1: \documentclass[twocolumn,english]{revtex4}
2: \usepackage[T1]{fontenc}
3: \usepackage[latin1]{inputenc}
4: \usepackage{graphicx}
5: \pdfoutput=1
6:
7: \makeatletter
8:
9: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% LyX specific LaTeX commands.
10: %% Bold symbol macro for standard LaTeX users
11: \providecommand{\boldsymbol}[1]{\mbox{\boldmath $#1$}}
12:
13:
14: \usepackage{babel}
15: \makeatother
16: \begin{document}
17:
18: \title{Pinwheel stability, pattern selection and the geometry of visual
19: space }
20:
21: \begin{abstract}
22: It has been proposed that the dynamical stability of topological defects
23: in the visual cortex reflects the Euclidean symmetry of the visual
24: world. We analyze defect stability and pattern selection in a generalized
25: Swift-Hohenberg model of visual cortical development symmetric under
26: the Euclidean group E(2). Euclidean symmetry strongly influences the
27: geometry and multistability of model solutions but does not directly
28: impact on defect stability.
29: \end{abstract}
30:
31: \author{Michael Schnabel$^{1,2}$, Matthias Kaschube$^{1,2,3,4}$ and Fred
32: Wolf$^{1,2}$}
33:
34:
35: \affiliation{$^{1}$Max Planck Institute for Dynamics and Self-Organization, Goettingen,
36: Germany, $^{2}$Bernstein Center for Computational Neuroscience, University
37: of Goettingen, Goettingen, Germany, $^{3}$Lewis-Sigler Institute,
38: $^{4}$Physics Department, Princeton University, Princeton, New Jersey,
39: USA}
40:
41: \maketitle
42: When an object in the visual world is rotated about the axis of sight
43: the orientations of distant contours change in a coordinated manner,
44: leaving relative orientations of distant contours invariant. In the
45: visual cortex of the brain, it has been predicted that this geometrical
46: property of visual space imposes the so called shift-twist symmetry
47: on joint representations of contour position and contour orientation
48: \cite{Bressloff:2001}. This symmetry requires the equivariance of
49: dynamical models for these representations under the Euclidean group
50: E(2). In particular, it has been hypothesized that shift-twist symmetry
51: stabilizes topological pinwheel-defects in models for the emergence
52: of orientation selectivity during cortical development \cite{Mayer:2002,Lee:2003,Bressloff:2005}.
53: Pinwheel defects are singular points in the visual cortex around which
54: each orientation is represented exactly once \cite{Swindale:1987b,Bonhoeffer:1991,Ohki:2006}.
55: They are initially generated in the visual cortex at the time of eye
56: opening \cite{White:2001b,Chapman:1996}. This process has been theoretically
57: explained by spontaneous symmetry breaking \cite{Swindale:1982b,Wolf:1998}.
58: Why in the brain, pinwheels remain present at all developmental stages,
59: although they are dynamically unstable in many models of visual cortical
60: development \cite{Wolf:1998,Koulakov:2001,Mayer:2002,Cho:2004,Lee:2003,Wolf:2005b}
61: and in analogous physical systems \cite{Cross:1993,Vilenkin:1994,Bodenschatz:2000},
62: remains unclear.
63:
64: Previous studies \cite{Wolf:1998,Koulakov:2001,Mayer:2002,Lee:2003,Cho:2004,Wolf:2005b}
65: found pinwheels dynamically unstable only in models exhibiting an
66: E(2)xU(1) symmetry, which is higher than the E(2) symmetry of visual
67: perceptual space. In contradistinction, models exhibiting only Euclidean
68: symmetry have been shown to exhibit stable pinwheels \cite{Mayer:2002,Lee:2003,Thomas:2004}
69: suggesting that the stabilization of pinwheel defects may be closely
70: related to the 'reduced' Euclidean symmetry. There are, however, also
71: other scenarios that predict the emergence of stable pinwheels with
72: higher than Euclidean symmetry \cite{Wolf:2005b,Koulakov:2001} leaving
73: it hard to judge to actual role of Euclidean symmetry in orientation
74: map development.
75:
76: Here we analyze the impact of Euclidean symmetry on pattern selection,
77: i.e. the question of whether stable pinwheel arrangement exist and
78: what their geometric organization is. We construct a generalized Swift-Hohenberg
79: model \cite{Cross/Hohenberg:1993,SwiftHohenberg:1977} symmetric under
80: the Euclidean group E(2) that allows to study the transition from
81: higher E(2)xU(1) to lower E(2) symmetry by changing a parameter that
82: controls the strength of shift symmetry breaking (SSB). Using weakly
83: nonlinear analysis we derive amplitude equations for stationary planforms
84: and find three classes of stationary solutions: stripe patterns without
85: any pinwheels, pinwheel crystals with pinwheels regularly arranged
86: on a rhombic lattice, and quasi-periodic patterns containing a large
87: number of irregularly spaced pinwheels. We calculate the phase diagram
88: of these solutions depending on the strength of SSB, the effective
89: strength of nonlocal interactions, and the range of nonlocal interactions.
90: With increasing strength of SSB, pinwheel free patterns are progressively
91: replaced by pinwheel crystals in the phase diagram while both pinwheel
92: free patterns and pinwheel crystals remain stable. Phases of aperiodic
93: pinwheel rich patterns remain basically unaffected. A critical strength
94: of SSB exists above which multistable aperiodic patterns collapse
95: into a single aperiodic state. %
96: \begin{figure}
97: \includegraphics[clip,width=0.95\columnwidth]{fig_9}
98:
99:
100: \caption{\textbf{(a)} Plane waves with wavevector in horizontal (\emph{top})
101: and oblique (\emph{bottom}) direction for variable strength of SSB
102: \textbf{$(\epsilon=0,\,0.35,\,1)$ (b)} PWCs of varying intersection
103: angle $\alpha$, $\pi/4\le\alpha\le\pi/2$. \textbf{(c)} Energy of
104: solutions depends on $\epsilon$ and on $\alpha$. For sufficiently
105: large $\epsilon$ PWCs are energetically favored relative to plane
106: waves. (\emph{dashed}: energy of plane waves, \emph{plain}: energies
107: of PWCs for $\alpha=\pi/4,\,\pi/3,\,\pi/2$) \label{fig:1}}
108: \end{figure}
109:
110:
111: The spatial structure of an OPM can be represented by a complex field
112: $z(\mathbf{x})$ where $\mathbf{x}$ denotes the 2D position of neurons
113: in the visual cortex, $\theta(\mathbf{x})=\arg\left(z(\mathbf{x})\right)/2$
114: their preferred stimulus orientation, and the modulus $|z(\mathbf{x})|$
115: is a measure of their selectivity \cite{Swindale:1982b}. In this
116: representation, pinwheel centers are zeros of the field $z(\mathbf{x})$.
117: The simplest models for the formation of OPMs are defined by a dynamics
118: \begin{equation}
119: \partial_{t}z(\mathbf{x},t)=F[z](\mathbf{x},t).\label{eq:1}\end{equation}
120: where $t$ denotes time and $F[z]$ is a nonlinear operator. We assume
121: the dynamics equivariant under translation $T_{\mathbf{y}}z(\mathbf{x})=z(\mathbf{x}+\mathbf{y})$,
122: rotation $R_{\alpha}z(\mathbf{x})=e^{2i\alpha}z(\Omega_{-\alpha}\mathbf{x})$
123: with rotation matrix $\Omega_{\phi}$, and reflection at the cortical
124: $(1,0)$ axis $Pz(\mathbf{x})=\bar{z}(\bar{\mathbf{x}})$, thus expressing
125: the fact that within cortical layers there are no special locations
126: or directions \cite{Braitenberg/Schuz:1998}. In addition, if interactions
127: between OPM development and visuotopy are neglected it is also equivariant
128: under global shifts of orientation preference $S_{\beta}z(\mathbf{x})=e^{i\beta}z(\mathbf{x})$
129: (\emph{shift symmetry}). Rotations $R$ thus consist of a composition
130: of phase shifts $S$ and coordinate rotations $D$, i.e. $R_{\alpha}=S_{2\alpha}\circ D_{\alpha}$
131: with $D_{\alpha}z(\mathbf{x})=z(\Omega_{-\alpha}\mathbf{x})$.
132:
133: We consider the general class of variational models \cite{Lee:2003,Wolf:2005b,Swindale:1982b}
134: for which $F[z]$ has the form
135:
136: \begin{equation}
137: F[z]=Lz+\epsilon M\bar{z}+N_{3}[z].\label{eq:F[z]}\end{equation}
138: Here $L$ is a linear, translation invariant and self-adjoint operator,
139: that accounts for a finite wavelength instability. $N_{3}$ is a cubic
140: nonlinearity which stabilizes the dynamics. The second term involves
141: a complex conjugation $Cz=\bar{z}$ and thus manifestly breaks shift
142: symmetry when $\epsilon\neq0$. $M$ is assumed to be linear, translation
143: invariant and bounded. Equivariance under rotations, $[MC,\, R_{\alpha}]=0$,
144: requires \begin{equation}
145: D_{\alpha}MD_{\alpha}^{-1}=S_{-4\alpha}M\label{eq:shtw_trof_rule}\end{equation}
146: and equivariance under parity $[M,\, P]=0$.
147:
148: As a concrete example we will consider the model \begin{eqnarray}
149: L & = & r-(k_{c}^{2}+\nabla^{2})^{2}\label{eq:L}\\
150: N_{3}[z] & = & (1-g)|z(\mathbf{x})|^{2}z(\mathbf{x})-\frac{2-g}{2\pi\sigma^{2}}\label{eq:N[z]}\\
151: & \times & \int d^{2}y\left(|z(\mathbf{y})|^{2}z(\mathbf{x})+\frac{1}{2}z(\mathbf{y})^{2}\bar{z}(\mathbf{x})\right)e^{-|\mathbf{y}-\mathbf{x}|^{2}/2\sigma^{2}}\nonumber \\
152: M & = & r{(\partial}_{x}+i\partial_{y})^{4}{(\partial}_{xx}+\partial_{yy})^{-2}\label{eq:M}\end{eqnarray}
153: where $L$ is the Swift-Hohenberg operator \cite{Cross/Hohenberg:1993,SwiftHohenberg:1977}
154: with critical wavenumber $k_{c}$ and instability parameter $r$.
155: $N_{3}$ is adopted from \cite{Wolf:2005b}, where $\sigma$ sets
156: the range of the nonlocal interactions and $g$ determines whether
157: the local $(g>1)$ or the nonlocal term $(g<1)$ stabilizes the dynamics.
158: $M$ is the simplest differential operator which transforms according
159: to Eq.(\ref{eq:shtw_trof_rule}). It is unitary with spectrum $\propto e^{4i\arg(\mathbf{k})}$.
160:
161: We used weakly nonlinear analysis \cite{Manneville:1990} to study
162: potential solutions of Eq.(\ref{eq:F[z]}). We consider planforms
163: $z(\mathbf{x})=\sum_{j=0}^{2n-1}A_{j}e^{i\mathbf{k}_{j}\mathbf{x}},\,\,\mathbf{k}_{j}=k_{c}(\cos\alpha_{j},\,\sin\alpha_{j})$
164: with $2n$ modes where we require that to each mode also its antiparallel
165: mode is in the set. By symmetry, the dynamics of the amplitudes $A_{j}$
166: at threshold has the form\begin{equation}
167: \dot{A}_{j}=A_{j}+\epsilon\bar{A}_{j-}e^{4i\alpha_{j}}-\sum_{k=0}^{2n-1}g_{jk}|A_{k}|^{2}A_{j}-\sum_{k=0}^{2n-1}f_{jk}A_{k}A_{k-}\bar{A}_{j-}\label{eq:amplitude}\end{equation}
168: where $j^{-}$ denotes the index of the mode antiparallel to mode
169: $j$, $\mathbf{k}_{j^{-}}=-\mathbf{k}_{j}$ and with real valued and
170: symmetric matrices $g_{jk}$ and $f_{jk}$ which determine the coupling
171: and competition between modes. They can be expressed in terms of angle-dependent
172: interaction functions $g(\alpha)$ and $f(\alpha)$, which are obtained
173: from the nonlinearity $N_{3}[z]$ (cf.\cite{Wolf:2005b,Cross/Hohenberg:1993,Manneville:1990}).
174: For simplicity we restrict the following analysis to the class of
175: permutation symmetric models, defined in \cite{Wolf:2005b}, for which
176: $g(\alpha)=g(\alpha+\pi)$. %
177: \footnote{For the nonlinearity Eq.(\ref{eq:N[z]}) one obtains $g(\alpha)=g+(2-g)e(\alpha)$
178: and $f(\alpha)=\frac{1}{2}g(\alpha)$ with $e(\alpha)=2\exp(-\sigma^{2}k_{c}^{2})\cosh(\sigma^{2}k_{c}^{2}\cos\alpha)$
179: (cf.\cite{Wolf:2005b} for details). The coupling coefficients are
180: given by $g_{jk}=(1-\frac{1}{2}\delta_{jk})g(|\alpha_{k}-\alpha_{j}|)$
181: and $f_{jk}=(1-\delta_{jk}-\delta_{jk^{-}})f(|\alpha_{k}-\alpha_{j}|)$. %
182: }
183:
184: The simplest solution to Eq.(\ref{eq:amplitude}) is obtained for
185: $n=1$ and consists of plane waves with wavevector $\mathbf{k}=k_{c}(\cos\alpha,\sin\alpha)$.
186: For $|\epsilon|\le1/2$ it is given by \begin{equation}
187: z(\mathbf{x})=\frac{e^{2i\alpha}}{\sqrt{g_{00}}}[\sqrt{1+2\epsilon}\,\cos(\mathbf{k}\mathbf{x}+\phi)+i\sqrt{1-2\epsilon}\sin(\mathbf{k}\mathbf{x}+\phi)]\label{eq:plane wave}\end{equation}
188: with arbitrary phase $\phi$. Hence with SSB orientation angles are
189: no longer equally represented. For $\epsilon>0$ cortical area for
190: orientations $\alpha$ and $\alpha+\pi/2$ is recruited at the expense
191: of $\alpha+\pi/4$ and $\alpha+3\pi/4$ (and vice versa for $\epsilon<0$).
192: Beyond a critical strength of SSB, $\epsilon_{*}=1/2$, patterns only
193: contain two orientations, $z(\mathbf{x})=e^{2i\alpha}\mathcal{N}\,\cos(\mathbf{k}\mathbf{x}+\phi)$
194: for $\epsilon>\epsilon_{*}$ and $z(\mathbf{x})=ie^{2i\alpha}\mathcal{N}\,\sin(\mathbf{k}\mathbf{x}+\phi)$
195: for $\epsilon<-\epsilon_{*}$ with $\mathcal{N=}\sqrt{4(1+|\epsilon|)/3g_{00}}$
196: (Fig.\ref{fig:1}a).
197:
198: Another class of solutions, rhombic pinwheel crystals (PWCs), exist
199: for $n=2$ and consist of two pairs of antiparallel modes forming
200: an angle $0<\alpha\le\pi/2$ which are characterized by $|A_{0}|=|A_{0^{-}}|=a=|A_{1}|=|A_{1^{-}}|$.
201: We consider w.l.o.g. the case $\alpha_{0}=-\alpha/2$ and $\alpha_{1}=\alpha/2$
202: (Fig.\ref{fig:1}b). With $A_{0,1}=ae^{i\mu_{0,1}},\, A_{0^{-},1^{-}}=ae^{i\nu_{0,1}}$
203: and $\Sigma_{0,1}:=\mu_{0,1}+\nu_{0,1}$ the stationary state is given
204: by $\Sigma_{1}=-\Sigma_{0}$ and $a^{2}=(1+\epsilon\cos(\Sigma_{0}+2\alpha))/\zeta$
205: where $\zeta=3g_{00}+2g_{01}+2f_{01}\cos2\Sigma_{0}$. The phase $\Sigma_{0}$
206: is the solution to $0=\sin2\Sigma_{0}+\epsilon[\sin(\Sigma_{0}-2\alpha)-(2+3g_{00}/g_{01})\sin(\Sigma_{0}+2\alpha)]$
207: which bifurcates from $\Sigma_{0}=\pm\pi/2$ for $\epsilon=0$. The
208: energy is given by $E_{PWC}=-4a�(1+|\epsilon|\cos(\Sigma_{0}+2\alpha))+2a^{4}\zeta$.
209: For the model Eqs.(\ref{eq:L}-\ref{eq:M}) the $\epsilon$ and $\alpha$
210: dependence of the energy is shown in Fig.\ref{fig:1}c. The solution
211: then reads $z(\mathbf{x})=2a\left[e^{i\Sigma_{0}/2}\cos(\mathbf{k}_{0}\mathbf{x}+\Delta_{0}/2)+e^{-i\Sigma_{0}/2}\sin(\mathbf{k}_{1}\mathbf{x}+\Delta_{1}/2)\right]$
212: with arbitrary $\Delta_{0}$ and $\Delta_{1}$.
213:
214: A large set of quasiperiodic solutions originates from the essentially
215: complex planforms (ECP) $z(\mathbf{x})=\sum_{j=0}^{n-1}A_{j}^{+}e^{il_{j}\mathbf{k}_{j}\mathbf{x}}$
216: that solve Eq. (\ref{eq:amplitude}) for $\epsilon=0$ \cite{Wolf:2005b}.
217: Here, wave vectors $\mathbf{k}_{j}=k_{c}(\cos\frac{\pi}{N}j,\,\sin\frac{\pi}{N}j)$
218: $(j=0,\dots,\, n-1)$ are distributed equidistantly on the upper half
219: of the critical circle and binary variables $l_{j}=\pm1$ determine
220: whether the mode with wave vector $\mathbf{k}_{j}$ or with wave vector
221: $-\mathbf{k}_{j}$ is active (Fig.\ref{fig:2}(a) \emph{left column}).
222: %
223: \begin{figure}
224: \includegraphics[clip,width=0.95\columnwidth]{fig_8}
225:
226:
227: \caption{Attractors of amplitude equations (Eq.\ref{eq:amplitude}) with full,
228: partially broken and completely broken shift symmetry. \textbf{(a)}
229: ECPs. Preferred orientations are color coded {[}see bars in (b)].
230: Arrangement of active modes on the critical circle and corresponding
231: OPMs. For n = 3 and 8 there are 2 and 15 different classes of ECPs,
232: respectively. Complete (partial, no) suppression of opposite modes
233: for full (weakly broken, maximally broken) shift symmetry \emph{(left,
234: middle, right column).} \textbf{(b)} OPM in tree shrew V1 (data: L.E.White,
235: Duke Univ., USA). Arrows pinwheel centers. Scale bar 1 mm. \textbf{(c)}
236: With increasing degree of symmetry breaking $\epsilon$ amplitudes
237: of antiparallel modes $A_{-}$ grow and eventually (at $\epsilon=\epsilon_{*}$)
238: reach the same absolute value as active modes $A_{+}.$\label{fig:2}}
239: \end{figure}
240: %
241: \begin{figure}
242: \includegraphics[clip,width=0.95\columnwidth]{fig_3}
243:
244:
245: \caption{\textbf{(a)}-\textbf{(c)} Pinwheel densities for all realizations
246: of ECPs with $3\le n\le17$ and different degrees of shift symmetry
247: breaking $\epsilon$. \textbf{(d)} Pinwheel densities for $n=17$
248: \emph{(dots)} and for $n\to\infty$ in the Gaussian approximation
249: \emph{(gray region)}.}
250: \end{figure}
251:
252:
253: For $|\epsilon|>0$ we find that ECPs generalize to $z(\mathbf{x})=\sum_{j=0}^{n-1}\left[A_{j}^{+}e^{il_{j}\mathbf{k}_{j}\mathbf{x}}+A_{j}^{-}e^{-il_{j}\mathbf{k}_{j}\mathbf{x}}\right]$.
254: For each $n$ there exists a critical value $\epsilon_{*}:=\gamma{[g}_{00}+2\sum_{j=0}^{n-1}f_{ij}\cos\frac{4\pi}{n}j]$,
255: where $\gamma=[2\sum_{j=0}^{n-1}g_{ij}]^{-1}$. When $|\epsilon|\le{|\epsilon}_{*}|$
256: stationary amplitudes $a^{\pm}=|A_{j}^{\pm}|$ fulfill $a_{\pm}^{2}=\gamma(1\pm\sqrt{1-\epsilon^{2}/\epsilon_{*}^{2}})$.
257: When $|\epsilon|\ge{|\epsilon}_{*}|$ the amplitude of antiparallel
258: and active modes are equal $a_{\pm}^{2}=\gamma[1+|\epsilon|]/[1+\epsilon_{*}]$
259: (Fig.\ref{fig:2}c). A simple measure of the degree to which SSB affect
260: $n$-ECPs is \[
261: q:=\frac{\sum_{j}A_{j}^{+}A_{j}^{-}e^{-i\frac{4\pi}{n}j}+c.c.}{\sum_{j}\left|A_{j}^{+}\right|^{2}+\left|A_{j}^{-}\right|^{2}}\,.\]
262: For a stationary $n$-ECP we find $q=\epsilon/|\epsilon_{*}|$ if
263: $|\epsilon|\le{|\epsilon}_{*}|$ and $q=\mbox{sign}\,\epsilon$ if
264: $|\epsilon|>{|\epsilon}_{*}|$.
265:
266: Stationary phases $\phi_{j}^{\pm}=\arg A_{j}^{\pm}$ fulfill the condition
267: $\phi_{j}^{+}+\phi_{j}^{-}=(4\pi/n)\, j$ if $\epsilon>0$ and $\phi_{j}^{+}+\phi_{j}^{-}=(4\pi/n)\, j+\pi$
268: if $\epsilon<0$. This implies that all orientations are represented
269: in patterns with $n\ge3$. The solution can be written \[
270: z(\mathbf{x})=\sqrt{2\gamma}\sum_{j=0}^{n-1}[\sqrt{1+q}\, z_{j}^{e}(\mathbf{x},\phi_{j})+\sqrt{1-q}\, z_{j}^{o}(\mathbf{x},\phi_{j})]\]
271: with $z_{j}^{e}(\mathbf{x},\phi_{j})=e^{i\frac{2\pi}{n}j}\cos(l_{j}\mathbf{k}_{j}\mathbf{x}+\phi_{j})$
272: and $z_{j}^{o}(\mathbf{x},\phi_{j})=ie^{i\frac{2\pi}{n}j}\sin(l_{j}\mathbf{k}_{j}\mathbf{x}+\phi_{j})$
273: and arbitrary phases $\phi_{j}$. For $n=1$ this is in agreement
274: with Eq.(\ref{eq:plane wave}) where $\epsilon_{*}=1/2$. Reflection
275: at the axis parallel to $\mathbf{k}_{j}$ acts on the functions $z_{j}^{e}$
276: and $z_{j}^{o}$ as $+1$ and $-1$, respectively. Thus $z_{j}^{e}$
277: and $z_{j}^{o}$ correspond to the even and odd eigenvectors of the
278: nullspace of $L+\epsilon MC$ (cf.\cite{Thomas:2004}). For $\epsilon>0$
279: $(\epsilon<0)$ the even (odd) part dominates the solution.
280:
281: The dynamics Eq.(\ref{eq:amplitude}) exhibits a potentially exceedingly
282: high number of multistable solutions. The energy of $n$-ECPs is given
283: by $E_{n}=-n\gamma[1+\epsilon^{2}/\epsilon_{*}]$ for $|\epsilon|\le\epsilon_{*}$
284: and $E_{n}=-n\gamma(1+|\epsilon|)^{2}/[1+\epsilon_{*}]$ for $|\epsilon|\ge\epsilon_{*},$
285: respectively, and does not depend on the variables $l_{j}$ which
286: identify a particular $n$-ECP. Due to the growth of antiparallel
287: modes with increasing $|\epsilon|$ patterns for all different realizations
288: $l_{j}$ with phases $\phi_{j}:=l_{j}\Phi_{j}+\frac{1}{4}(1-\mbox{sign}(\epsilon))(1-l_{j})\pi$
289: ($\Phi_{j}$ arbitrary but fixed) eventually collapse in a single
290: state $z(\mathbf{x})\propto\sum_{j=0}^{n-1}z_{j}^{e/o}(\mathbf{x},\phi_{j})$
291: (Fig.\ref{fig:2}a).%
292: \begin{figure}
293: \includegraphics[clip,width=0.95\columnwidth]{fig_2}
294:
295:
296: \caption{Phase diagrams of the model, Eqs.(\ref{eq:L}-\ref{eq:M}), near
297: criticality for variable SSB $\epsilon$. The graph shows the regions
298: of the $g-\sigma/\Lambda$ plane in which \emph{n}-ECPs and PWCs have
299: minimal energy ($n=1-25,\, n>25$ dots). Regions of maximally broken
300: shift symmetry {[}$\epsilon\ge\epsilon_{*}(N,g,\sigma)$] shaded in
301: \emph{gray}. Regions where $\alpha$-PWCs prevail is shaded in \emph{blue,}
302: intensity level codes for the relative angle \emph{$\alpha$. (light
303: blue: $\pi/4\le\alpha\le\pi/2\,:$dark blue)} \label{fig:Phase-diagrams}}
304: \end{figure}
305: %
306: \begin{figure}
307: \includegraphics[clip,width=0.95\columnwidth]{fig_4}
308:
309:
310: \caption{Stability regions of ECPs with $n=20$ for different $\epsilon$.
311: Dashed line denotes the critical line $\epsilon=\epsilon_{*}(N,g,\sigma)$,
312: above which $q=1$. Shaded region denotes region in the $g-\sigma/\Lambda$
313: plane for which that planform is a stable solution of the dynamics
314: and coexists with planforms of nearby values of $n$, e.g. $n=18,\,19,\,21,\,22$.
315: In the inner region (marked by the two inner lines) this solution
316: minimizes energy. \label{fig:4}}
317: \end{figure}
318:
319:
320: This collapse manifests itself in the pinwheel densities $\rho_{n}$
321: of various ECPs shown in Fig.3a-c for different strength of SSB. The
322: pinwheel density of an $n$-ECP in the large $n$ limit is $\rho(\chi)=\pi\sqrt{1-\frac{8}{\pi^{2}}(1-\epsilon^{2}/\epsilon_{*}^{2})\zeta^{2}}$
323: and depends through $\zeta=\frac{1}{4n}|\sum_{j=0}^{n-1}l_{j}\mathbf{k}_{j}|\le1$
324: on the configuration of active modes. Fig.3d shows that pinwheel densities
325: fill a band of values. With increasing degree of SSB this band narrows
326: and pinwheel densities eventually equal $\pi$ at the critical value
327: $|\epsilon|=\epsilon_{*}.$
328:
329: To reveal how SSB affects pattern selection we calculated the phase
330: diagram for the model specified in Eqs.(\ref{eq:L}-\ref{eq:M}) for
331: various values of $\epsilon$. Fig.\ref{fig:Phase-diagrams} shows
332: the configurations of $n$-ECPs and PWCs minimizing the energy %
333: \footnote{The dynamics Eq. (\ref{eq:amplitude}) has the real valued energy
334: functional $E=\sum_{j}|A_{j}|�+\epsilon\sum_{j}(\bar{A}_{j}\bar{A}_{j-}e^{4i\frac{\pi}{n}j}+c.c.)-\frac{1}{2}\sum_{jk}g_{jk}|A_{j}|^{2}|A_{k}|^{2}-\frac{1}{2}\sum_{jk}f_{jk}A_{k}A_{k-}\bar{A}_{j}\bar{A}_{j-}$.%
335: }. Plane waves are progressively replaced by $\alpha$-PWCs with increasing
336: SSB. Depending on the location in parameter space and on $\epsilon$,
337: a particular angle $\alpha$ minimizing the energy (c.f. Fig.\ref{fig:1}c).
338: Large $n-$ECPs are selected when the dynamics is stabilized by long-range
339: interactions ($g<1$, $\sigma>\Lambda$). In this parameter regime
340: plane waves and pinwheel crystals are unstable. The degree of SSB
341: $q$ manifest in a given $n-$ECP attractor depends on $\epsilon$
342: and on the location in the phase diagram. Above a critical line defined
343: by $|\epsilon_{*}(N,g,\sigma)|=|\epsilon|$ antiparallel modes are
344: maximal and $|q|=1$ (\emph{gray area}), below that line $|q|\le1$.
345: Figs.~\ref{fig:Phase-diagrams} and \ref{fig:4} show the high sensitivity
346: of the dynamics to even small amounts of SSB, a substantial area in
347: phase is occupied by ECPs with $|q|=1$ even for $\epsilon=0.02$.
348:
349: Our analysis of pattern selection in visual cortical development demonstrates
350: that dynamical models of orientation map development are very sensitive
351: to the presence of interactions imposed by Euclidean E(2) symmetry.
352: It also reveals that the impact of the Euclidean symmetry of perceptual
353: visual space is not an all or none phenomenon. For weak SSB our Euclidean
354: model closely mimics the behavior of models possessing the higher
355: E(2)xU(1) symmetry. The only qualitative change that we found in the
356: transition from higher to lower symmetry was the collapse of multistable
357: solutions. This collapse, however, only happens when a finite critical
358: strength of SSB is reached. Up to this threshold strength, models
359: exhibiting E(2) and E(2)xU(1) symmetry seem to be topologically conjugate
360: to one another.
361:
362: Our analysis also reveals that the impact of Euclidean symmetry differs
363: qualitatively for aperiodic and periodic patterns. For aperiodic patterns,
364: (i) the parameter regime in which they possess minimal energy is virtually
365: unaffected by the strength of SSB, (ii) sets of different multistable
366: solutions become progressively more similar and finally merge forming
367: a single unique ground state, and (iii) Euclidean symmetry geometrically
368: manifests itself in specific two point correlations that are absent
369: in the E(2)xU(1) symmetric limit. For periodic patterns such as pinwheel
370: crystals and pinwheel free states, (i) the parameter regime in which
371: these solutions possess minimal energy depends on the strength of
372: SSB, and (ii) Euclidean symmetry geometrically manifests itself in
373: a specifically selected tilt angle of rhombic pinwheel crystals and
374: a wave vector dependent underrepresentation of particular orientations
375: for pinwheel free states.
376:
377: The Swift-Hohenberg model of Euclidean symmetry considered here predicts
378: that aperiodic pinwheel rich patterns resembling the architecture
379: of the primary visual cortex are only stable when long-range interactions
380: dominate pattern selection, confirming previous predictions of a model
381: of higher E(2)xU(1) symmetry. Our analysis predicts that in this regime,
382: two point correlations provide a sensitive measure of the strength
383: of SSB. Such correlations are therefore a promising tool for probing
384: the impact of Euclidean symmetry in visual cortex development. \bibliographystyle{apsrev}
385: \bibliography{bib}
386:
387: \end{document}
388: