: We now turn to the main results of our study. 
974: Convergence of the method has been tested by changing one of the three
975: numerical approximation
976: parameters $\rp$, $\Nr$ and $\Nt$ while keeping the other two at the fixed
977: values set by the ``best example''. More specifically, while changing
978: $\rp$, we consider two variants: one with fixed $\Nt$, and one with changing
979: $\Nt$ so that the combination $\rp\dt$, which is the size of the
980: outermost computational cells in the angular direction, remains
981: constant. 
982: 
983: \Fig{convergence} illustrates the results of the study, where the four
984: columns correspond to different series of calculations, and the three
985: rows correspond to the three different methods of assessing the
986: accuracy: closeness of the eigenvalues to the theoretical values,
987: distance between ``numerical'' and ``analytical'' GMs and
988: orthogonality between non-dual RFs and GMs. The scales of $\Delta \rho$, $\dt$
989: and the error estimates are logarithmic, and the scales of $\rp$ are
990: linear. Here shown is the distance between the ``numerical'' and ``analytical'' 
991: Goldstone modes in $L_2$ norm, the distance in $C_0$ norm looks similar.
992: 
993: A typical feature on many of the curves is a ``knee''-shape, when the
994: measure of the error decreases as $\rp$ grows or $\dt$ or $\Delta \rho$
995: decrease, but only until a certain point, beyond which it 
996: reaches a plateau. This behavior is expected and explicable. 
997: The calculation error is affected by many factors, and if the factor
998: varied in a particular series becomes negligible, then the error
999: remains at a constant level determined by fixed values of other factors.
1000: 
1001: The position of the ``knees'' on the curves indicates that the
1002: accuracy of the rotational ($n=0$) modes would be improved if $\dt$ were
1003: further decreased (there are no knees on the curves corresponding to the rotational modes, red online, in the fourth,
1004: \ie\ rightmost column), whereas the limiting parameter for the
1005: translational ($n=1$) modes is $\dr$ (there are no knees on the curves corresponding to the tanslational modes, blue online, in
1006: the third column). The analysis of the first two columns is more
1007: complicated. The errors estimates at the maximal $\rp$ are similar in
1008: both columns as they correspond to the same ``best'' spiral. These
1009: limit values are achieved, \ie\ plateaux are observed, at much smaller
1010: $\rp$ values if $\dt=\const$, than if $\rp\dt=\const$.  This is
1011: because reduction of $\rp$ at fixed $\dt$ produces an additional
1012: improvement of approximation due to the angular discretization.  When
1013: $\rp\dt$ is kept fixed, as in the first column, the dependence of the
1014: solution on the disk radius is without this extra benefit.
1015: 
1016: The rates of convergence with respect to parameters can be assessed by the
1017: slopes of the curves above the knees before they plateau. In some cases the
1018: data is somewhat irregular, primarily at parameters corresponding to lower
1019: values of error estimates. This is not unexpected and we attribute it to
1020: incomplete convergence of the iterative procedures (see below).  On the whole,
1021: the slopes can be determined clearly from these plots.
1022: 
1023: The constant slope in the first (leftmost) and the second columns
1024: corresponds to the exponential convergence with $\rp$. The constant
1025: slope in the third column corresponds to power-law convergence, and
1026: the typical slope is 2. This is well seen on the curves for
1027: translational modes, blue online, and not well on the curves for rotational
1028: modes, red online, which are very small anyway. 
1029: Slope 2 in the third column is to be
1030: expected as our discretization is second-order in $\dr$ in all
1031: cases. The curves in the fourth (rightmost) column are convex, which
1032: is consistent with the fact that the order of approximation is $\Nt$,
1033: which varies along the curve as $\dt$ varies, since $\Nt=2\pi/\dt$, so
1034: the slope is bigger for smaller $\dt$. 
1035: %
1036: In other words, the high order of the Fornberg approximation of the
1037: $\theta$ derivatives implies the convergence in $\dt$ is faster than
1038: any fixed power.
1039: 
1040: 
1041: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1042: \dblfigure{
1043:   \includegraphics{fig5.eps}
1044: }{
1045:   (a,b)
1046:   Effect of the accuracy of the unperturbed spiral wave solution on the convergence:
1047:   (a) Newton-iteration tolerance $10^{-8}$.
1048:   (b) Newton iterations until the norm of the residual
1049:   stopped decreasing. 
1050:   (c) Convergence of the response functions in $\dr$. 
1051: }{comp}
1052: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1053: 
1054: 
1055: The irregular shape of some of the curves in
1056: \fig{convergence} at very low values of the error estimates is related to
1057: the accuracy of finding the spiral solution and is ulitmately affected by the
1058: precision of floating point computations. 
1059: Note that all calculations in \fig{convergence}
1060: have been performed with %Newton's 
1061: a tolerance of $10^{-8}$ for Newton iterations of the spiral wave and some of
1062: the curves fall as low as $10^{-15}$ \ie\ close to machine
1063: epsilon. A change in the % Newton's iteration 
1064: tolerance of the Newton iteration reduces irregularities in the curves
1065: at low values, as shown in \fig{comp}(a,b). 
1066: 
1067: 
1068: Finally, \fig{comp}(c) illustrates convergence of
1069: numerical RFs $\numRF{0,1}$ as $\dr\to0$, calculated as the
1070: $L_2$-distance $\interr{0,1}$ between the solutions at a given
1071: resultion $\dr$ and the ``best'' solution calculated at the smallest
1072: $\dr_*=25/1280$.  As explained in the Sec.~\ref{sec:anal}, this
1073: comparsion has been restricted to the series of calculations with varying
1074: $\dr$, where grids at lower resolutions were subgrids of those with
1075: higher resolutions. The graphs of $C_0$ distances $\maxerr{0,1}$
1076: looked similar and are not shown here.  
1077: 
1078: 
1079: \section{Discussion}
1080: 
1081: 
1082: The main result of this paper is
1083: a general, robust method for obtaining response functions
1084: for rigidly rotating spiral waves in excitable media
1085: with required accuracy.
1086: 
1087: We have tested the method 
1088: on the FitzHugh-Nagumo model
1089: and we have studied the convergence of 
1090: spiral wave solutions and eigenfunctions, 
1091: both the Goldstone modes and the
1092: response functions,
1093: with respect to the numerical approximation parameters $\rp$, $\Nr$
1094: and $\Nt$. The rates of convergence are found to agree with the order of
1095: approximation and indicate the accuracy with which solutions can be found for
1096: particular numerical parameters.
1097: 
1098: 
1099: The slowest (second-order) convergence is, as expected, in the
1100: parameter $\Nr$. Thus in a typical situation, an improvement of
1101: accuracy requires, other things being equal, an increase of $\Nr$,
1102: with associated increase in memory and time demands. Thus, the
1103: most promising avenue of further development of the method is via increase of
1104: the approximation order of the radial derivatives.
1105: This is, of course, subject to usual caveat that the degree of approximation
1106: should be consistent with the actual smoothness of the solutions.
1107: 
1108: 
1109: The method used here to solve the eigenvalue problems for operators $\mx{L}$
1110: relies on successive application of 
1111: % linear 
1112: transformations of $\mx{L}$ applied to 
1113: a sequence of vectors, alternating with Gram-Schmidt
1114: orthogonalization. These are typical ideas, also used
1115: in \cite{Henry-Hakim-2002,Biktasheva-etal-2006}. The difference is
1116: that in \cite{Henry-Hakim-2002,Biktasheva-etal-2006}, the 
1117: linear transformations were polynomial functions of 
1118: $\mx{L}$ whereas we use rational functions of
1119: $\mx{L}$.  
1120: %
1121: The polynomial iterations used in
1122: \cite{Henry-Hakim-2002,Biktasheva-etal-2006} were in fact equivalent to
1123: solving a Cauchy problem for equation
1124: $\d{\u}/\d{t}=\mx{L}\u$ by the explicit Euler method.  Therefore, those
1125: methods require a large number of iterations, and
1126: convergence speed of the iterations depends on the smallness of the
1127: absolute difference of the real parts of the eigenvalues of interest
1128: compared to those of other eigenvalues. 
1129: One requires at least $\O(10^5)$ and typically 
1130: $\O(10^6)$ sparse matrix-vector
1131: multiplications to achieve the desired solutions to the eigenvalue problem using
1132: such an approach. 
1133: 
1134: In contrast, with the
1135: complex shift and inversion of $\mx{L}$ used in this paper, the
1136: convergence speed of the iterations depends on the smallness of the
1137: distance of the eigenvalues from their theoretical values used in
1138: the complex shift, compared to the distance to other eigenvalues. Hence
1139: the number of iterations required is very small, typically $\O(10)$.
1140: More specifically, with Krylov subspace dimensionality 3, the number of matrix
1141: multiplications with matrix $\mx{B}$ of \eq{mxB} 
1142: did not exceed 7 per one eigenpair; with Krylov subspace
1143: dimensionality 10, this number rose to 10.  
1144: The price to pay for this acceleration is the necessity to solve 
1145: large systems of linear equations.
1146: However, the key observation is that since the linear system is fixed, 
1147: it needs to be factorized only once, for a given complex shift, and used for 
1148: all iterations. Multiplication by matrix $\mx{B}$ is achieved with only
1149: inexpensive back/forward solves. 
1150: %
1151: Moreover, due to the way we ordered the unknowns in the discretized problem,
1152: the sparcity of matrix $\mx{B}$ does not depend on the order of
1153: approximation of $\theta$-derivatives. Hence, we are able to employ high-order
1154: approximations requiring far fewer points in the $\theta$ direction
1155: for the same accuracy as the second-order finite difference
1156: discretization used in \cite{Henry-Hakim-2002}, thereby further improving
1157: the efficiency of our method. 
1158: 
1159: Discounting the factorization step, each
1160: iteration, which involves multiplication by $\mx{B}$, 
1161: is comparable to multiplications by $\mx{L}$. 
1162: In practice we find that the factorization itself does not require
1163: more than the equivalent of four to six actions of $\mx{B}$.
1164: %
1165: On a MacPro with 3\,GHz Intel processor, 
1166: the factorization step
1167: takes \eg\ about 7.5\,sec for the grid $\Nr=1280$, $\Nt=64$, and 
1168: 0.67\,sec for the grid $\Nr=640$, $\Nt=32$; the computation times 
1169: per $\mx{B}$-multiplication were 1.23~sec and 0.17\,sec
1170: respectively.  
1171: 
1172: The comparison of our present method
1173: with \cite{Biktasheva-etal-2006} is unequivocal: matrix inverses
1174: were not used there, and 
1175: it was admitted already in \cite{Biktasheva-etal-2006}
1176: that the resulting accuracy of solutions was severely limited.
1177: %
1178: While direct accuracy and timing comparisons with \cite{Henry-Hakim-2002}
1179: would be most convincing, that code is not publicly available. However, for
1180: reasons already noted, on any given polar grid, the method we report is more
1181: accurate due to the angular discretization and considerably faster in
1182: floating-point operations.
1183: 
1184: 
1185: 
1186: The computed response functions are localized in the vicinity of the
1187: spiral wave tip and exponentially decay with distance from it. This
1188: localization ensures convergence of the convolution integral in
1189: \eq{forces} in an unbounded domain.
1190: 
1191: The eigenvectors of the linearized operator, \ie\ Goldstone modes and
1192: of its adjoint, \ie\ the response functions have been computed
1193: using the same technique, so the qualitatively different behavior of
1194: these solutions at large $\rho$ is not a numerical artefact, as it was
1195: not in any way assumed in the numerical method.
1196: 
1197: Although the method has been used here to compute the response functions in
1198: the FitzHugh-Nagumo model, none of the details of the method depends
1199: on any specifics of the particular
1200: reaction kinetics and should be widely applicable to the computation of
1201: response functions of rigidly rotating waves in any other model of excitable
1202: tissue, as long as its right-hand sides are continuously differentiable so the linarized theory is applicable.
1203: Moreover, the method can also be extended in a straightforward way to include
1204: additional effects, such as the effect of uniform twist along scroll waves
1205: with linear filaments in three
1206: dimensions~\cite{%
1207:   Biktashev-1989a,%
1208:   Margerit-Barkley-2001,%
1209:   Henry-Hakim-2002%
1210: }.
1211: 
1212: \section*{Acknowledgement}
1213: This study has been supported in part by EPSRC grants EP/D074789/1 and EP/D074746/1. 
1214: 
1215: 
1216: %\bibliography{rf}
1217: \begin{thebibliography}{48}
1218: \expandafter\ifx\csname natexlab\endcsname\relax\def\natexlab#1{#1}\fi
1219: \expandafter\ifx\csname bibnamefont\endcsname\relax
1220:   \def\bibnamefont#1{#1}\fi
1221: \expandafter\ifx\csname bibfnamefont\endcsname\relax
1222:   \def\bibfnamefont#1{#1}\fi
1223: \expandafter\ifx\csname citenamefont\endcsname\relax
1224:   \def\citenamefont#1{#1}\fi
1225: \expandafter\ifx\csname url\endcsname\relax
1226:   \def\url#1{\texttt{#1}}\fi
1227: \expandafter\ifx\csname urlprefix\endcsname\relax\def\urlprefix{URL }\fi
1228: \providecommand{\bibinfo}[2]{#2}
1229: \providecommand{\eprint}[2][]{\url{#2}}
1230: 
1231: \bibitem[{\citenamefont{Frisch et~al.}(1994)\citenamefont{Frisch, Rica,
1232:   Coullet, and Gilli}}]{Frisch-etal-1994}
1233: \bibinfo{author}{\bibfnamefont{T.}~\bibnamefont{Frisch}},
1234:   \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Rica}},
1235:   \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Coullet}}, \bibnamefont{and}
1236:   \bibinfo{author}{\bibfnamefont{J.~M.} \bibnamefont{Gilli}},
1237:   \bibinfo{journal}{Phys. Rev. Lett.} \textbf{\bibinfo{volume}{72}},
1238:   \bibinfo{pages}{1471} (\bibinfo{year}{1994}).
1239: 
1240: \bibitem[{\citenamefont{Yu et~al.}(1999)\citenamefont{Yu, Lu, and
1241:   Harrison}}]{Yu-etal-1999}
1242: \bibinfo{author}{\bibfnamefont{D.~J.} \bibnamefont{Yu}},
1243:   \bibinfo{author}{\bibfnamefont{W.~P.} \bibnamefont{Lu}}, \bibnamefont{and}
1244:   \bibinfo{author}{\bibfnamefont{R.~G.} \bibnamefont{Harrison}},
1245:   \bibinfo{journal}{Journal of Optics B --- Quantum and Semiclassical Optics}
1246:   \textbf{\bibinfo{volume}{1}}, \bibinfo{pages}{25} (\bibinfo{year}{1999}).
1247: 
1248: \bibitem[{\citenamefont{Madore and Freedman}(1987)}]{Madore-Freedman-1987}
1249: \bibinfo{author}{\bibfnamefont{B.~F.} \bibnamefont{Madore}} \bibnamefont{and}
1250:   \bibinfo{author}{\bibfnamefont{W.~L.} \bibnamefont{Freedman}},
1251:   \bibinfo{journal}{Am. Sci.} \textbf{\bibinfo{volume}{75}},
1252:   \bibinfo{pages}{252} (\bibinfo{year}{1987}).
1253: 
1254: \bibitem[{\citenamefont{Schulman and Seiden}(1986)}]{Schulman-Seiden-1986}
1255: \bibinfo{author}{\bibfnamefont{L.~S.} \bibnamefont{Schulman}} \bibnamefont{and}
1256:   \bibinfo{author}{\bibfnamefont{P.~E.} \bibnamefont{Seiden}},
1257:   \bibinfo{journal}{Science} \textbf{\bibinfo{volume}{233}},
1258:   \bibinfo{pages}{425} (\bibinfo{year}{1986}).
1259: 
1260: \bibitem[{\citenamefont{Zhabotinsky and
1261:   Zaikin}(1971)}]{Zhabotinsky-Zaikin-1971}
1262: \bibinfo{author}{\bibfnamefont{A.~M.} \bibnamefont{Zhabotinsky}}
1263:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{A.~N.}
1264:   \bibnamefont{Zaikin}}, in \emph{\bibinfo{booktitle}{Oscillatory processes in
1265:   biological and chemical systems}}, edited by
1266:   \bibinfo{editor}{\bibfnamefont{E.~E.} \bibnamefont{Selkov}},
1267:   \bibinfo{editor}{\bibfnamefont{A.~A.} \bibnamefont{Zhabotinsky}},
1268:   \bibnamefont{and} \bibinfo{editor}{\bibfnamefont{S.~E.} \bibnamefont{Shnol}}
1269:   (\bibinfo{publisher}{Nauka}, \bibinfo{address}{Pushchino},
1270:   \bibinfo{year}{1971}), p. \bibinfo{pages}{279}, \bibinfo{note}{in Russian}.
1271: 
1272: \bibitem[{\citenamefont{Jakubith et~al.}(1990)\citenamefont{Jakubith,
1273:   Rotermund, Engel, von Oertzen, and Ertl}}]{Jakubith-etal-1990}
1274: \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Jakubith}},
1275:   \bibinfo{author}{\bibfnamefont{H.~H.} \bibnamefont{Rotermund}},
1276:   \bibinfo{author}{\bibfnamefont{W.}~\bibnamefont{Engel}},
1277:   \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{von Oertzen}},
1278:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Ertl}},
1279:   \bibinfo{journal}{Phys. Rev. Lett.} \textbf{\bibinfo{volume}{65}},
1280:   \bibinfo{pages}{3013} (\bibinfo{year}{1990}).
1281: 
1282: \bibitem[{\citenamefont{Agladze and Steinbock}(2000)}]{Agladze-Steinbock-2000}
1283: \bibinfo{author}{\bibfnamefont{K.}~\bibnamefont{Agladze}} \bibnamefont{and}
1284:   \bibinfo{author}{\bibfnamefont{O.}~\bibnamefont{Steinbock}},
1285:   \bibinfo{journal}{J.Phys.Chem. A} \textbf{\bibinfo{volume}{104 (44)}},
1286:   \bibinfo{pages}{9816} (\bibinfo{year}{2000}).
1287: 
1288: \bibitem[{\citenamefont{Allessie et~al.}(1973)\citenamefont{Allessie, Bonk, and
1289:   Schopman}}]{Allessie-etal-1973}
1290: \bibinfo{author}{\bibfnamefont{M.~A.} \bibnamefont{Allessie}},
1291:   \bibinfo{author}{\bibfnamefont{F.~I.~M.} \bibnamefont{Bonk}},
1292:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{Schopman}},
1293:   \bibinfo{journal}{Circ. Res.} \textbf{\bibinfo{volume}{33}},
1294:   \bibinfo{pages}{54} (\bibinfo{year}{1973}).
1295: 
1296: \bibitem[{\citenamefont{Gorelova and Bures}(1983)}]{Gorelova-Bures-1983}
1297: \bibinfo{author}{\bibfnamefont{N.~A.} \bibnamefont{Gorelova}} \bibnamefont{and}
1298:   \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{Bures}}, \bibinfo{journal}{J.
1299:   Neurobiol.} \textbf{\bibinfo{volume}{14}}, \bibinfo{pages}{353}
1300:   (\bibinfo{year}{1983}).
1301: 
1302: \bibitem[{\citenamefont{Alcantara and Monk}(1974)}]{Alcantara-Monk-1974}
1303: \bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{Alcantara}} \bibnamefont{and}
1304:   \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Monk}}, \bibinfo{journal}{J.
1305:   Gen. Microbiol.} \textbf{\bibinfo{volume}{85}}, \bibinfo{pages}{321}
1306:   (\bibinfo{year}{1974}).
1307: 
1308: \bibitem[{\citenamefont{Lechleiter et~al.}(1991)\citenamefont{Lechleiter,
1309:   Girard, Peralta, and Clapham}}]{Lechleiter-etal-1991}
1310: \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{Lechleiter}},
1311:   \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Girard}},
1312:   \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{Peralta}}, \bibnamefont{and}
1313:   \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{Clapham}},
1314:   \bibinfo{journal}{Science} \textbf{\bibinfo{volume}{252}}
1315:   (\bibinfo{year}{1991}).
1316: 
1317: \bibitem[{\citenamefont{Carey et~al.}(1978)\citenamefont{Carey, Giles, Jr., and
1318:   Mclean}}]{Carey-etal-1978}
1319: \bibinfo{author}{\bibfnamefont{A.~B.} \bibnamefont{Carey}},
1320:   \bibinfo{author}{\bibfnamefont{R.~H.} \bibnamefont{Giles}},
1321:   \bibinfo{author}{\bibnamefont{Jr.}}, \bibnamefont{and}
1322:   \bibinfo{author}{\bibfnamefont{R.~G.} \bibnamefont{Mclean}},
1323:   \bibinfo{journal}{Am. J. Trop. Med. Hyg.} \textbf{\bibinfo{volume}{27}},
1324:   \bibinfo{pages}{573} (\bibinfo{year}{1978}).
1325: 
1326: \bibitem[{\citenamefont{Murray et~al.}(1986)\citenamefont{Murray, Stanley, and
1327:   Brown}}]{Murray-etal-1986}
1328: \bibinfo{author}{\bibfnamefont{J.~D.} \bibnamefont{Murray}},
1329:   \bibinfo{author}{\bibfnamefont{E.~A.} \bibnamefont{Stanley}},
1330:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{D.~L.} \bibnamefont{Brown}},
1331:   \bibinfo{journal}{Proc. R. Soc. Lond. ser. B} \textbf{\bibinfo{volume}{229}},
1332:   \bibinfo{pages}{111} (\bibinfo{year}{1986}).
1333: 
1334: \bibitem[{\citenamefont{Biktashev}(1989{\natexlab{a}})}]{Biktashev-1989}
1335: \bibinfo{author}{\bibfnamefont{V.~N.} \bibnamefont{Biktashev}}, Ph.D. thesis,
1336:   \bibinfo{school}{Moscow Institute of Physics and Technology}
1337:   (\bibinfo{year}{1989}{\natexlab{a}}).
1338: 
1339: \bibitem[{\citenamefont{Biktashev and Holden}(1995)}]{Biktashev-Holden-1995}
1340: \bibinfo{author}{\bibfnamefont{V.}~\bibnamefont{Biktashev}} \bibnamefont{and}
1341:   \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Holden}},
1342:   \bibinfo{journal}{Chaos, Solitons and Fractals} \textbf{\bibinfo{volume}{5}},
1343:   \bibinfo{pages}{575} (\bibinfo{year}{1995}).
1344: 
1345: \bibitem[{\citenamefont{Barkley}(1992)}]{Barkley-1992}
1346: \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{Barkley}},
1347:   \bibinfo{journal}{Phys. Rev. Lett.} \textbf{\bibinfo{volume}{68}},
1348:   \bibinfo{pages}{2090} (\bibinfo{year}{1992}).
1349: 
1350: \bibitem[{\citenamefont{Biktashev}(1989{\natexlab{b}})}]{Biktashev-1989a}
1351: \bibinfo{author}{\bibfnamefont{V.~N.} \bibnamefont{Biktashev}},
1352:   \bibinfo{journal}{Physica D} \textbf{\bibinfo{volume}{36}},
1353:   \bibinfo{pages}{167} (\bibinfo{year}{1989}{\natexlab{b}}).
1354: 
1355: \bibitem[{\citenamefont{Sandstede and
1356:   Scheel}(2000{\natexlab{a}})}]{Sandstede-Scheel-PhysD-2000}
1357: \bibinfo{author}{\bibfnamefont{B.}~\bibnamefont{Sandstede}} \bibnamefont{and}
1358:   \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Scheel}},
1359:   \bibinfo{journal}{Physica D} pp. \bibinfo{pages}{233--277}
1360:   (\bibinfo{year}{2000}{\natexlab{a}}).
1361: 
1362: \bibitem[{\citenamefont{Sandstede and
1363:   Scheel}(2000{\natexlab{b}})}]{Sandstede-Scheel-PhysRevE-2000}
1364: \bibinfo{author}{\bibfnamefont{B.}~\bibnamefont{Sandstede}} \bibnamefont{and}
1365:   \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Scheel}},
1366:   \bibinfo{journal}{Phys. Rev. E} pp. \bibinfo{pages}{7708--7714}
1367:   (\bibinfo{year}{2000}{\natexlab{b}}).
1368: 
1369: \bibitem[{\citenamefont{Sandstede and
1370:   Scheel}(2001)}]{Sandstede-Scheel-PhysLett-2001}
1371: \bibinfo{author}{\bibfnamefont{B.}~\bibnamefont{Sandstede}} \bibnamefont{and}
1372:   \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Scheel}},
1373:   \bibinfo{journal}{Phys. Rev. Lett.} \textbf{\bibinfo{volume}{86}},
1374:   \bibinfo{pages}{171} (\bibinfo{year}{2001}).
1375: 
1376: \bibitem[{\citenamefont{Wheeler and Barkley}(2006)}]{Wheeler-Barkley-2006}
1377: \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Wheeler}} \bibnamefont{and}
1378:   \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{Barkley}},
1379:   \bibinfo{journal}{SIAM Journal on Applied Dynamical Systems}
1380:   \textbf{\bibinfo{volume}{5(1)}}, \bibinfo{pages}{157} (\bibinfo{year}{2006}).
1381: 
1382: \bibitem[{\citenamefont{Davydov et~al.}(1988)\citenamefont{Davydov, Zykov,
1383:   Mikhailov, and Brazhnik}}]{Davydov-etal-1988}
1384: \bibinfo{author}{\bibfnamefont{V.~A.} \bibnamefont{Davydov}},
1385:   \bibinfo{author}{\bibfnamefont{V.~S.} \bibnamefont{Zykov}},
1386:   \bibinfo{author}{\bibfnamefont{A.~S.} \bibnamefont{Mikhailov}},
1387:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{P.~K.}
1388:   \bibnamefont{Brazhnik}}, \bibinfo{journal}{Izvestia VUZov - Radiofizika}
1389:   \textbf{\bibinfo{volume}{31}}, \bibinfo{pages}{574} (\bibinfo{year}{1988}),
1390:   \bibinfo{note}{in Russian}.
1391: 
1392: \bibitem[{\citenamefont{Zykov}(1987)}]{Zykov-1987}
1393: \bibinfo{author}{\bibfnamefont{V.~S.} \bibnamefont{Zykov}},
1394:   \bibinfo{journal}{Biofizika} \textbf{\bibinfo{volume}{32}},
1395:   \bibinfo{pages}{337} (\bibinfo{year}{1987}), \bibinfo{note}{in Russian}.
1396: 
1397: \bibitem[{\citenamefont{Keener and Tyson}(1990)}]{Keener-Tyson-1990}
1398: \bibinfo{author}{\bibfnamefont{J.~P.} \bibnamefont{Keener}} \bibnamefont{and}
1399:   \bibinfo{author}{\bibfnamefont{J.~J.} \bibnamefont{Tyson}},
1400:   \bibinfo{journal}{Physica D} \textbf{\bibinfo{volume}{44}},
1401:   \bibinfo{pages}{191} (\bibinfo{year}{1990}).
1402: 
1403: \bibitem[{\citenamefont{Keener and Tyson}(1991)}]{Keener-Tyson-1991}
1404: \bibinfo{author}{\bibfnamefont{J.~P.} \bibnamefont{Keener}} \bibnamefont{and}
1405:   \bibinfo{author}{\bibfnamefont{J.~J.} \bibnamefont{Tyson}},
1406:   \bibinfo{journal}{Physica D} \textbf{\bibinfo{volume}{53}},
1407:   \bibinfo{pages}{151} (\bibinfo{year}{1991}).
1408: 
1409: \bibitem[{\citenamefont{Davydov et~al.}(1991)\citenamefont{Davydov, Zykov, and
1410:   Mikhailov}}]{Davydov-etal-1991}
1411: \bibinfo{author}{\bibfnamefont{V.~A.} \bibnamefont{Davydov}},
1412:   \bibinfo{author}{\bibfnamefont{V.~S.} \bibnamefont{Zykov}}, \bibnamefont{and}
1413:   \bibinfo{author}{\bibfnamefont{A.~S.} \bibnamefont{Mikhailov}},
1414:   \bibinfo{journal}{Usp. Fiz. Nauk} \textbf{\bibinfo{volume}{161}},
1415:   \bibinfo{pages}{45} (\bibinfo{year}{1991}), \bibinfo{note}{in Russian}.
1416: 
1417: \bibitem[{\citenamefont{Biktashev and Holden}(1994)}]{Biktashev-Holden-1994}
1418: \bibinfo{author}{\bibfnamefont{V.~N.} \bibnamefont{Biktashev}}
1419:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{A.~V.}
1420:   \bibnamefont{Holden}}, \bibinfo{journal}{J. Theor. Biol.}
1421:   \textbf{\bibinfo{volume}{169}}, \bibinfo{pages}{101} (\bibinfo{year}{1994}).
1422: 
1423: \bibitem[{\citenamefont{Biktashev}(1998)}]{Biktashev-1996}
1424: \bibinfo{author}{\bibfnamefont{V.~N.} \bibnamefont{Biktashev}},
1425:   \bibinfo{journal}{Int. J. of Bifurcation and Chaos}
1426:   \textbf{\bibinfo{volume}{8}}, \bibinfo{pages}{677} (\bibinfo{year}{1998}).
1427: 
1428: \bibitem[{\citenamefont{Krinsky et~al.}(1996)\citenamefont{Krinsky, Hamm, and
1429:   Voignier}}]{Krinsky-etal-1996}
1430: \bibinfo{author}{\bibfnamefont{V.}~\bibnamefont{Krinsky}},
1431:   \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{Hamm}}, \bibnamefont{and}
1432:   \bibinfo{author}{\bibfnamefont{V.}~\bibnamefont{Voignier}},
1433:   \bibinfo{journal}{Phys. Rev. Lett.} \textbf{\bibinfo{volume}{76}},
1434:   \bibinfo{pages}{3854} (\bibinfo{year}{1996}).
1435: 
1436: \bibitem[{\citenamefont{Henry}(2004)}]{Henry-2004}
1437: \bibinfo{author}{\bibfnamefont{H.}~\bibnamefont{Henry}},
1438:   \bibinfo{journal}{Phys. Rev. E} \textbf{\bibinfo{volume}{70}},
1439:   \bibinfo{pages}{026204} (\bibinfo{year}{2004}).
1440: 
1441: \bibitem[{\citenamefont{Keener}(1988)}]{Keener-1988}
1442: \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{Keener}},
1443:   \bibinfo{journal}{Physica D} \textbf{\bibinfo{volume}{31}},
1444:   \bibinfo{pages}{269} (\bibinfo{year}{1988}).
1445: 
1446: \bibitem[{\citenamefont{Biktashev et~al.}(1994)\citenamefont{Biktashev, Holden,
1447:   and Zhang}}]{Biktashev-etal-1994}
1448: \bibinfo{author}{\bibfnamefont{V.~N.} \bibnamefont{Biktashev}},
1449:   \bibinfo{author}{\bibfnamefont{A.~V.} \bibnamefont{Holden}},
1450:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{H.}~\bibnamefont{Zhang}},
1451:   \bibinfo{journal}{Philos. Trans. R. Soc. London ser. A}
1452:   \textbf{\bibinfo{volume}{347}}, \bibinfo{pages}{611} (\bibinfo{year}{1994}).
1453: 
1454: \bibitem[{\citenamefont{Biktasheva et~al.}(1998)\citenamefont{Biktasheva,
1455:   Elkin, and Biktashev}}]{Biktasheva-etal-1998}
1456: \bibinfo{author}{\bibfnamefont{I.~V.} \bibnamefont{Biktasheva}},
1457:   \bibinfo{author}{\bibfnamefont{Y.~E.} \bibnamefont{Elkin}}, \bibnamefont{and}
1458:   \bibinfo{author}{\bibfnamefont{V.~N.} \bibnamefont{Biktashev}},
1459:   \bibinfo{journal}{Phys. Rev. E} \textbf{\bibinfo{volume}{57}},
1460:   \bibinfo{pages}{2656} (\bibinfo{year}{1998}).
1461: 
1462: \bibitem[{\citenamefont{Biktasheva et~al.}(1999)\citenamefont{Biktasheva,
1463:   Elkin, and Biktashev}}]{Biktasheva-etal-1999}
1464: \bibinfo{author}{\bibfnamefont{I.~V.} \bibnamefont{Biktasheva}},
1465:   \bibinfo{author}{\bibfnamefont{Y.~E.} \bibnamefont{Elkin}}, \bibnamefont{and}
1466:   \bibinfo{author}{\bibfnamefont{V.~N.} \bibnamefont{Biktashev}},
1467:   \bibinfo{journal}{J. Biol. Phys.} \textbf{\bibinfo{volume}{25}},
1468:   \bibinfo{pages}{115} (\bibinfo{year}{1999}).
1469: 
1470: \bibitem[{\citenamefont{Biktasheva}(2000)}]{Biktasheva-2000}
1471: \bibinfo{author}{\bibfnamefont{I.~V.} \bibnamefont{Biktasheva}},
1472:   \bibinfo{journal}{Phys. Rev. E} \textbf{\bibinfo{volume}{62}},
1473:   \bibinfo{pages}{8800} (\bibinfo{year}{2000}).
1474: 
1475: \bibitem[{\citenamefont{Biktasheva and
1476:   Biktashev}(2001)}]{Biktasheva-Biktashev-2001}
1477: \bibinfo{author}{\bibfnamefont{I.~V.} \bibnamefont{Biktasheva}}
1478:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{V.~N.}
1479:   \bibnamefont{Biktashev}}, \bibinfo{journal}{J. Nonlin. Math. Phys.}
1480:   \textbf{\bibinfo{volume}{8 Supl.}}, \bibinfo{pages}{28}
1481:   (\bibinfo{year}{2001}).
1482: 
1483: \bibitem[{\citenamefont{Biktasheva and
1484:   Biktashev}(2003)}]{Biktasheva-Biktashev-2003}
1485: \bibinfo{author}{\bibfnamefont{I.~V.} \bibnamefont{Biktasheva}}
1486:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{V.~N.}
1487:   \bibnamefont{Biktashev}}, \bibinfo{journal}{Phys. Rev. E}
1488:   \textbf{\bibinfo{volume}{67}}, \bibinfo{pages}{026221}
1489:   (\bibinfo{year}{2003}).
1490: 
1491: \bibitem[{\citenamefont{Sandstede and Scheel}(2004)}]{Sandstede-Scheel-2004}
1492: \bibinfo{author}{\bibfnamefont{B.}~\bibnamefont{Sandstede}} \bibnamefont{and}
1493:   \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Scheel}},
1494:   \bibinfo{journal}{SIAM Journal on Applied Dynamical Systems} pp.
1495:   \bibinfo{pages}{1--68} (\bibinfo{year}{2004}).
1496: 
1497: \bibitem[{\citenamefont{Hamm}(1997)}]{Hamm-1997}
1498: \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{Hamm}}, Ph.D. thesis,
1499:   \bibinfo{school}{Universit\'e de Nice - Sophia Antipolice / Institut Non
1500:   Lin\'eair de Nice} (\bibinfo{year}{1997}).
1501: 
1502: \bibitem[{\citenamefont{Henry and Hakim}(2002)}]{Henry-Hakim-2002}
1503: \bibinfo{author}{\bibfnamefont{H.}~\bibnamefont{Henry}} \bibnamefont{and}
1504:   \bibinfo{author}{\bibfnamefont{V.}~\bibnamefont{Hakim}},
1505:   \bibinfo{journal}{Phys. Rev. E} \textbf{\bibinfo{volume}{65}},
1506:   \bibinfo{pages}{046235} (\bibinfo{year}{2002}).
1507: 
1508: \bibitem[{\citenamefont{Biktasheva et~al.}(2006)\citenamefont{Biktasheva,
1509:   Holden, and Biktashev}}]{Biktasheva-etal-2006}
1510: \bibinfo{author}{\bibfnamefont{I.~V.} \bibnamefont{Biktasheva}},
1511:   \bibinfo{author}{\bibfnamefont{A.~V.} \bibnamefont{Holden}},
1512:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{V.~N.}
1513:   \bibnamefont{Biktashev}}, \bibinfo{journal}{IJBC}
1514:   \textbf{\bibinfo{volume}{16}}, \bibinfo{pages}{1547} (\bibinfo{year}{2006}).
1515: 
1516: \bibitem[{\citenamefont{Fornberg}(1998)}]{Fornberg-1998}
1517: \bibinfo{author}{\bibfnamefont{B.}~\bibnamefont{Fornberg}},
1518:   \emph{\bibinfo{title}{A Practical Guide to Pseudospectral Methods}}
1519:   (\bibinfo{publisher}{Cambridge University Press}, \bibinfo{year}{1998}).
1520: 
1521: \bibitem[{\citenamefont{Press et~al.}(1992)\citenamefont{Press, Flannery,
1522:   Teukolsky, and Vetterling}}]{Numerical-Recipes}
1523: \bibinfo{author}{\bibfnamefont{W.}~\bibnamefont{Press}},
1524:   \bibinfo{author}{\bibfnamefont{B.}~\bibnamefont{Flannery}},
1525:   \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Teukolsky}},
1526:   \bibnamefont{and}
1527:   \bibinfo{author}{\bibfnamefont{W.}~\bibnamefont{Vetterling}},
1528:   \emph{\bibinfo{title}{Numerical Recipes in {C}}}
1529:   (\bibinfo{publisher}{Cambridge University Press},
1530:   \bibinfo{address}{Cambridge, UK}, \bibinfo{year}{1992}).
1531: 
1532: \bibitem[{\citenamefont{Lehoucq et~al.}(1998)\citenamefont{Lehoucq, Sorensen,
1533:   and Yang}}]{ARPACK}
1534: \bibinfo{author}{\bibfnamefont{R.~B.} \bibnamefont{Lehoucq}},
1535:   \bibinfo{author}{\bibfnamefont{D.~C.} \bibnamefont{Sorensen}},
1536:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{C.}~\bibnamefont{Yang}},
1537:   \emph{\bibinfo{title}{{ARPACK} Users' Guide}} (\bibinfo{publisher}{SIAM},
1538:   \bibinfo{year}{1998}), ISBN \bibinfo{isbn}{0898714079, 9780898714074}.
1539: 
1540: \bibitem[{\citenamefont{Hagan}(1982)}]{Hagan-1982}
1541: \bibinfo{author}{\bibfnamefont{P.~S.} \bibnamefont{Hagan}},
1542:   \bibinfo{journal}{SIAM J. Appl. Math.} \textbf{\bibinfo{volume}{42}},
1543:   \bibinfo{pages}{762} (\bibinfo{year}{1982}).
1544: 
1545: \bibitem[{\citenamefont{Biktashev}(1989{\natexlab{c}})}]{Biktashev-1989-inNW}
1546: \bibinfo{author}{\bibfnamefont{V.~N.} \bibnamefont{Biktashev}}, in
1547:   \emph{\bibinfo{booktitle}{Nonlinear Waves II. Dynamics and evolution}},
1548:   edited by \bibinfo{editor}{\bibfnamefont{A.~V.}
1549:   \bibnamefont{Gaponov-Grekhov}}, \bibinfo{editor}{\bibfnamefont{M.~I.}
1550:   \bibnamefont{Rabinovich}}, \bibnamefont{and}
1551:   \bibinfo{editor}{\bibfnamefont{J.}~\bibnamefont{Engelbrecht}}
1552:   (\bibinfo{publisher}{Springer}, \bibinfo{address}{Berlin},
1553:   \bibinfo{year}{1989}{\natexlab{c}}), pp. \bibinfo{pages}{87--96}.
1554: 
1555: \bibitem[{\citenamefont{Sandstede}(2005)}]{Sandstede-private}
1556: \bibinfo{author}{\bibfnamefont{B.}~\bibnamefont{Sandstede}},
1557:   \bibinfo{howpublished}{private communication} (\bibinfo{year}{2005}).
1558: 
1559: \bibitem[{\citenamefont{Margerit and Barkley}(2001)}]{Margerit-Barkley-2001}
1560: \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{Margerit}} \bibnamefont{and}
1561:   \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{Barkley}},
1562:   \bibinfo{journal}{Phys. Rev. Lett.} \textbf{\bibinfo{volume}{86}},
1563:   \bibinfo{pages}{175} (\bibinfo{year}{2001}).
1564: 
1565: \end{thebibliography}
1566: 
1567: \end{document}
1568: