1: %Ch9.tex
2: %
3: % These are the notes for Lecture 7 of my IHP course
4: %
5: % Frank Sottile
6: % 27 November 2005
7: % Darmstadt
8: %
9: %
10: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
11: \chapter{The Shapiro Conjecture for Rational Functions}\label{Ch:EG}
12: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
13:
14: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
15: %\section*{Introduction}
16:
17: We continue our study of the Shapiro Conjecture, which asserts that
18: if $f_1(t),f_2(t),\dotsc,f_m(t)$ are polynomials of degree $m{+}p{-}1$ whose
19: Wronskian is $\Phi(t)$ has degree $mp$ with distinct real roots,
20: then the linear span of $f_1(t),f_2(t),\dotsc,f_m(t)$ is real.
21: We consider an apparently degenerate case, that of rational
22: functions, which is when $m=2$.
23:
24: Eremenko and Gabrielov~\cite{EG02} originally gave a proof of the Shapiro
25: Conjecture when $m=2$ using essentially the uniformization theorem from complex
26: analysis.
27: This proof was also complicated and challenging for {\sl this} reader.
28: They subsequently found a second, significantly more elementary proof~\cite{EG05}.
29: We will first discuss that proof, and then a generalization concerning rational functions that
30: are constant on prescribed sets, which leads to a generalization of the Shapiro Conjecture that
31: we will discuss in Chapter~\ref{Ch:Frontier}.
32:
33: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
34: \section{The Shapiro Conjecture for rational functions}\label{S:SCRF}
35:
36: The Shapiro conjecture for $m=2$ asserts that if $f_1(t)$ and $f_2(t)$ are univariate
37: polynomials whose Wronskian
38: %
39: \begin{equation}\label{E9:Wronski}
40: \Wr(f_1,f_2)\ =\ f'_1(t)f_2(t)\ -\ f_1(t)f'_2(t)
41: \end{equation}
42: %
43: has only real roots, then the complex linear span $\langle f_1,f_2\rangle$ is real in that
44: there are real polynomials $g_1$ and $g_2$ with $\langle f_1,f_2\rangle=\langle g_1,g_2\rangle$.
45:
46: This has a natural interpretation in terms of rational functions.
47: The quotient of two univariate polynomials $f_1$ and $f_2$ defines a rational function
48: $\rho\colon\P^1\to\P^1$ which on $\C\subset\P^1$ is
49: \[
50: \rho\ \colon\ t\ \longmapsto\ f_1(t)/f_2(t)\ .
51: \]
52: Since
53: \[
54: \rho'(t)\ =\ \frac{f'_1(t)f_2(t)\ -\ f_1(t)f'_2(t)}{f_2(t)^2}
55: \ =\ \frac{\Wr(f_1,f_2)}{f_2(t)^2}\,,
56: \]
57: if $f_1$ and $f_2$ are relatively prime, the critical points of $\rho$ in $\P^1$ are
58: the roots of their Wronskian.
59:
60:
61: Two rational functions $\rho_1,\rho_2\colon\P^1\to\P^1$ are
62: \DeCo{{\sl equivalent}} if they differ by a fractional linear transformation on the target $\P^1$.
63: Since a fractional linear transformation on $f_1/f_2$ is a change of basis in the
64: linear span $\langle f_1,f_2\rangle$, an equivalence class of rational functions is
65: simply a two-dimensional space of polynomials.
66: An equivalence class is \DeCo{{\sl real}} if the corresponding linear space is real.
67: We state the theorem of Eremenko and Gabrielov.
68:
69: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
70: \begin{thm}[Shapiro's conjecture for rational functions~\cite{EG02,EG05}]\label{T9:EG}
71: If a rational function $\rho\colon\P^1\to\P^1$ has only real critical points, then
72: $\rho$ is equivalent to a real rational function.
73: \end{thm}
74: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
75:
76: Theorem~\ref{T8:asymptotic} and the Grassmann duality of Theorem~\ref{T8:GrassmannDuality}
77: ensure the existence of a polynomial $\Phi_0(t)\in\R_{2p}[t]$ with only real roots such
78: that every space of polynomials with Wronskian $\Phi_0(t)$ is real, and there are exactly
79: $\#_{2,p}=\frac{1}{p+1}\binom{2p}{p}$ such spaces of polynomials.
80: The elementary proof of Theorem~\ref{T9:EG} analytically continues these
81: $\#_{2,p}$ real spaces of polynomials as the $2p$ distinct real roots of $\Phi_0(t)$ vary.
82: This continuation will produce fewer than $\#_{2,p}$ real spaces of polynomials for any
83: $\Phi(t)$ only if some of the spaces become complex during the continuation.
84: But this can happen only if two spaces of polynomials first become equal during the
85: continuation.
86:
87: The proof shows that such a collision cannot occur by associating discrete objects called nets,
88: to the real rational functions that are distinct from each of the $\#_{2,p}$ spaces of
89: polynomials with Wronskian $\Phi_0(t)$, and which are preserved under a continuation that
90: varies the roots of $\Phi_0(t)$.
91: Thus no collisions are possible, which will imply Theorem~\ref{T9:EG}.
92:
93:
94: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
95: \subsection{Continuity and nets of rational functions}
96:
97: A point $p$ in the Grassmannian $\DeCo{\Gr(2,\C_{p{+}1}[t])}\simeq\Gr(2,p{+}2)$ is a
98: two-dimensional space $V$ of univariate polynomials of degree at most $p{+}1$.
99: Each such point gives an equivalence class of rational functions
100: $\rho\colon\P^1\to\P^1$ of degree $p{+}1{-}\delta$, where $\delta$ is the maximum
101: degree of a common factor of the polynomials in $V$.
102: Working with this equivalence is awkward, so we will instead use the
103: \DeCo{{\sl Stiefel manifold}}, \DeCo{$\St_{p{+}1}$}, which is a $GL(2,\R)$-fiber bundle
104: over $\Gr(2,\C_{p{+}1}[t])$.
105:
106: The points of $\St_{p{+}1}$ are pairs of nonproportional univariate polynomials of
107: degree at most $p{+}1$.
108: Hence $\St_{p{+}1}$ is an open subset of $\R^{2p+4}$, with coordinates the
109: coefficients of the polynomials $f$ and $g$.
110: We give $\St_{p{+}1}$ the subspace topology.
111: The association $\St_{p{+}1}\ni(f,g)\longmapsto f/g$ defines a map (written
112: \DeCo{$\pi$}) from $\St_{p{+}1}$ to the
113: space of rational functions.
114: While this map is not continuous as a map of spaces, it does have the weak
115: continuity property given in Proposition~\ref{P9:Weak_cont} below.
116:
117: Let $\DeCo{Z}\subset \St_{p{+}1}$ be the locus of pairs $(f,g)$ with either
118: %%%%%%%%%%%%%%
119: \begin{center}
120: $\deg \gcf (f,g)>0$ \quad or \quad $\deg(f)<d$ and $\deg(g)<p{+}1$.
121: \end{center}
122: %%%%%%%%%%%%%%
123: That is, $f$ and $g$ either have a common root in $\C$ or else a common root at
124: $\infty$, and thus define a rational function $f/g$ of degree less than $p{+}1$.
125:
126: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
127: \begin{prop}\label{P9:Weak_cont}
128: Let $\{\varphi_j\mid j\in\N\}\subset \St_{p{+}1}\setminus Z$ be a sequence of points that
129: converges to some point $\varphi=(f,g)\in Z$.
130: Let $z_1,\dotsc,z_k$ be the common roots of $f$ and $g$ (including $\infty$ if
131: $\deg(f)$ and $\deg(g)$ are both less than $d$).
132: Then the sequence of functions $\{\pi(\varphi_j)\mid j\in\N\}$ converges to
133: $\pi(\varphi)$ uniformly on compact subsets of\/ $\P^1\setminus\{z_1,\dotsc,z_k\}$.
134: \end{prop}
135: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
136:
137: We give an elementary proof of this proposition.
138: Let $K\subset\P^1$ be a compact subset disjoint from the common roots
139: $\{z_1,\dotsc,z_k\}$ of $f$ and $g$.
140: We may cover $\P^1$ by the standard affine charts $\C_0$ and $\C_\infty$ whose
141: coordinates are $t$ and $1/t$, respectively.
142: Then $K=K_0\cup K_\infty$, where $K_0\subset\C_0$ and $K_\infty\subset\C_\infty$ are
143: compact subsets of the two affine charts.
144: It suffices to show that the sequence of functions $\{\pi(\varphi_i)\}$ converges
145: uniformly to $\pi(\varphi)$ on each set $K_0$ and $K_\infty$.
146:
147: Now $K_0$ is itself covered by compact sets $K_0^f$ and $K_0^g$, where $K_0^f$ contains
148: no root of $f$ and $K_0^g$ contains no root of $g$.
149: Removing finitely many members of the sequence
150: $\{\varphi_i=(f_i,g_i)\mid i\in\N\}$, we may assume that no $f_i$ has a root in $K_0^f$
151: and no $g_i$ has a root in $K_0^g$.
152: As $(f_i,g_i)$ converges to $(f,g)$ in $\St_{p{+}1}$, and no $g_i$ has a root
153: in $K_0^g$, both sequences of functions
154: \[
155: \{ f_i(t)\mid i\in\N\}\qquad\mbox{and}\qquad
156: \{(g_i(t))^{-1}\mid i\in\N\}
157: \]
158: are uniformly bounded in $K_0^g$.
159: Therefore, the sequence of functions
160: \[
161: \frac{f_i(t)}{g_i(t)}\ \colon\ K_0^g\ \longrightarrow\ \C
162: \]
163: is uniformly bounded and converges pointwise on the compact set $K_0^g$ to
164: $f(t)/g(t)$.
165: Thus this convergence is uniform on $K_0^g$, and it remains
166: uniform under $\C\xrightarrow{\,\sim\,}\C_0\subset\P^1$.
167: The same arguments work for $K_0^f$, as well as $K_\infty$, which proves the proposition.
168: \hfil\QED
169: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
170:
171: This proposition is half of the engine of this proof
172: of Eremenko and Gabrielov.
173: The other half is the asymptotic proof of Shapiro's conjecture, Theorem~\ref{T8:asymptotic}.
174: \medskip
175:
176: We now explain how to associate an embedded graph with distinguished vertices to each real
177: rational function.
178: Let \DeCo{$R_{p{+}1}$} be the set of nonconstant real rational functions of degree at
179: most $p{+}1$, all of whose critical points are real.
180: If $\rho\in R_{p{+}1}$, then $\rho^{-1}(\R\P^1)\subset\P^1$ defines an embedded (multi-)
181: graph $\Gamma$ with the following properties:
182: %%%%%%%%%%%%%%%%%%
183: \begin{enumerate}
184: \item[(i)] $\Gamma$ is stable under complex conjugation and
185: $\R\P^1\subset \Gamma$.
186: \end{enumerate}
187: %%%%%%%%%%%%%%%%%%
188: Call any edge in $\Gamma\setminus\R\P^1$ an \DeCo{{\sl interior edge}}.
189: %%%%%%%%%%%%%%%%%%
190: \begin{enumerate}
191: \item[(ii)] The vertices of $\Gamma$ lie on $\R\P^1$ and are the critical points
192: of the rational function $\rho$.
193: The valence of a vertex is even and it equals twice the order of ramification of $\rho$
194: at the critical point, which we call the \DeCo{{\sl local degree}} of $\Gamma$ at the vertex.
195: \end{enumerate}
196: %%%%%%%%%%%%%%%%
197: The set-theoretic difference $\P^1\setminus\Gamma$ is a union of $2d$ cells,
198: where $d$ is the degree of $\rho$.
199: The closure of each cell is homeomorphic to a disc, and the boundary of each
200: cell maps homeomorphically to $\R\P^1$.
201: This is because the cells (and their closures) are the inverse images of one of
202: the two discs in $\P^1\setminus \R\P^1$ (or their closures), and there are no
203: critical points in the interior of any cell.
204: We deduce the following additional property of these multi-graphs.
205: %%%%%%%%%%%%%%%%%%
206: \begin{enumerate}
207: \item[(iii)] No interior edge of $\Gamma$ can begin and end at the same vertex.
208: \end{enumerate}
209: %%%%%%%%%%%%%%%%
210: Indeed, if an interior edge $e$ begins and ends at the same vertex, then $r(e)=\R\P^1$ as $v$ is
211: the only critical point on $e$.
212: But then $e$ must be the boundary of any cell adjacent to $e$, which implies that
213: $\Gamma$ consists of only two cells and one edge $e$ and so $\rho$ has degree 1,
214: and in fact $e$ was not an interior edge after all.
215:
216:
217: Here are three pictures of such embedded (multi-) graphs for quintic rational functions with
218: evenly spaced critical points.
219: We have drawn $\R\P^1$ as a circle with the upper half plane in its interior.
220: The point $\sqrt{-1}$ is at the center of the circle,
221: $-\sqrt{-1}$ is the point at infinity, and complex
222: conjugation is inversion in the circle.
223: \[
224: \includegraphics[height=80pt]{figures/9/G26.1.eps}\qquad
225: \includegraphics[height=80pt]{figures/9/G26.5.eps}\qquad
226: \includegraphics[height=80pt]{figures/9/G26.7.eps}
227: \]
228:
229: We seek to analytically continue rational functions whose Wronskians
230: lie in a curve of polynomials $\{\Phi_z(t)\mid z\in[0,1]\}$ where each
231: $\Phi_z(t)$ has degree $2p$ with distinct real roots
232: $s_1(z), s_2(z), \dotsc,s_{2p}(z)$, and where each $s_i$ is a continuous
233: function of $z$.
234: The vertices of the graph $\rho^{-1}(\R\P^1)$ associated to a rational function $\rho$ with
235: Wronskian $\Phi_z(t)$ are labeled by these roots, or equivalently by the numbers
236: $1,2,\dotsc,2p$.
237: Since the relative order of these roots $s_1(z), s_2(z), \dotsc,s_{2p}(z)$
238: does not change as $z$ varies (because each polynomial $\Phi_z(t)$ has distinct
239: roots), we may capture this information by labeling only one root, say $s_1(z)$
240: (which is a vertex of the corresponding graph), and assuming
241: that the roots are ordered in a manner consistent with a fixed orientation of
242: $\R\P^1$.
243: It is these labeled graphs that we wish to consider up to isotopy (deformation
244: in $\P^1$).
245:
246:
247: \begin{defn}
248: A \DeCo{{\sl net}} is an (isotopy) equivalence
249: class of such embedded multi-graphs in $\P^1$ satisfying (i), (ii), and
250: (iii), with a distinguished vertex.
251: \end{defn}
252:
253: Here are the five nets with 6 vertices, each with local degree 2 at every vertex.
254: These correspond to rational functions of degree four with simple ramification.
255: %
256: \begin{equation}\label{Eq:quartic_nets}
257: \raisebox{-30pt}{%
258: \includegraphics[height=65pt]{figures/9/G25.1.eps}\quad
259: \raisebox{-7.5pt}{\includegraphics[height=80pt]{figures/9/G25.3.eps}}\quad
260: \includegraphics[height=65pt]{figures/9/G25.2.eps}\quad
261: \raisebox{4.2pt}{\includegraphics[height=65pt]{figures/9/G25.4.eps}}\quad
262: \raisebox{-9.8pt}{\includegraphics[height=65pt]{figures/9/G25.5.eps}}}
263: \end{equation}
264: %
265:
266: The uniform convergence of Proposition~\ref{P9:Weak_cont} implies a certain continuity
267: of nets.
268: Two subsets $X,Y\subset\P^1$ have \DeCo{{\sl Hausdorff distance}} $\epsilon$ if every point of
269: $X$ lies within a distance $\epsilon$ of $Y$ and vice-versa.
270: This gives the \DeCo{{\sl Hausdorff metric}} on subsets of $\P^1$.
271:
272: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
273: \begin{prop}\label{P9:Conv_nets}
274: Let $\{\varphi_j\}\subset\St_{p{+}1}$ be a convergent sequence with limit $\varphi$.
275: Then the sets $\{\pi(\varphi_j)^{-1}(\R\P^1)\}$ converge in the Hausdorff metric to
276: the set $\{\pi(\varphi)^{-1}(\R\P^1)\}$.
277: \end{prop}
278: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
279:
280: We deduce two corollaries from this proposition.
281:
282: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
283: \begin{cor}\label{C9:analytic_continuation}
284: Suppose that $\{\rho_z\mid z\in[0,1]\}$ is a continuous path in $R_{2p}$ where each
285: $\rho_z$ has $2p$ critical points.
286: Let $v_1(z)$ be a continuous function of $z$ which is equal to a critical point
287: of $\rho_z$, for each $z$.
288: Then the net of the pair
289: \[
290: \bigl(\rho_z^{-1}(\R\P^1),\, v_1(z)\bigr)
291: \]
292: does not depend upon $z$.
293: \end{cor}
294: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
295:
296: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
297: \begin{cor}\label{C9:create_edge}
298: Suppose that $\{\varphi_z\mid z\in[0,1]\}$ is a continuous path in the Stiefel
299: manifold $\St_{2p}$.
300: Suppose that for $0<z$ we have $\pi(\varphi_z)\in R_{2p}$ and $\pi(\varphi_z)$ has distinct
301: critical points $v_1(z),v_2(z),\dotsc,v_n(z)$,
302: where $v_i(z)$ is a continuous function of $z$,
303: and also that, at $z=0$, we have $v_1(0)=v_2(0)$, but all other
304: critical points are distinct.
305: Then the degree of $\pi(\varphi_z)$ is constant for $z\in(0,1]$, \Blue{and}
306: $\deg(\pi(\varphi_0))<\deg(\pi(\varphi_1))$ if and only if the net of $\pi(\varphi_1)$
307: has an interior edge between $v_1(1)$ and $v_2(1)$.
308: \end{cor}
309: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
310:
311: Here are two nets for quartic rational functions~\eqref{Eq:quartic_nets}
312: as two of their vertices collide.
313: \[
314: \begin{picture}(286,165)
315: \put(0,88){\begin{picture}(281,77)(0,-2)
316: \put(0,0){\includegraphics{figures/9/N21.eps}}
317: \put(3,18){$v_2$} \put(33,1){$v_1$}
318: \put(75,35){$\longrightarrow$}
319: \put(110,1){\includegraphics{figures/9/N22.eps}}
320: \put(124,5){$v_2$} \put(153,1){$v_1$}
321: \put(188,35){$\longrightarrow$}
322: \put(220,9){\includegraphics{figures/9/N23.eps}}
323: \put(249,1){$v_1$}
324: \end{picture}}
325: % fig2dev -Leps -m0.4 N11.fig N11.eps
326: % fig2dev -Leps -m0.4 N12.fig N12.eps
327: % fig2dev -Leps -m0.4 N13.fig N13.eps
328: % fig2dev -Leps -m0.4 N21.fig N21.eps
329: % fig2dev -Leps -m0.4 N22.fig N22.eps
330: % fig2dev -Leps -m0.4 N23.fig N23.eps
331: \put(5,0){\begin{picture}(281,77)(0,-2)
332: \put(0,0){\includegraphics{figures/9/N11.eps}}
333: \put(4,8){$v_2$} \put(28,1){$v_1$}
334: \put(70,35){$\longrightarrow$}
335: \put(107,0){\includegraphics{figures/9/N12.eps}}
336: \put(115,9){$v_2$} \put(140,1){$v_1$}
337: \put(182,35){$\longrightarrow$}
338: \put(218,0){\includegraphics{figures/9/N13.eps}}
339: \put(257,2){$v_1$}
340: \end{picture}}
341: \end{picture}
342: \]
343: In the first row, there is an edge (in fact two) of $\Gamma\setminus\R\P^1$ between the
344: vertices $v_2$ and $v_1$.
345: This edge collapses in the limit as $v_2$ approaches $v_1$, which causes two regions of
346: $\P^1\setminus\Gamma$ to collapse, so that the limiting net has two fewer regions in its
347: complement, and thus corresponds to a rational function of degree $3$.
348: There is no such edge in the nets of the second row, and the limiting net still has eight
349: regions and thus its rational function still has degree four.\medskip
350:
351: \noindent{\it Proof of Corollary$~\ref{C9:create_edge}$.}
352: The degree of a rational function $\rho\in R_{2p}$ is one-half the number of
353: cells in the complement $\P^1\setminus \rho^{-1}(\R\P^1)$ of the net of $\rho$.
354: Set $\rho_z:=\pi(\varphi_z)$.
355: The only way for the number of cells in the complement of the net of $\rho_z$ to
356: change at some $z_0\in[0,1]$ would be if some edge of $\rho_z^{-1}(\R\P^1)$
357: collapsed as $z\to z_0$.
358: Since the vertices of $\rho_z^{-1}(\R\P^1)$ are the critical points
359: $v_1(z),\dotsc,v_n(z)$, which are distinct for $z\in(0,1]$,
360: we see that the degree of $\rho_z$ is constant for $z\in(0,1]$.
361:
362: If the degree of $\rho_0$ is less than that of $\rho_1$, then some edges which bound a cell must
363: disappear in the limit as $z\to 0$.
364: But this can only happen if a cell is bounded by edges between $v_1(z)$ and $v_2(z)$,
365: as they are the only critical points which collide in the limit as $z\to 0$.
366: By (iii), such a cell must be bounded by more than one edge which implies that there was
367: an interior edge between $v_1(1)$ and $v_2(1)$ outside of $\R\P^1$.
368: This shows the necessity of such an interior edge between $v_1(1)$ and $v_2(1)$ for the degree
369: to drop.
370:
371: For sufficiency, note that if there is an interior edge between $v_1(1)$ and $v_2(1)$, then it
372: must collapse in the limit as $z\to 0$ for otherwise
373: condition (iii) for nets would be violated.
374: \QED
375:
376:
377: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
378: %
379: \subsection{Schubert induction for rational functions}
380: In Chapter~\ref{Ch:ScGr}, we used Schubert induction to construct a sequence of points
381: $s_1,\dotsc,s_{mp}\in\R$ and sufficiently many real points in each
382: Schubert variety $X_\alpha\Fdot(0)$ which also lie in
383: $X_\I\Fdot(s_i)$ for $i=1,\dotsc,|\alpha|$.
384: Without re-running that proof, we will describe what that
385: construction gives for rational functions.
386:
387: The construction of Theorem~\ref{T8:asymptotic} relevant for rational functions was
388: in the Grassmannian $\Gr(p,p{+}2)$.
389: Under the Grassmann duality of Theorem~\ref{T8:GrassmannDuality}, this becomes a
390: construction in $\Gr(2,\C_{p+1}[t])$ and involves a Schubert variety
391: $X_\alpha\Edot(t)$ with $\alpha\in\binom{[p{+}2]}{2}$.
392: In fact, the statements become identical after replacing $\Fdot(t)$ by $\Edot(t)$.
393: We will briefly recall these statements in this setting.
394:
395: A point in the Schubert cell $X^\circ_\alpha\Edot(s)$ where $\alpha\colon\alpha_1<\alpha_2$
396: is a two-dimensional subspace $\langle f,g\rangle$ of polynomials of degree $p{+}1$ such
397: that
398: %
399: \begin{equation}\label{eq:ord}
400: \ord_s(f)\ =\ p{+}2{-}\alpha_2
401: \qquad\mbox{and}\qquad
402: \ord_s(g)\ =\ p{+}2{-}\alpha_1\,.
403: \end{equation}
404: %
405: In particular,
406: \[
407: (t-s)^{p+2-\alpha_2}|| f
408: \qquad\mbox{and}\qquad
409: (t-s)^{p+2-\alpha_1}|| g\,.
410: \]
411: (Here, $a^k||b$ means that $a^k$ divides $b$, but $a^{k+1}$ does not divide $b$.)
412:
413: Given $\langle f,g\rangle\in X^\circ_\alpha\Edot(s)$, a consequence of~\eqref{eq:ord} is
414: that
415: \[
416: \Wr(f,g)\ =\ f'(t)g(t)-f(t)g'(t)
417: \]
418: vanishes to order $p{+}1{-}\alpha_2{+}p{+}2{-}\alpha_1=2p{-}|\alpha|$ at $s$.
419:
420: By Lemma~\ref{L8:schubert_induction}, there exist real numbers $s_1,\dotsc,s_{2p}$
421: (in fact, we have $s_1>\dotsb>s_{2p}>0$) such that for all
422: $\alpha\in\binom{[p{+}2]}{2}$, the intersection
423: %
424: \begin{equation}\label{Eq9:Inductive_step}
425: X_{\alpha}\Edot(0)\ \cap\ \bigcap_{i=1}^{|\alpha|} X_\I\Edot(s_i)
426: \end{equation}
427: %
428: is transverse, and it consists of $\delta(\alpha)$ real points.
429: Any point $\langle f,g\rangle$ in the intersection~\eqref{Eq9:Inductive_step} will have
430: Wronskian
431: \[
432: f'(t)g(t)-f(t)g'(t)\ =\
433: \mbox{constant}\cdot t^{2p-|\alpha|}\cdot\prod_{i=1}^{|\alpha|}(t-s_i)
434: \]
435:
436: In fact, as noted in Remark~\ref{R8:construct}, the proof of
437: Lemma~\ref{L8:schubert_induction} did much more.
438: Suppose that $|\alpha|>0$, and define
439: \[
440: \DeCo{\beta^1}\ :=\ \alpha_1{-}1<\alpha_2
441: \qquad\mbox{and}\qquad
442: \DeCo{\beta^2}\ :=\ \alpha_1<\alpha_2{-}1\,,
443: \]
444: when possible.
445: ($\beta^1$ is only defined if $1<\alpha_1$ and $\beta^2$ is only defined if
446: $\alpha_1{+}1<\alpha_2$)
447: Then the proof constructed $\delta(\alpha)=\delta(\beta^1)+\delta(\beta^2)$
448: families $\{\langle f_z,g_z\rangle\mid z\in[0,s_{|\alpha|}]\}$ of polynomials such that
449:
450: \begin{enumerate}
451: \item For $z\neq 0$, $(f_z,g_z)\in X^\circ_\alpha$.
452: \item ${\displaystyle f_z'(t)g_z(t)-f_z(t)g_z'(t)\ =\
453: \mbox{constant}\cdot t^{2p-|\alpha|}\cdot \Bigl(\
454: \prod_{i=1}^{|\alpha|-1}(t-s_i)\Bigr)\cdot
455: (t-z)}$.
456: \item Exactly $\delta(\beta^i)$ of these families began in $X_{\beta^i}\Edot(0)$.
457: That is, for $\delta(\beta^i)$ of these families, we have
458: $(f_0,g_0)\in X_{\beta^i}\Edot(0)$.
459: \end{enumerate}
460:
461:
462: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
463: \subsection{Schubert induction for nets}
464:
465: The main idea in the proof is that the rational functions constructed in
466: Lemma~\ref{L8:schubert_induction} each have different nets.
467:
468: \begin{thm}\label{T9:Nets_Diff}
469: The $\delta(\alpha)$ rational functions in $X^\circ_\alpha\Edot(0)$ constructed in
470: Lemma~$\ref{L8:schubert_induction}$ each have
471: different nets.
472: \end{thm}
473:
474: Suppose that $\langle f,g\rangle$ is a point in the intersection~\eqref{Eq9:Inductive_step}
475: where $f$ and $g$ satisfy~\eqref{eq:ord} for $s=0$.
476: Then its Wronskian vanishes to order $2p{-}|\alpha|$ at $0$ and to order 1 at the points
477: $s_1,\dotsc,s_{|\alpha|}$.
478: In particular, 0 is the only common zero of $f$ and $g$.
479: Removing the common factor $t^{p+2-\alpha_2}$ from both $f$ and $g$ gives relatively prime
480: polynomials of degree at most $\alpha_2{-}1$.
481: Indeed, if $f$ and $g$ had a common root $s$, then a linear combination of them would vanish
482: to order at least 2 and so their Wronskian would vanish to order at least $2$ at $s$.
483: Then the rational function $\DeCo{r}:=f/g$ has degree $\alpha_2{-}1$ with Wronskian
484: \[
485: \mbox{constant}\cdot t^{\alpha_2-\alpha_1-1}\cdot \prod_{i=1}^{|\alpha|}(t-s_i)\,.
486: \]
487:
488: The point $\langle f,g\rangle$ corresponds to a unique path in the Bruhat order from
489: $12$ to $\alpha$ in the Bruhat order.
490: We claim that this path may be recovered from the net $\rho^{-1}(\R\P^1)$ of $\rho$.
491:
492: Indeed, consider the $i$th step in the construction, when the critical point $s_i$ was
493: created.
494: By Corollary~\ref{C9:create_edge}, the interior edge from $s_i$ has other endpoint $0$ if the
495: degree of the rational function increased at the $i$th step, and if its degree did not
496: increase, then the other endpoint of that edge is at some critical point $s_k$ with $s_k>s_i$
497: and so $k<i$.
498: Subsequent steps in the construction will not affect an edge from $s_i$ to $s_k$ with $k<i$,
499: but an edge between $0$ and $s_i$ may be moved to an edge between $s_i$ and $s_j$ where $j>i$.
500:
501: Thus, the degree of the rational function increased at step $i$ if and only if the other
502: endpoint of an interior edge from $s_i$ is at $s_j$ with $j>i$.
503: If $\beta\lessdot\beta'$ is the $i$th step in the chain corresponding to our rational function
504: $\rho$, then
505: \begin{enumerate}
506: \item $\beta_2+1=\beta'_2$, so the degree of the rational function increased, if the interior
507: edge from $s_i$ has endpoint $s_j$ with $j>i$ (so $s_j<s_i$), and
508: \item $\beta_1+1=\beta'_1$, so the degree of the rational function did not increase, if the
509: interior edge from $s_i$ has endpoint $s_k$ with $k<i$ (so $s_k>s_i$).
510: \end{enumerate}
511: This completes the proof.
512: \QED\smallskip
513:
514: Figure~\ref{F9:nets} illustrates the formation of the nets during the Schubert induction
515: for quartic rational functions, as well as the recursion for $\delta(\alpha)$.
516:
517: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
518: \begin{figure}[!ht]
519: \[
520: \begin{picture}(420,585)(4,-17)
521: \thicklines
522:
523: \put(335,-17){
524: \begin{picture}(80,160)
525: \put(-2,0){\includegraphics{figures/9/Y4.eps}}
526: % fig2dev -Leps -m0.23 Y4.fig Y4.eps
527: \put( 0,150){45}
528: \put(25,125){35}
529: \put( 0,100){35} \put(50,100){25}
530: \put(25, 75){24} \put(75, 75){15}
531: \put( 0, 50){23} \put(50,50){14}
532: \put(25, 25){13}
533: \put( 0, 0){12}
534: \end{picture}
535: }
536:
537:
538: \put( 54,-13){\includegraphics{figures/9/C0.eps}}%
539: \put(85, 29){$s_1$}
540: \put(99, 8){$s_2$}
541: \put(86,-16){$s_3$}
542: \put(55,-16){$s_4$}
543: \put(42, 8){$s_5$}
544: \put(53, 29){$s_6$}
545:
546: \put( 94,24){\Blue{\vector(3,2){50}}}
547:
548: \put(147,57){\includegraphics{figures/9/C1.eps}}%
549:
550: \put(142,95){\Blue{\vector(-1,1){40}}}
551:
552: \put(187,95){\Blue{\vector(3,2){60}}}
553:
554: \put( 67.5,137){\includegraphics{figures/9/C21.eps}}%
555: \put(249,137){\includegraphics{figures/9/C22.eps}}%
556:
557: \put(102,185){\Blue{\vector(1,2){19}}}
558: \put(252,185){\Blue{\vector(-1,1){42}}}
559: \put(298,185){\Blue{\vector(3,2){69}}}
560:
561: \put(108,228.75){\includegraphics{figures/9/C31.eps}}%
562: \put(169,228){\includegraphics{figures/9/C32.eps}}%
563:
564: \put(372,233.9){\includegraphics{figures/9/C33.eps}}%
565:
566: \put(114,285){\Blue{\vector(-4,3){56}}}%
567: \put(173,285){\Blue{\vector(-4,3){56}}}%
568: \put(151,285){\Blue{\vector(4,3){56}}}%
569: \put(210,285){\Blue{\vector(4,3){56}}}%
570:
571: \put(387,285){\Blue{\vector(-1,2){21}}}%
572:
573: \put( 16,327){\includegraphics{figures/9/C41.eps}}%
574: \put( 77,327){\includegraphics{figures/9/C42.eps}}%
575:
576: \put(190,327){\includegraphics{figures/9/C43.eps}}%
577: \put(251,327){\includegraphics{figures/9/C44.eps}}%
578: \put(312,327){\includegraphics{figures/9/C45.eps}}%
579:
580: \put( 39,384){\Blue{\vector(1,2){18}}}%
581: \put(100,384){\Blue{\vector(1,2){18}}}%
582:
583: \put(193,384){\Blue{\vector(-1,2){18}}}%
584: \put(254,384){\Blue{\vector(-1,2){18}}}%
585: \put(315,384){\Blue{\vector(-1,2){18}}}%
586:
587: \put( 30,423){\includegraphics{figures/9/C51.eps}}%
588: \put( 91,419){\includegraphics{figures/9/C52.eps}}%
589: \put(152,418){\includegraphics{figures/9/C53.eps}}%
590: \put(213,419){\includegraphics{figures/9/C54.eps}}%
591: \put(274,423){\includegraphics{figures/9/C55.eps}}%
592:
593: \put( 51,472){\Blue{\vector(-1,2){18}}}%
594: \put(112,472){\Blue{\vector(-1,2){18}}}%
595: \put(173,472){\Blue{\vector(-1,2){18}}}%
596: \put(234,472){\Blue{\vector(-1,2){18}}}%
597: \put(295,472){\Blue{\vector(-1,2){18}}}%
598:
599: \put( 4,510){\includegraphics{figures/9/C61.eps}}%
600: \put( 65,507){\includegraphics{figures/9/C62.eps}}%
601: \put(126,507){\includegraphics{figures/9/C63.eps}}%
602: \put(187,509){\includegraphics{figures/9/C64.eps}}%
603: \put(248,513){\includegraphics{figures/9/C65.eps}}%
604: %\thinlines\put(20,335){\line(1,0){390}}
605: \end{picture}
606: \]
607: \caption{Formation of nets during Schubert induction.}\label{F9:nets}
608: \end{figure}
609: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
610:
611: We complete the proof of the Shapiro conjecture for rational curves.
612:
613: \begin{thm}
614: Let $\Phi(t)$ be a real polynomial of degree $2p$ whose roots are all real.
615: Then there are exactly $\delta(p{+}1,p{+}2)$ real equivalence classes of rational
616: functions with Wronskian $\Phi(t)$.
617: \end{thm}
618:
619: \noindent{\it Proof.}
620: Let $s_1>s_2>\dotsb>s_{2p}\in\R$ be the numbers such that
621: the intersection
622: \[
623: \bigcap_{i=1}^{2p} X_\bI\Edot(s_i)
624: \]
625: transverse with all points real.
626: Each point in the intersection is an equivalence class of rational functions with
627: Wronskian
628: \[
629: \Phi_0(t)\ =\ \prod_{i=1}^{2p}(t-s_i)\,.
630: \]
631:
632: Let $\{\Phi_z\mid z\in[0,1]\}$ be a continuous family of polynomials of degree $2p$ all
633: with distinct real roots and with $\Phi_1(t)=\Phi(t)$.
634: We attempt to analytically continue each point in the fiber $\Wr^{-1}(\Phi_z)$ from
635: $z=0$ to $z=1$.
636: The only way this continuation could fail would be if it encountered a fiber
637: $\Wr^{-1}(\Phi_z)$ with a multiple point.
638: Then two of the rational functions in this fiber would have to coincide.
639: In particular two would have the same net, (where we have labeled the
640: nets by the root of the $\Phi_z(t)$ corresponding to $s_1$).
641: This implies that two of the original rational functions in $\Wr^{-1}(\Phi_0)$ have
642: the same net, by Corollary~\ref{C9:analytic_continuation}.
643: But this contradicts Theorem~\ref{T9:Nets_Diff}.
644: \QED
645:
646: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
647: %
648: %
649: \section{Rational functions with prescribed coincidences}
650: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
651: The results of Section~\ref{S:SCRF} can be used to prove a result about real rational functions
652: that satisfy a certain interpolation condition.
653: This is is due to Eremenko, Gabrielov, Shapiro, and Vainstein~\cite{EGSV}, and
654: may be interpreted in the Grassmannian $\Gr(p,p{+}2)$ as an appealing
655: generalization of the Shapiro conjecture.
656: We will discuss this generalization, the Secant Conjecture, in Chapter~\ref{Ch:Frontier}.
657:
658: We first describe the interpolation problem.
659: Let $A_1,\dotsc,A_n$ be disjoint finite subsets of $\P^1$ where the set $A_i$ has $1+a_i$
660: elements with $1\leq a_i\leq p$ and $a_1+\dotsb+a_n=2p$.
661: Write $\ba$ for this sequence $(a_1,\dotsc,a_n)$ of numbers.
662: The interpolation problem is to determine the equivalence classes of rational functions $\rho$ of
663: degree $p{+}1$ that satisfy
664: \[
665: \rho|_{A_i}\mbox{ is constant for }i=1,\dotsc,n\,.
666: \]
667: There are in fact finitely many such equivalence classes of rational functions when the sets
668: $A_i$ are general.
669: We will later prove this finiteness and show the number of equivalence classes is a Kostka
670: number $K_\ba$~\cite[p.25]{Fu97},\cite[I,6]{Mc95}.
671: Heuristically, there are finitely many equivalence classes because the
672: condition that a rational function is constant on a finite set of $1{+}a$ elements gives $a$
673: equations in the Stiefel coordinates for rational functions.
674:
675:
676: A collection of sets $A_i\subset\R\P^1$ for $i=1,\dotsc,n$ is \DeCo{{\sl separated}} if there
677: exist disjoint intervals $I_1,\dotsc,I_n$ of $\R\P^1$ with $A_i\subset I_i$ for
678: $i=1,\dotsc,n$.
679:
680:
681: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
682: \begin{thm}[\cite{EGSV}]\label{T9:EGSV}
683: Let $\ba=(a_1,\dotsc,a_n)$ with $1\leq a_i\leq p$ and $a_1+\dotsb+a_b=2p$.
684: For general separated subsets of $\R\P^1$, $A_1,\dotsc,A_n$ with
685: $|A_1|=1{+}a_i$, there are exactly $K_{\ba}$ real equivalence classes of rational
686: functions $\rho$ such that
687: %
688: \begin{equation}\label{E9:constant}
689: \rho|_{A_i}\mbox{ is constant for }i=1,\dotsc,n\,.
690: \end{equation}
691: %
692: \end{thm}
693: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
694:
695: Given separated subsets $A_1,\dotsc,A_n$ of $\R\P^1$, the proof first constructs such a
696: real rational function satisfying~\eqref{E9:constant} for every net with a certain
697: property, described below~\eqref{E9:hypothesis}.
698: Next, it relates this interpolation problem to a problem in the Schubert calculus that has
699: $K_\ba$ solutions, and finally shows that $K_\ba$ is the number of nets with the
700: property~\eqref{E9:hypothesis}.
701: This will imply that we have constructed all the solutions for general sets $A_i$.
702:
703: Theorem~\ref{T9:EGSV} generalizes Theorem~\ref{T9:EG}.
704: Suppose that we have subsets $\{A_z\mid z\in(0,1]\}$ of $\R\P^1$ depending continuously on
705: $z$, each of cardinality $a+1$, whose limit as $z\to 0$ consists of a single point,
706: \[
707: \lim_{z\to 0} A_z\ =\ \{s\}\,.
708: \]
709: Suppose further that we have a family $\{\rho_z\mid z\in[0,1]\}$ of rational functions that
710: depends continuously on $z$, and such that for $z>0$, $\rho_z$ is constant on $A$.
711: Then $\rho_0$ will have a critical point at $s$ of order at least $a$.
712:
713:
714: In this way, Theorem~\ref{T9:EGSV} implies Theorem~\ref{T9:EG} by simply considering the limit
715: as the points in each set $A_i$ collide.
716: In fact, this analysis will enable us to deduce a stronger form of Theorem~\ref{T9:EG}
717:
718: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
719: \begin{thm}\label{T9:Pieri}
720: Let $a_1,\dotsc,a_n$ be positive integers with $1\leq a_i\leq p$ and $a_1+\dotsb+a_n=2p$.
721: Then every rational function of degree $p+1$ with $n$ real critical points of multiplicities
722: $a_1,\dotsc,a_n$ is real.
723: There are exactly $K_\ba$ classes of such rational functions, and the corresponding Schubert
724: varieties meet transversally.
725: \end{thm}
726: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
727:
728: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
729: %
730: \subsection{Proof of Theorems~\ref{T9:EGSV} and~\ref{T9:Pieri}}
731:
732: Let $\calR_{p+1}$ be the set of rational function of degree $p{+}1$ with exactly $2p$ critical
733: points of multiplicity 1.
734: We use two consequences of the work in Section~\ref{S:SCRF}.
735: \begin{enumerate}
736: \item If $\rho_1$ and $\rho_2$ are rational functions in $\calR_{p+1}$ with the same critical points
737: and isotopic nets (where we use the same critical point for each net), then
738: $\rho_1$ is equivalent to $\rho_2$.
739:
740: \item For every net $\Gamma\subset\P^1$ with a given vertex set $V$ (and distinguished vertex
741: $v_1\in V$), there is a unique equivalence class of rational functions in $\calR_{p+1}$ with
742: critical set $V$ and net (with distinguished vertex $v_1$) isotopic to $\Gamma$.
743: \end{enumerate}
744:
745: Actually, we only showed the second point for nets with local degree 2 at each critical
746: point.
747: The stronger statement follows from results in~\cite{EG02}.
748: To prove Theorem~\ref{T9:EGSV}, fix separated subsets $A_1,\dotsc,A_n$ of $\R\P^1$
749: satisfying the hypotheses.
750: Choose $2p$ additional points $s_1,\dotsc,s_{2p}$ where, for each $i$, $a_i$ of the points
751: interlace the $a_i{+}1$ points of $A_i$.
752: Figure~\ref{F9:interlace} shows an example when $p=5$ and $\ba=(3,2,2,2,1)$.
753: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
754: \begin{figure}[htb]
755: \[
756: \begin{picture}(152,136)(-2,1)
757: \put( 15, 13){\includegraphics{figures/9/G27.interlace.eps}}
758: \put( 49.5, 73){$s_1,\dotsc,s_{10}$}
759: \put(137, 75){$A_1$}
760: \put(108,126){$A_2$}
761: \put( 21,125){$A_3$}
762: \put( -2, 73){$A_4$}
763: \put( 70, 2){$A_5$}
764: \end{picture}
765: \]
766: \caption{Interlacing points.}
767: \label{F9:interlace}
768: \end{figure}
769: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
770: Each point $s_j$ lies between two points of some set $A_i$.
771: Write $[x_j,y_j]$ for the interval that contains $s_j$ and note that $x_j,y_j\in A_i$,
772: for some $i$.
773:
774: Consider nets with the vertices $s_1,\dotsc,s_{2p}$ that have local degree $2$ at each
775: vertex and satisfy the additional hypothesis:
776: %
777: \begin{equation}\label{E9:hypothesis}
778: \mbox{There are no edges between points $s_j$ and $s_k$ that interlace the same set
779: $A_i$}
780: \end{equation}
781: %
782: There are five nets satisfying~\eqref{E9:hypothesis} for the points $A_i$ of
783: Figure~\ref{F9:interlace} (we only draw the edges in the upper half plane, which is the
784: interior of the circle).
785: %
786: \begin{equation}\label{Eq9:interpolate_nets}
787: \raisebox{-35pt}{
788: \includegraphics[height=70pt]{figures/9/G27.interpolate.1.eps}\quad
789: \includegraphics[height=70pt]{figures/9/G27.interpolate.2.eps}\quad
790: \includegraphics[height=70pt]{figures/9/G27.interpolate.3.eps}\quad
791: \includegraphics[height=70pt]{figures/9/G27.interpolate.4.eps}\quad
792: \includegraphics[height=70pt]{figures/9/G27.interpolate.5.eps}}
793: \end{equation}
794: %
795:
796: Suppose that we have critical points $s_1,\dotsc,s_{2p}$ interlacing the sets
797: $A_i$, where all the points $s_i$ are fixed and chosen arbitrarily within their intervals,
798: except for one, $s_j$, which is allowed to vary within its interval $[x_j,y_j]$.
799: Also fix an isotopy class $\Gamma$ of nets satisfying~\eqref{E9:hypothesis}.
800: For each $s\in[x_j,y_j]$ let $\rho_s\in\calR_{p+1}$ be a rational function with the critical
801: points $s_1,\dotsc,s_{2p}$ and net $\Gamma$.
802: We may suppose that $\{\rho_s\mid s\in[x_j,y_j]\}$ is a continuous family.
803:
804: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
805: \begin{lemma}\label{L9:interpolate}
806: There exists a point $s_j\in[x_j,y_j]$ such that $\rho_{s_j}(x_j)=\rho_{s_j}(y_j)$.
807: \end{lemma}
808: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
809:
810: To see this, we may assume that $\rho_s$ is normalized so that $\rho_s(s)=0$ and $\rho_s$ maps the
811: interior edge of $\Gamma$ terminating in $s_j$ to the interval $[-\infty,0]$ in $\R\P^1$.
812: Then the difference
813: \[
814: \rho_s(x_j)\ -\ \rho_s(y_j)
815: \]
816: is positive when $s$ is near $y_j$ and negative when $s$ is near $x_j$, so it changes sign
817: on the interval $[x_j,y_j]$ and therefore takes the value zero at some $s_j\in[x_j,y_j]$.
818: We give a picture below.
819: The arrows point in the direction of increase of $\rho_s(t)$ for $t$ in the net.
820: \[
821: \begin{picture}(120,76)(0,-13)
822: \put(0,0){\includegraphics{figures/9/IVTII.eps}}
823: \put(80,3){$x_j$} \put(80,60){$y_j$}
824: \put(99,53){$s$}
825: \put(0,-13){$\rho_s(x_j)-\rho_s(y_j)>0$}
826: \end{picture}
827: \qquad
828: \begin{picture}(120,76)(0,-13)
829: \put(0,0){\includegraphics{figures/9/IVTI.eps}}
830: \put(80,3){$x_j$} \put(80,60){$y_j$}
831: \put(103,33){$s$}
832: \end{picture}
833: \qquad
834: \begin{picture}(120,76)(0,-13)
835: \put(0,0){\includegraphics{figures/9/IVTIII.eps}}
836: \put(80,3){$x_j$} \put(80,60){$y_j$}
837: \put(99,8){$s$}
838: \put(0,-13){$\rho_s(x_j)-\rho_s(y_j)<0$}
839: \end{picture}
840: \]
841:
842:
843: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
844: \begin{lemma}\label{L9:reality}
845: If\/ $\Gamma$ is a net satisfying~\eqref{E9:hypothesis}, then there is a choice of critical
846: points $s_1,\dotsc,s_{2p}$ interlacing the points of the sets $A_i$ such that every rational
847: function $\rho$ of degree $p{+}1$ with the net $\Gamma$ and critical points $s_j$ satisfies
848: %
849: \begin{equation}\label{E9:reprise}
850: \rho|_{A_i}\mbox{ is constant for }i=1,\dotsc,n\,.
851: \end{equation}
852: %
853: \end{lemma}
854: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
855:
856: The set of possible critical points $\bs=(s_1,\dotsc,s_{2p})\in(\R\P^1)^{2p}$ interlacing the
857: sets $A_i$ forms the interior of a closed cube
858: \[
859: Q\ :=\ [x_1,y_1]\times[x_2,y_2]\times\dotsb\times[x_{2p},y_{2p}]\,.
860: \]
861: By Lemma~\ref{L9:interpolate}, for every $j$, the function
862: $\varphi_j(\bs):=\rho_{\bs}(x_j)-\rho_{\bs}(y_j)$ (defined as described in the proof of
863: Lemma~\ref{L9:interpolate}) is positive on the face $s_j=y_j$ and negative on the face
864: $s_j=x_j$.
865: This implies (via variant of Brower's fixed point theorem) that there is a point $\bs$ in the
866: interior of $Q$ where $\varphi_j(\bs)=0$ for all $j$, that is
867: $\rho_{\bs}(x_j)=\rho_{\bs}(y_j)$ for all $j$.
868: Since these intervals interlace the sets $A_i$, this implies~\eqref{E9:reprise}.
869: \QED
870:
871:
872: The next step is to show that the number of nets satisfying~\eqref{E9:hypothesis} for sets
873: $A_1,\dotsc,A_n$ where $A_i$ has $1{+}a_i$ members and $a_1+\dotsb+a_n=2p$ is the Kostka number
874: $K_{\ba}$.
875: This Kostka number is the number of Young tableaux of shape $2\times p$ and content
876: $\ba$~\cite[p.25]{Fu}.
877: These are arrays consisting of two rows of integers, each of length $p$ such that the integers
878: increase weakly across each row and strictly down each column.
879: For example, here are the five Young tableaux of shape $2\times p$ and content
880: $(3,2,2,2,1)$, showing that $K_{(3,2,2,2,1)}=5$.
881: %
882: \begin{equation}\label{Eq9:tableaux}
883: \raisebox{-11pt}{
884: \begin{picture}(68,29)(-3.4,-3)
885: \put(-3.4,-3){\includegraphics{figures/9/YT.eps}}
886: \put(0,13){1} \put(13,13){1} \put(26,13){1} \put(39,13){2} \put(52,13){2}
887: \put(0, 0){3} \put(13, 0){3} \put(26, 0){4} \put(39, 0){4} \put(52, 0){5}
888: \end{picture}
889: \quad
890: \begin{picture}(63,27)(-3.4,-3)
891: \put(-3.4,-3){\includegraphics{figures/9/YT.eps}}
892: \put(0,13){1} \put(13,13){1} \put(26,13){1} \put(39,13){3} \put(52,13){4}
893: \put(0, 0){2} \put(13, 0){2} \put(26, 0){3} \put(39, 0){4} \put(52, 0){5}
894: \end{picture}
895: \quad
896: \begin{picture}(63,27)(-3.4,-3)
897: \put(-3.4,-3){\includegraphics{figures/9/YT.eps}}
898: \put(0,13){1} \put(13,13){1} \put(26,13){1} \put(39,13){3} \put(52,13){3}
899: \put(0, 0){2} \put(13, 0){2} \put(26, 0){4} \put(39, 0){4} \put(52, 0){5}
900: \end{picture}
901: \quad
902: \begin{picture}(63,27)(-3.4,-3)
903: \put(-3.4,-3){\includegraphics{figures/9/YT.eps}}
904: \put(0,13){1} \put(13,13){1} \put(26,13){1} \put(39,13){2} \put(52,13){4}
905: \put(0, 0){2} \put(13, 0){3} \put(26, 0){3} \put(39, 0){4} \put(52, 0){5}
906: \end{picture}
907: \quad
908: \begin{picture}(63,27)(-3.4,-3)
909: \put(-3.4,-3){\includegraphics{figures/9/YT.eps}}
910: \put(0,13){1} \put(13,13){1} \put(26,13){1} \put(39,13){2} \put(52,13){3}
911: \put(0, 0){2} \put(13, 0){3} \put(26, 0){4} \put(39, 0){4} \put(52, 0){5}
912: \end{picture}}
913: \end{equation}
914: %
915:
916: We only describe the map from nets to Young tableaux.
917: Given a net satisfying~\eqref{E9:hypothesis}, we will successively place
918: integers into a left-justified two-rowed array while traversing $\R\P^1$.
919: This starts from the first (in the canonical ordering on $\R\P^1$) point in $A_1$ and
920: begins with an empty array.
921: When a critical point $s_j$ is encountered, there will be an interior edge of the net
922: with endpoint $s_j$.
923: Place the integer $j$ in the second row if the other endpoint of that edge has already
924: been encountered, and in the first row if it has not been encountered.
925: For example, the tableaux in~\eqref{Eq9:tableaux} correspond, in order, to the nets
926: in~\eqref{Eq9:interpolate_nets}.
927: (There, the order on $\R\P^1$ is counterclockwise on the circles.)
928:
929:
930: This bijection shows that we have constructed $K_{\ba}$ equivalence classes of rational
931: functions satisfying~\eqref{E9:constant}.
932: To complete the proof of Theorem~\ref{T9:EGSV}, we first show that the Kostka number $K_\ba$ is
933: the expected number of equivalence classes of rational functions
934: satisfying~\eqref{E9:constant}, and then that there is some choice of the sets $A_i$ for which
935: there are exactly $K_{\ba}$ equivalence classes of rational functions.
936: This last step will also prove Theorem~\ref{T9:Pieri}.
937:
938: Recall that a polynomial $f$ of degree $p{+}1$ may be considered to be a linear map on
939: $\C^{p+2}$ so that the composition with the rational normal curve
940: $\gamma(t)\colon\C\to\C^{p+2}$ gives the polynomial $f(t)$.
941: We used this to relate ramification to osculating flags as in Section~\ref{S8:duality}.
942: A two-dimensional space $\langle f,g\rangle$ of polynomials gives a map
943: \[
944: \C\ \xrightarrow{\ \gamma(t)\ }\
945: \C^{p+2}\ \xrightarrow{\ (f,g)\ }\ \C^2\,.
946: \]
947: The kernel $H$ of the map $\C^{p+2} \xrightarrow{\ ( f,g)\ }\C^2$
948: corresponds to $\langle f,g\rangle$ under Grassmann duality.
949:
950: Suppose that the rational function $\rho=f/g$ is constant on some finite subset,
951: $A$.
952: This means that the line $(f(a),g(a))\subset\C^2$ is constant, for $a\in A$.
953: This implies that in $\C^{p+2}$ we have
954: \[
955: \langle H, \gamma(a)\rangle\ =\ \langle H, \gamma(b)\rangle\,,
956: \]
957: for any $a,b\in A$.
958: In particular, $H$ has exceptional position with respect to the $|A|$-plane
959: $\DeCo{S(A)}:=\langle \gamma(a)\mid a\in A\rangle$ in that the two subspaces do not span
960: $\C^{p+2}$.
961:
962: Thus, the equivalence classes of rational functions $\rho$ of degree $p+1$ that
963: satisfy~\eqref{E9:reprise} correspond to the $p$-planes $H$ in $\C^{p+2}$ such that
964: %
965: \begin{equation}\label{Eq9:span}
966: \Span (H, S(A_i))\ \neq\ \C^{p+2}\qquad
967: \mbox{for}\ i=1,\dotsc,n\,.
968: \end{equation}
969: %
970: Those $H$ which satisfy~\eqref{Eq9:span} are the points of an intersection of Schubert
971: varieties.
972: Let $\DeCo{\sigma_a}:=(2,3,\dotsc,a{+}1,\; a{+}3,\dotsc,p{+}2)\in\binom{[p{+}2]}{p}$.
973: Then $X_{\sigma_a}\Fdot$ consists of those $H\in\Gr(p,p{+}2)$ such that
974: \[
975: \Span (H, F_{a+1})\ \neq\ \C^{p+2}\,.
976: \]
977: We will also write \DeCo{$X_{\sigma_a}F_{a{+}1}$} for this Schubert variety, which has
978: dimension $|\sigma_a|=2p-a$.
979: Thus the solutions to the interpolation problem~\eqref{E9:reprise} correspond to the
980: intersection of Schubert varieties
981: \[
982: X_{\sigma_{a_1}}S(A_1)\;\cap\;
983: X_{\sigma_{a_2}}S(A_2)\;\cap\; \dotsb\;\cap\;
984: X_{\sigma_{a_n}}S(A_n)\,,
985: \]
986: which is expected to be zero dimensional.
987: These are special Schubert varieties, so the expected number of points in this
988: intersection may be computed using the Pieri formula, and it is the Kostka number
989: $K_{\ba}$~\cite[p.25]{Fu}.
990:
991: All that remains to show is that there is some choice of the sets $A_i$ for which there
992: are finitely many equivalence classes of rational functions satisfying the interpolation
993: condition~\eqref{E9:reprise}.
994: We show that indirectly, by passing to the limit as each set $A_i$ collapses to a single
995: point, $s_i$.
996: If we consider the rational functions for a given net in this limit, we see that the
997: limiting rational function still has degree $p{+}1$, by Corollary~\ref{C9:create_edge} as no
998: interior edges were collapsed in the limit, by Condition~\eqref{E9:hypothesis}.
999: The limiting rational function has a critical point at each $s_i$ of multiplicity $a_i$,
1000: and is necessarily real.
1001:
1002: It is easy to see that there are still $K_\ba$ nets that have critical points of
1003: multiplicity $a_i$ at points $s_i$---the same bijection works.
1004: For example, here are the nets of rational functions of degree five with critical points
1005: having multiplicities $(3,2,2,2,1)$.
1006: \[
1007: \includegraphics[height=75pt]{figures/9/G27.pieri.1.eps}\quad
1008: \includegraphics[height=75pt]{figures/9/G27.pieri.5.eps}\quad
1009: \includegraphics[height=75pt]{figures/9/G27.pieri.4.eps}\quad
1010: \includegraphics[height=75pt]{figures/9/G27.pieri.3.eps}\quad
1011: \includegraphics[height=75pt]{figures/9/G27.pieri.2.eps}
1012: \]
1013: Moreover, the corresponding intersection of Schubert varieties is expected to have
1014: $K_\ba$ points.
1015: Thus we have constructed the expected number of real rational functions with the desired
1016: critical points.
1017: The proof of Theorem~\ref{T9:Pieri} now follows from the results of~\cite{EG02}, which gave a
1018: bijective correspondence between nets and rational functions.
1019:
1020: