1: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2: \chapter{Overview}\label{ch:1}
3: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
4:
5:
6: In mathematics and its applications, we are often faced with a system of polynomial
7: equations, and we need to study or find the solutions.
8: Such systems that arise naturally typically possess some geometric or combinatorial
9: structure that may be exploited to study their solutions.
10: Such structured systems are studied in enumerative algebraic geometry, which has given us
11: the deep and powerful tools of intersection theory~\cite{Fu} to count and analyze their
12: {\sl complex} solutions.
13: A companion to this theoretical work are algorithms, both symbolic (based on Gr\"obner
14: bases~\cite{CLO,GBCP}) and numerical (based on homotopy continuation~\cite{SW05}) for
15: solving and analyzing systems of polynomial equations.
16:
17: Despite these successes, this line of research largely sidesteps the often primary goal of
18: formulating problems as solutions to systems of equations---namely to determine or study
19: their real solutions.
20: This deficiency is particularly acute in applications, from control~\cite{Byrnes},
21: Kinematics~\cite{BoRo}, statistics~\cite{PRW}, and computational biology~\cite{ASCB}, for
22: it is typically the real solutions that applications seek.
23: One reason that traditional algebraic geometry ignores the real solutions is that there
24: are few elegant theorems or general results available to study real solutions.
25: Nevertheless, the demonstrated importance of understanding the real solutions to systems
26: of equations demands our attention.
27:
28: In the 19th century and earlier, many elegant and powerful methods were developed to study
29: the real roots of univariate polynomials (Sturm sequences, Budan-Fourier Theorem,
30: Routh-Hurwitz criterion), which are now standard tools in some applications of
31: mathematics.
32: In contrast, it has only been in the pat few decades that serious attention has been paid
33: toward understanding the real solutions to systems of polynomial equations.
34:
35: This work has concentrated on systems possessing some, particularly geometric, structure.
36: The reason for this is two-fold:
37: systems from nature typically possess some special structure that should be exploited in
38: their study, and it is unlikely that any results of substance hold for general or
39: unstructured systems.
40: In this period, a story has emerged of bounds (both upper and lower) on the number of real
41: solutions to certain classes of systems, as well as the discovery and study of systems
42: that have only real solutions.
43: This Overview will sketch this emerging landscape and the subsequent chapters will treat
44: these developments in more detail.
45:
46:
47: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
48: %
49: \section*{Introduction}
50: %
51: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
52:
53: Our goal will be to say something
54: meaningful about the real solutions to a system of equations.
55: For example, consider a system
56: %
57: \begin{equation}\label{E1:system}
58: f_1(x_1,\dotsc,x_n)\ =\
59: f_2(x_1,\dotsc,x_n)\ =\ \dotsb\ =\
60: f_N(x_1,\dotsc,x_n)\ =\ 0\,,
61: \end{equation}
62: %
63: of $N$ real polynomials in $n$ variables.
64: Let \DeCo{$r$} be its number of real solutions and \DeCo{$d$} its number of complex
65: solutions\fauxfootnote{We shall always
66: assume that our systems are \DeCo{{\sl generic}} in the sense
67: that all of their solutions occur with multiplicity 1, and the
68: number \DeCo{$d$} of complex solutions is stable under certain
69: \DeCo{allowed} perturbations of the coefficients.}.
70: Since every real number is complex, and since nonreal solutions come
71: in conjugate pairs, we have the following trivial inequality
72: \[
73: d\ \geq\ r\ \geq\ d\!\mod 2\ \;\in\ \;\{0,1\}\,.
74: \]
75: We can say nothing more unless the equations have some structure,
76: and a particularly fruitful class of structures are those which come from
77: geometry.
78: The main point of this book is that we can identify
79: structures in equations that will allow us to do better
80: than this trivial inequality.
81:
82: Our discussion will have three themes:
83: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
84: \begin{enumerate}
85: \item[(I)] Sometimes, the upper bound $d$ is not sharp and there is a
86: smaller bound for $r$.
87: \item[(II)] For many problems from enumerative geometry, the
88: upper bound is sharp.
89: \item[(III)] The lower bound for $r$ may be significantly
90: larger than $d\mod 2$.
91: \end{enumerate}
92: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
93: A lot of time will be spent on the Shapiro Conjecture (Theorem of Mukhin, Tarasov, and
94: Varchenko~\cite{MTV_Sh}) and its generalizations, which
95: is a situation where the upper bound of $d$ is also the lower
96: bound---all solutions to our system are real.
97:
98: We will not describe how to actually find the solutions to a system~\eqref{E1:system} and
99: there will be no discussion of algorithms nor any
100: complexity analysis.
101: The book of Basu, Pollack, and Roy~\cite{BPR03} is an excellent place to learn about
102: algorithms for computing real algebraic varieties and finding real solutions.
103: %
104: % Cite new oberwolfach seminar book?
105: %
106: We remark that some of the techniques we employ to study real solutions
107: underlie numerical algorithms to compute the solutions.\medskip
108:
109: One class of systems that we will study are systems of sparse
110: polynomials.
111: Integer vectors $a=(a_1,\dotsc,a_n)\in\Z^n$ are exponents for (Laurent) monomials
112: %
113: \[
114: \Z^n\ni a\ \leftrightarrow\ \DeCo{x^a}\ :=\ x_1^{a_1}x_2^{a_2}\dotsb
115: x_n^{a_n}\ \in\ \C[x_1,\dotsc,x_n,x_1^{-1},\dotsc,x_n^{-1}]\,.
116: \]
117: %
118: Sometimes, we will just call elements of $\Z^n$ \DeCo{{\sl monomials}}.
119: Let $\DeCo{\calA}\subset\Z^n$ be a finite set of monomials.
120: A linear combination
121: \[
122: f\ =\ \sum_{a\in\calA} c_a x^a\qquad c_a\in\R
123: \]
124: of monomials from $\calA$ is a \DeCo{{\sl sparse polynomial}} with
125: \DeCo{{\sl support} $\calA$}.
126: Sparse polynomials naturally define functions on the complex torus $(\C^\times)^n$.
127: A system~\eqref{E1:system} of $N=n$ polynomials in $n$ variables, where
128: each polynomial has support $\calA$, will be called a
129: \DeCo{{\sl system}} (of polynomials) with \DeCo{{\sl support} $\calA$}.
130: These are often called \DeCo{{\sl unmixed systems}} in contrast to
131: \DeCo{{\sl mixed systems}} where each polynomial may have different support.
132: While sparse systems occur naturally---multilinear or multihomogeneous polynomials are
133: an example---they also occur in problem formulations for the simple reason
134: that we humans seek simple formulations of problems, and this may mean polynomials with
135: few terms.
136:
137:
138: A fundamental result about unmixed systems is the Kushnirenko bound
139: on their number of complex solutions.
140: The \DeCo{{\sl Newton polytope}} of a polynomial with support $\calA$
141: is the convex hull \DeCo{$\Delta_\calA$} of the set $\calA$ of monomials.
142: Write \DeCo{$\vol(\Delta)$} for the Euclidean volume of a polytope $\Delta$.
143:
144: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
145: \begin{thm}[Kushnirenko~\cite{BKK}]\label{T1:Koushnirenko}
146: A system of $n$ polynomials in $n$ variables with common support $\calA$
147: has at most $n!\vol(\Delta_\calA)$ isolated solutions in
148: $(\C^\times)^n$, and exactly this number when the polynomials are generic
149: polynomials with support $\calA$.
150: \end{thm}
151: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
152:
153: Bernstein generalized this to mixed systems.
154: The Minkowski sum $P+Q$ of two polytopes in $\R^n$ is their pointwise sum as sets of
155: vectors in $\R^n$.
156: Let $P_1,\dotsc,P_n\subset\R^n$ be polytopes.
157: The volume
158: \[
159: \vol( t_1 P_1\ +\ t_2P_2\ +\ \dotsb\ +\ t_nP_n)
160: \]
161: is a homogeneous polynomial of degree $n$ in the variables
162: $t_1,\dotsc,t_n$~\cite[Exercise 15.2.6]{Grunbaum}.
163: The \DeCo{{\sl mixed volume} MV$(P_1,\dotsc,P_n)$} of $P_1,\dotsc,P_n$ is the coefficient
164: of the monomial $t_1t_2\dotsb t_n$ in this polynomial.
165:
166:
167: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
168: \begin{thm}[Bernstein~\cite{Be75}]\label{T1:Bernstein}
169: A system of $n$ polynomials in $n$ variables where the polynomials have supports
170: $\calA_1,\dotsc,\calA_n$ has at most $\mbox{\rm
171: MV}(\Delta_{\calA_1},\dotsc,\Delta_{\calA_n})$ isolated solutions in
172: $(\C^\times)^n$, and exactly this number when the polynomials are generic for their given
173: support.
174: \end{thm}
175: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
176:
177: Since $\mbox{\rm MV}(P_1,\dotsc,P_n)=n!\vol(P)$ when $P_1=\dotsb=P_n=P$, this generalizes
178: Kushnirenko's Theorem.
179: We will prove Kushnirenko's Theorem in Chapter~\ref{Ch:sparse},
180: but will not present a proof of Bernstein's Theorem.
181: Instead, we suggest two excellent sources by Sturmfels.
182: Both are similar, but the first is self-contained and superbly written.
183:
184: \begin{itemize}
185:
186: \item{\cite{St98}} \ \
187: \emph{Polynomial equations and convex polytopes}, Amer. Math.
188: Monthly \textbf{105} (1998), no.~10, 907--922.
189:
190: \item{\cite{SPE}} \ \ Chapter 3 in
191: \emph{Solving systems of polynomial equations}, CBMS, vol.~97,
192: American Mathematical Society, Providence, RI, 2002.
193:
194: \end{itemize}
195:
196: The bound of Theorem~\ref{T1:Koushnirenko} and its generalization
197: Theorem~\ref{T1:Bernstein} is often called the
198: \DeCo{{\sl BKK bound}} for \DeCo{B}ernstein, \DeCo{K}hovanskii,
199: and \DeCo{K}ushnirenko~\cite{BKK}.
200:
201: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
202: \section{Upper bounds}
203:
204: While the number of complex roots of a univariate polynomial is
205: typically equal to its degree, the number of real roots depends upon
206: the length of the expression for the polynomial.
207: Indeed, by Descartes's rule of signs~\cite{D1637} (see Section~\ref{S2:Descartes}),
208: a univariate polynomial with $m{+}1$ terms has at most $m$ positive roots,
209: and thus at most $2m$ nonzero real roots.
210: For example, the polynomial $x^d-a$ with $a\neq 0$ has 0, 1, or 2 real roots,
211: but always has $d$ complex roots.
212: Khovanskii generalized this type of a bound to multivariate polynomials with
213: his fundamental \DeCo{{\sl fewnomial bound}}.
214:
215: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
216: \begin{thm}[Khovanskii~\cite{Kh80}]\label{T1:Khovanski}
217: A system of\/ $n$ polynomials in $n$ variables having a total of \/
218: $l{+}n{+}1$ distinct monomials has at most
219: \[
220: 2^{\binom{l+n}{2}}(n+1)^{l+n}
221: \]
222: nondegenerate positive real solutions.
223: \end{thm}
224: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
225:
226: There are two reasons for this restriction to positive solutions.
227: Most fundamentally is that Khovanskii's proof requires this restriction.
228: This restriction also excludes the following type of trivial zeroes:
229: Under the substitution $x_i\mapsto x_i^2$, each positive solution becomes $2^n$ real
230: solutions, one in each of the $2^n$ orthants.
231: More subtle substitutions lead to similar extra trivial zeroes which differ
232: from the positive solutions only by some sign patterns.
233:
234:
235: This is the first of many results which verified the principle of Bernstein and
236: Kushnirenko that the topological complexity of a set defined by real polynomials should
237: depend on the number of terms in the polynomials and not on their degree.
238: Khovanskii's work was also a motivation for the notion of o-minimal structures~\cite{vdD,PiSt}.
239: The main point of Khovanskii's theorem is the existence of such a bound and not the actual
240: bound itself.
241: For each $l,n\geq 1$, we define the \DeCo{{\sl Khovanskii number} $X(l,n)$} to be the
242: maximum number of nondegenerate positive solutions to a system of $n$ polynomials in $n$
243: variables with $l+n+1$ monomials.
244: Khovanskii's Theorem gives a bound for $X(l,n)$, but that bound is enormous.
245: For example, when $l=n=2$, the bound is 5184.
246: Because of this, it was expected to be far from sharp.
247: Despite this expectation, the first nontrivial improvement was only given in 2003.
248:
249: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
250: \begin{thm}[Li, Rojas, and Wang~\cite{LRW03}]\label{T1:LRW}
251: Two trinomials in two variables have at most $5$ nondegenerate positive real
252: solutions.
253: \end{thm}
254: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
255:
256: This bound sharp.
257: Haas~\cite{Ha02} had shown that the system of two
258: trinomials in $x$ and $y$
259: %
260: \begin{equation}\label{E1:Haas}
261: 10x^{106} + 11 y^{53} - 11y\ =\
262: 10y^{106} + 11 x^{53} - 11x\ =\ 0\,,
263: \end{equation}
264: %
265: has 5 positive solutions.
266:
267: Since we may multiply one of the trinomials in~\eqref{E1:Haas} by an arbitrary monomial
268: without changing the solutions, we can assume that the two trinomials~\eqref{E1:Haas} share a
269: common monomial, and so there are at most $3+3-1=5=2+2+1$ monomials between the two
270: trinomials, and so two trinomials give a fewnomial system with $l=n=2$.
271: While 5 is less than 5184, Theorem~\ref{T1:LRW} does
272: not quite show that $X(2,2)=5$ as two trinomials do not constitute a general
273: fewnomial system with $l=n=2$.
274: Nevertheless, Theorem~\ref{T1:LRW} gave strong evidence that Khovanskii's bound may be
275: improved.
276: Such an improved bound was given in~\cite{BS07}.
277:
278: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
279: \begin{thm}\label{T1:New_Bounds}
280: ${\displaystyle X(l,n)\ <\ \tfrac{e^2+3}{4}2^{\binom{l}{2}}n^l}$.
281: \end{thm}
282: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
283:
284: For small values of $l$, it is not hard to improve this.
285: For example, when $l=0$, the support $\calA$ of the system is a simplex, and there will be
286: at most 1 positive real solution, so $X(0,n)=1$.
287: Theorem~\ref{T1:New_Bounds} was inspired by the sharp bound of
288: Theorem~\ref{Th1:BBS} when $l=1$~\cite{BBS}.
289: A set $\calA$ of exponents is \DeCo{{\sl primitive}} if $\calA$ affinely spans
290: the full integer lattice $\Z^n$.
291:
292: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
293: \begin{thm}\label{Th1:BBS}
294: If $l=1$ and the set $\calA$ of exponents is primitive, then
295: there can be at most $2n{+}1$ nondegenerate nonzero real solutions, and this
296: is sharp in that for any $n$ there exist systems with $n{+}2$ monomials and $2n{+}1$
297: nondegenerate real solutions whose exponent vectors affinely span $\Z^n$.
298: \end{thm}
299: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
300:
301: Observe that this bound is for all real solutions, not just positive solutions.
302: We will discuss this in Section~\ref{S4:circuit}.
303: Further analysis gives the sharp bound for $X(1,n)$.
304:
305: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
306: \begin{thm}[Bihan~\cite{Bihan}]
307: $X(1,n)=n+1$.
308: \end{thm}
309: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
310:
311: In contrast to these results establishing absolute upper bounds for the
312: number of real solutions which improve the trivial bound of the number
313: $d$ of complex roots, there are a surprising number of problems that
314: come from geometry for which all solutions can be real.
315: For example, Sturmfels~\cite{St94} proved the following.
316:
317: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
318: \begin{thm}\label{T1:Sturm}
319: Suppose that a lattice polytope $\Delta\subset\Z^n$ admits a regular
320: triangulation with each simplex having minimal volume $\frac{1}{n!}$.
321: Then there is a system of sparse polynomials with support
322: $\Delta\cap\Z^n$ having all solutions real.
323: \end{thm}
324: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
325:
326: For many problems from enumerative geometry, it is similarly possible
327: that all solutions can be real.
328: This will be discussed in Chapter~\ref{S:ERAG}.
329: The state of affairs in 2001 was presented in~\cite{So03}.
330:
331:
332: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
333: %
334: \section{The Wronski map and the Shapiro Conjecture}
335: %
336: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
337:
338: The \DeCo{{\sl Wronskian}} of univariate polynomials
339: $f_1(t), f_2(t), \dotsc,f_m(t)$ is the determinant
340: \[
341: \DeCo{\Wr(f_1,\dotsc,f_m)}\ :=\ \det \bigl(
342: ({\textstyle \frac{\partial}{\partial t}})^{j-1} f_i(t)
343: \bigr)_{i,j=1,\dotsc,m}
344: \ .
345: \]
346: When the polynomials $f_i$ have degree $m{+}p{-}1$ and are linearly independent, the
347: Wronskian has degree at most $mp$.
348: For example, if $m=2$, then $W(f,g)=f'g-fg'$, which has degree $2p$ as the
349: coefficients of $t^{2p+1}$ in this expression cancel.
350: Up to a scalar, the Wronskian depends only upon the linear span of the
351: polynomials $f_1,f_2,\dotsc,f_m$.
352: Removing these ambiguities gives the \DeCo{{\sl Wronski map}},
353: %
354: \begin{equation}\label{E1:Wronski}
355: \DeCo{\Wr}\ :=\ \Gr(m,\C_{m{+}p{-}1}[t])\ \longrightarrow\ \P^{mp}\,,
356: \end{equation}
357: %
358: where \DeCo{$\Gr(m,\C_{m{+}p{-}1}[t])$} is the \DeCo{{\sl Grassmannian}} of $m$-dimensional
359: subspaces of the linear space $\C_{m{+}p{-}1}[t]$ of polynomials of degree $m{+}p{-}1$ in
360: the variable $t$, and $\P^{mp}$ is the projective space of
361: polynomials of degree at most $mp$, which has dimension equal to the
362: the dimension of the Grassmannian.
363:
364: Work of Schubert in 1886~\cite{Sch1886b}, combined with a result of
365: Eisenbud and Harris in 1983~\cite{EH83} shows that the Wronski map is surjective
366: and the general polynomial $\Phi\in\P^{mp}$ has
367: %
368: \begin{equation}\label{E1:WronDeg}
369: \DeCo{\#_{m,p}}\ :=\
370: \frac{1!2!\dotsb(m{-}1)!\cdot(mp)!}%
371: {m!(m{+}1)!\dotsb (m{+}p{-}1)!}
372: \end{equation}
373: %
374: preimages under the Wronski map.
375: These results concern the complex Grassmannian and complex projective space.
376:
377: Boris Shapiro and Michael Shapiro made a
378: conjecture in 1993/4 about the Wronski map from
379: the real Grassmannian to real projective space.
380: This was proven when $\min(k,d{+}1{-}k)=2$ by Eremenko and Gabrielov~\cite{EG02}, and finally
381: settled by Mukhin, Tarasov, and Varchenko~\cite{MTV_Sh}.
382: They have subsequently found another proof~\cite{MTV_R}.
383:
384: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
385: \begin{thm}\label{T1:Shap_Conj}
386: If the polynomial $\Phi\in\P^{mp}$ has only real zeroes, then every point in
387: $\Wr^{-1}(\Phi)$ is real.
388: Moreover, if $\Phi$ has $mp$ simple real zeroes
389: then there are $\#_{m,p}$ real points in $\Wr^{-1}(\Phi)$.
390: \end{thm}
391: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
392:
393: We will sketch the proof of Mukhin, Tarasov, and Varchenko in Chapter~\ref{ch:MTV}.
394: This \DeCo{{\sl Shapiro Conjecture}} has appealing geometric interpretations, enjoys links
395: to several areas of mathematics, and has many theoretically satisfying generalizations
396: which we will discuss in Chapters~\ref{Ch:ScGr},~\ref{Ch:EG}, and~\ref{Ch:Frontier}.
397: We now mention two of its interpretations.
398:
399:
400: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
401: %
402: \subsection{The problem of four lines}
403: %
404: A geometric interpretation of the Wronski map and the Shapiro Conjecture
405: when $m=p=2$ is a variant of the classical problem of the
406: lines in space which meet four given lines.
407: Points in $\Gr(2,\C_3[t])$ correspond to lines in $\C^3$ as follows.
408: The \DeCo{{\sl moment curve} $\gamma$} in $\C^3$ is the curve with
409: parameterization
410: \[
411: \gamma(t)\ :=\ (t, t^2, t^3)\,.
412: \]
413: A cubic polynomial $f$ is the composition of $\gamma$ and an affine-linear map
414: $\C^3\to\C$, and so a two-dimensional space of cubic polynomials is to a
415: two-dimensional space of affine-linear maps whose common kernel is the corresponding line
416: in $\C^3$.
417: (This description is not exact, as some points in $\Gr(2,\C_3[t])$ correspond to lines at
418: infinity.)
419:
420:
421: Given a polynomial $\Phi(t)$ of degree 4 with distinct real roots,
422: points in the fiber $\Wr^{-1}(\Phi)$ correspond to the lines in space
423: which meet the four lines tangent to the moment curve $\gamma$ at its
424: points coming from the roots of $\Phi$.
425: There will be two such lines, and the Shapiro conjecture asserts that
426: both will be real.
427:
428: It is not hard to see this directly.
429: Any fractional linear change of parameterization of the moment curve
430: is realized by a projective linear transformation of 3-dimensional space which
431: stabilizes the image of the moment curve.
432: Thus we may assume that the polynomial $\Phi(t)$ is equal to $(t^3-t)(t-s)$,
433: which has roots $-1,0,1$, and $s$, where $s\in (0,1)$.
434: Applying an affine transformation to 3-dimensional space, the moment curve becomes
435: the curve with parameterization
436: \[
437: \gamma\ :\ t\ \longmapsto\
438: (6t^2-1, \ \tfrac{7}{2}t^3+\tfrac{3}{2}t,\ \tfrac{3}{2}t-\tfrac{1}{2}t^3)\,.
439: %
440: %[-1+6*t^2, 7/2*t^3+3/2*t, -1/2*t^3+3/2*t]
441: %[12*t, 21/2*t^2+3/2, -3/2*t^2+3/2]
442: %
443: % gamma(-1) =( 5,-5,-1) gamma(0) =(-1,0,0) gamma(1) =(5,5,1)
444: % gamma'(-1)=(-1, 1, 0) gamma'(0)=( 0,3,3) gamma'(1)=(1,1,0)
445: \]
446: Then the lines tangent to $\gamma$ at the roots $-1,0,1$ of $\Phi$ have
447: parameterizations
448: \[
449: (-5-s,\;5+s,\;-1)\,,\ (-1,\;s,\;s)\,,\ (5+s\;,5+s\;,1)\qquad s\in\R\,.
450: \]
451: These lie on a hyperboloid \Brown{$Q$} of one sheet, which is defined by
452: %
453: \begin{equation}\label{Eq1:Hyperboloid}
454: 1 - x_1^2 + x_2^2 - x_3^2\ =\ 0\,.
455: \end{equation}
456: %
457: We display this geometric configuration in Figure~\ref{F1:hyperboloid}.
458: There, $\ell(i)$ is the line tangent to $\gamma$ at the point $\gamma(i)$.
459: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
460: \begin{figure}[htb]
461: \[
462: \begin{picture}(372,185.2)(0,1)
463: \put(0,0){\includegraphics[height=190pt]{figures/7/hyperboloid.eps}}
464: \put(170,0){\Red{$\ell(-1)$}} \put(354,0){\Red{$\ell(0)$}}
465: \put(0,136){\Red{$\ell(1)$}} \put(0,94){\Blue{$\gamma$}}
466: \put(280,178){\Red{$\ell(-1)$}} \put(337,174){\Blue{$\gamma$}}
467: \put(320,135){\Brown{$Q$}}
468: \end{picture}
469: \]
470: \caption{Quadric containing three lines tangent to $\gamma$\label{F1:hyperboloid}.}
471: \end{figure}
472: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
473: The quadric \Brown{$Q$} has two rulings.
474: One ruling contains our three tangent lines and
475: the other ruling (which is drawn on \Brown{$Q$}) consists of the lines which
476: meet our three tangent lines.
477:
478: Now consider the fourth line \ForestGreen{$\ell(s)$} which is tangent to
479: \Brown{$\gamma$} at the point \Brown{$\gamma(s)$}.
480: This has the parameterization
481: \[
482: \ForestGreen{\ell(s)}\ =\
483: \bigl( 6s^2-1\,,\
484: \tfrac{7}{2}s^3+\tfrac{3}{2}s\,,\
485: \tfrac{3}{2}s-\tfrac{1}{2}s^3 \bigr)\ +\
486: t \bigl(12s\,,\
487: \tfrac{21}{2}s^2+\tfrac{3}{2}\,,\
488: \tfrac{3}{2} -\tfrac{3}{2}s^2\bigr)\,.
489: \]
490: We compute the intersection of the fourth line with $Q$.
491: Substituting its parameterization into~\eqref{Eq1:Hyperboloid} and dividing by $-12$ gives
492: the equation
493: \[
494: (s^3-s)(s^3-s + t(6s^2-2)+9st^2)\ =\ 0\,.
495: \]
496: The first (nonconstant) factor $s^3-s$ vanishes when \ForestGreen{$\ell(s)$} is equal to
497: one of \Red{$\ell(-1)$}, \Red{$\ell(-0)$}, or \Red{$\ell(-1)$}--for these values of $s$
498: every point of \ForestGreen{$\ell(s)$} lies on the quadric \Brown{$Q$}.
499: The second factor has solutions
500: \[
501: t\ =\ -\; \frac{3s^2-1\pm\sqrt{3s^2+1}}{9s}\ .
502: \]
503: Since $3s^2+1>0$ for all $s$, both solutions will be real.
504:
505: We may also see this geometrically.
506: Consider the fourth line \ForestGreen{$\ell(s)$} for $0<s<1$.
507: In Figure~\ref{F1:line4}, we look down the throat of the hyperboloid
508: at the interesting part of this configuration.
509: This picture demonstrates that \ForestGreen{$\ell(s)$} must meet \Brown{$Q$} in
510: two real points.
511: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
512: \begin{figure}[htb]
513: \[
514: %
515: \begin{picture}(285, 159)(-34.5,-11.5)
516: \put(0,0){\includegraphics[height=130pt]{figures/1/shapiro.eps} }
517: \put(-12,63.5){\Red{$\ell(1)$}} \put(237,34.6){\Red{$\ell(-1)$}}
518: \put(193,138){\Red{$\ell(0)$}}
519: \put(8.4,34.5){\Blue{$\gamma$}} \put(2.4,111){\ForestGreen{$\ell(s)$}}
520: \put(117.6,131){\Brown{$Q$}}
521: \thicklines
522: \put(141.6,-13){\Brown{$\gamma(s)$}}
523: \put(156,2.4){\Brown{\vector(0,1){82}}}
524: \end{picture}
525: \]
526: \caption{The fourth tangent line meets \Brown{$Q$} in two real
527: points\label{F1:line4}.}
528: \end{figure}
529: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
530: Through each point, there is a real line in the second ruling which
531: meets all four tangent lines, and this proves Shapiro's conjecture
532: for $m=p=2$.
533:
534: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
535: %
536: \subsection{Rational functions with real critical points}
537: When $m=2$, the Shapiro conjecture may be interpreted in terms of
538: rational functions.
539: A rational function $\rho(t)=f(t)/g(t)$ is a quotient of two univariate
540: polynomials, $f$ and $g$.
541: This defines a map $\rho\colon \P^1\to\P^1$ whose critical points are those $t$ for which
542: $\rho'(t)=0$.
543: Since $\rho'(t)=(f'g-g'f)/g^2$, we see that the critical points are
544: the roots of the Wronskian of $f$ and $g$.
545: Composing the rational function $\rho\colon\P^1\to\P^1$ with an automorphism of the target
546: $\P^1$ gives an equivalent rational function, and the equivalence class of $\rho$ is
547: determined by the linear span of its numerator and denominator.
548: Thus Shapiro's conjecture asserts that a rational function having only real critical
549: points is equivalent to a real rational function.
550:
551: Eremenko and Gabrielov~\cite{EG02} proved exactly this statement in 2002, and
552: thereby established the Shapiro Conjecture in the case $m=2$.
553:
554: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
555: \begin{thm}
556: A rational function with only real critical points is equivalent to a
557: real rational function.
558: \end{thm}
559: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
560:
561: In Chapter~\ref{Ch:EG} we will present an elementary proof of this result
562: that Eremenko and Gabrielov found in 2005~\cite{EG05}.
563:
564:
565: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
566: %
567: \section{Lower bounds}\label{S1:lower}
568: %
569: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
570:
571: We begin with some of perhaps the most exciting recent
572: development in real algebraic geometry.
573: It begins with the fundamental observation of Euclid that two points determine a line.
574: Many people who have studied geometry know that five points on the plane determine a
575: conic. In general, if you have $m$ random points in the plane and you
576: want to pass a rational curve of degree $d$ through all of them,
577: there may be no solution to this interpolation problem (if $m$ is
578: too big), or an infinite number of solutions (if $m$ is too small),
579: or a finite number of solutions (if $m$ is just right). It turns
580: out that ``$m$ just right'' means $m=3d{-}1$ ($m=2$ for lines and
581: $m=5$ for conics).
582:
583: A harder question is, if $m=3d{-}1$, {\it how many\/} rational curves of degree
584: $d$ interpolate the points? Let's call this number $N_d$, so that
585: $N_1=1$ and $N_2=1$ because the line and conic of the previous
586: paragraph are unique. It has long been known that $N_3=12$, and
587: in 1873 Zeuthen~\cite{Ze1873} showed that $N_4=620$.
588: That was where matters stood until 1989, when Ran~\cite{R89} gave a recursion for these
589: numbers.
590: In the 1990's, Kontsevich and Manin~\cite{KM} used associativity in quantum
591: cohomology of $\mathbb{P}^2$ to give the elegant recursion
592: %
593: \begin{equation}\label{E1:Konts}
594: N_d\ =\ \sum_{a+b=d} N_a N_b \left( a^2b^2\binom{3d-4}{3a-2} -
595: a^3b\binom{3d-4}{3a-1}\right)\ ,
596: \end{equation}
597: %
598: which begins with the Euclidean declaration that two points determine a line ($N_1=1$).
599: These numbers grow quite fast, for example $N_5=87304$.
600:
601: The number of real rational curves which interpolate
602: a given $3d-1$ points in the real plane $\R\P^2$ will depend rather
603: subtly on the configuration of the points.
604: To say anything about the real rational curves would seem impossible.
605: However this is exactly what Welschinger~\cite{W} did.
606: He found an invariant which does not depend upon the
607: choice of points.
608:
609: A rational curve in the plane is necessarily singular---typically it has
610: $\binom{d-1}{2}$ nodes.
611: Real curves have three types of nodes.
612: Only two types are visible in $\R\P^2$, and we are familiar with them from
613: rational cubics.
614: The curve on the left below has a node with two real branches, and the curve
615: on the right has a \DeCo{{\sl solitary point}} `{\small$\bullet$}', where two complex conjugate
616: branches meet.
617: %
618: \[
619: \includegraphics[height=70pt]{figures/1/renodes.eps}\qquad\qquad\qquad\qquad\qquad
620: \includegraphics[height=70pt]{figures/1/conodes.eps}
621: \]
622: %
623: The third type of node is a pair of complex conjugate nodes, which are
624: not visible in $\R\P^2$.
625:
626: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
627: \begin{thm}[Welschinger~\cite{W}]
628: The sum,
629: %
630: \begin{equation}\label{Eq1:Wel}
631: \sum (-1)^{\#\{\mbox{\scriptsize solitary points, }\bullet, \mbox{\scriptsize in }C\}}\,,
632: \end{equation}
633: %
634: over all real rational curves $C$ of degree $d$ interpolating $3d{-}1$
635: general points in $\R\P^2$ does not depend upon the choice of the points.
636: \end{thm}
637: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
638:
639: Set $\DeCo{W_d}$ to be the sum~\eqref{Eq1:Wel}.
640: The absolute value of this \DeCo{{\sl Welschinger invariant}} is then a lower
641: bound for the number of real rational curves of degree $d$ interpolating $3d-1$
642: points in $\R\P^2$.
643: Since $N_1=N_2=1$, we have $W_1=W_2=1$.
644: Prior to Welschinger's discovery, Kharlamov~\cite[Proposition 4.7.3]{DeKh00} (see also
645: Section~\ref{sec:ratcubics}) showed that $W_3=8$.
646: The question remained whether any other Welschinger invariants were
647: nontrivial.
648: This was settled in the affirmative by Itenberg, Kharlamov, and
649: Shustin~\cite{IKS03,IKS04}, who used Mikhalkin's
650: Tropical Correspondence Theorem~\cite{Mi05} to show
651: \begin{enumerate}
652: \item If $d>0$, then $W_d\geq \frac{d!}{3}$. (Hence $W_d$ is positive.) \smallskip
653: \item ${\displaystyle \lim_{d\to\infty}
654: \frac{\log N_d}{\log W_d}=1}$. \
655: (In fact, $\log N_d \sim 3d\log d\sim \log N_d$.)
656: \end{enumerate}
657: In particular, there are always quite a few real rational curves
658: of degree $d$ interpolating $3d{-}1$ points in $\R\P^2$.
659: Since then, Itenberg, Kharlamov, and Shustin~\cite{IKS09} gave a recursive formula for
660: the Welschinger invariant which is based upon Gathmann and Markwig's~\cite{GM}
661: tropicalization of the Caporaso-Harris~\cite{CH98} formula.
662: This shows that $W_4=240$ and $W_5=18264$.
663: Solomon~\cite{Sol} has also found an intersection-theoretic interpretation for these invariants.
664:
665: These ideas have also found an application.
666: Gahleitner, J\"uttler, and Schicho~\cite{GJH} proposed a method to compute an
667: approximate parametrization of a plane curve using rational cubics.
668: Later, Fiedler-Le Touz\'e~\cite{FLT} used the result of Kharlamov
669: (that $W_3=8$), and an analysis of pencils of plane cubics to prove that this
670: method works.
671: \smallskip
672:
673: While the story of this interpolation problem is fairly well-known, it
674: was not the first instance of lower bounds in enumerative real
675: algebraic geometry.
676: In their investigation of the Shapiro conjecture, Eremenko and
677: Gabrielov found a similar invariant $\sigma_{m,p}$ which gives a lower bound on the
678: number of real points in the inverse image $\Wr^{-1}(\Phi)$ under the Wronski
679: map of a real polynomial $\Phi\in\R\P^{mp)}$.
680: Assume that $m\leq p$.
681: If $m+p$ is odd, set
682: \[
683: \DeCo{\sigma_{m,p}}\ :=\
684: \frac{1!2!\dotsb(m{-}1)!(p{-}1)!(p{-}2)!\dotsb(p{-}m{+}1)!(\frac{mp}{2})!}
685: {(p{-}m{+}2)!(p{-}m{+}4)!\dotsb(p{+}m{-}2)!\left(\frac{p-m+1}{2}\right)!%
686: \left(\frac{p-m+3}{2}\right)!\dotsb\left(\frac{p+m-1}{2}\right)!}\ .
687: \]
688: If $m+p$ is even, then set $\sigma_{m,p}=0$.
689: If $m>p$, then set $\sigma_{m,p}:=\sigma_{p,m}$.
690:
691:
692: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
693: \begin{thm}[Eremenko-Gabrielov~\cite{EG01}]\label{T1:E-G01}
694: If\/ $\Phi(t)\in\R\P^{mp}$ is a generic real polynomial of degree
695: $mp$ (a regular value of the Wronski map), then there are at least
696: $\sigma_{m,p}$ real $m$-dimensional subspaces of polynomials of degree $m{+}p{-}1$
697: with Wronskian $\Phi$.
698: \end{thm}
699: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
700:
701:
702: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
703: \begin{rmk}
704: Recall that the number of complex points in $\Wr^{-1}(\Phi)$ is
705: $\#_{m,p}$~\eqref{E1:WronDeg}.
706: It is instructive to compare these numbers.
707: We show them for $m{+}p=11$ and $m=2,\dotsc,5$.
708: \[
709: \begin{tabular}{|c|r|r|r|r|}\hline
710: $m$ & 2 & 3 & 4 & 5\\\hline
711: $\sigma_{m,p}$& 14 & 110 & 286 & 286 \\\hline
712: $\#_{m,p}$ &4862 &23371634&13672405890&396499770810 \\\hline
713: \end{tabular}
714: \]
715: % 4862 14
716: % 23371634 110
717: % 13672405890 286
718: % 396499770810 286
719: We also have $\sigma_{7,6}\approx 3.4\times 10^4$ and
720: $\#_{7,6}\approx 9.5\times 10^{18}$.
721: % 33592 9,490,348,077,234,178,440.
722: Despite this disparity in their magnitudes, the asymptotic ratio of
723: $\log(\sigma_{m,p})/\log(\#_{m,p})$ appears to be close to $1/2$.
724: We display this ratio in the table below, for different values of $m$ and $p$.
725: \[
726: \begin{tabular}{|c|r|c|c|c|c|c|c|}\hline
727: \multicolumn{2}{|c|}{\multirow{2}{80pt}{
728: ${\displaystyle \frac{\log(\sigma_{m,p})}{\log(\#_{m,p})}}$}}&
729: \multicolumn{6}{|c|}{$m$\rule{0pt}{13 pt}}\\\cline{3-8}
730: \multicolumn{2}{|c|}{\ }
731: &2& $\frac{m{+}p{-}1}{10}$&$2\frac{m{+}p{-}1}{10}$&$3\frac{m{+}p{-}1}{10}$
732: &$4\frac{m{+}p{-}1}{10}$&$5\frac{m{+}p{-}1}{10}$\raisebox{-6pt}{\rule{0pt}{19pt}}\\\hline
733: \multirow{6}{15pt}{\hspace{10pt}\raisebox{-28pt}{\begin{rotate}{90}$m{+}p{-}1$\end{rotate}}\hspace{-10pt}}
734: & 100& 0.47388& 0.45419& 0.43414& 0.41585& 0.39920& 0.38840\\\cline{2-8}
735: %%%% & 500& 0.49325& 0.47257& 0.45777& 0.44457& 0.43281& 0.42557\\\cline{2-8}
736: & 1000& 0.49627& 0.47677& 0.46358& 0.45185& 0.44144& 0.43510\\\cline{2-8}
737: %%%% & 5000& 0.49908& 0.48292& 0.47247& 0.46319& 0.45500& 0.45007\\\cline{2-8}
738: & 10000& 0.49951& 0.48468& 0.47510& 0.46660& 0.45909& 0.45459\\\cline{2-8}
739: %%%% & 50000& 0.49988& 0.48765& 0.47963& 0.47251& 0.46624& 0.46249\\\cline{2-8}
740: & 100000& 0.49994& 0.48860& 0.48111& 0.47445& 0.46860& 0.46511\\\cline{2-8}
741: %%%% & 500000& 0.49999& 0.49033& 0.48384& 0.47806& 0.47299& 0.46998\\\cline{2-8}
742: & 1000000& 0.49999& 0.49092& 0.48479& 0.47932& 0.47453& 0.47168\\\cline{2-8}
743: %%%% & 5000000& 0.50000& 0.49205& 0.48661& 0.48175& 0.47750& 0.47497\\\cline{2-8}
744: &10000000& 0.50000& 0.49246& 0.48726& 0.48263& 0.47857& 0.47616\\
745: % \cline{2-8} &50000000&&&&&&\\
746: \hline
747: \end{tabular}
748: \]
749: %
750: Thus, the lower bound on the number of real points in a fiber of the Wronski map
751: appears asymptotic to the square root of the number of complex solutions.
752:
753: It is interesting to compare this to the the result of Shub and Smale~\cite{ShSm93} that
754: the expected number of real solutions to a system of $n$ Gaussian random polynomials in
755: $n$ variables of degrees $d_1,\dotsc,d_n$ is $\sqrt{d_1\dotsb d_n}$, which is the square
756: root of the number of complex solutions to such a system of polynomials.
757: \end{rmk}
758: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
759:
760: The idea behind the proof of Theorem~\ref{T1:E-G01} is to compute the topological degree
761: of the real Wronski map, which is the restriction of the Wronski map to real subspaces of
762: polynomials,
763: \[
764: \Wr\!_\R\ :=\ \Wr\!|_{\sGr(m,\R_{m{+}p{-}1}[t])}\ \colon\
765: \Gr(m,\R_{m{+}p{-}1}[t])\ \longrightarrow\ \R\P^{mp}\,.
766: \]
767: This maps the Grassmannian of real subspaces to the space of real
768: Wronski polynomials.
769: Recall that the topological degree of a
770: map $f\colon X\to Y$ between two oriented manifolds $X$ and $Y$ of the same dimension
771: is the number $d$ such that $f_*[X]=d[Y]$, where $[X]$ and $[Y]$ are the fundamental
772: homology cycles of $X$ and $Y$, respectively, and $f_*$ is the functorial map in homology.
773: When $f$ is differentiable, this may be computed as follows.
774: Let $y\in Y$ be a regular value of $f$ so that at any point $x$ in the fiber $f^{-1}(y)$
775: above $y$ the derivative $d_xf\colon T_xX \to T_yY$ is an isomorphism.
776: Since $X$ and $Y$ are oriented, the isomorphism $d_xf$ either preserves the orientation or it reverses
777: it.
778: Let $P$ be the number of points $x\in f^{-1}(y)$ at which $d_xf$ preserves the
779: orientation and $R$ be the number of points where the orientation is reversed.
780: Then the degree of $f$ is the difference $P-R$.
781:
782: There is a slight problem in computing the degree of $\Wr\!_\R$, as neither the real
783: Grassmannian nor the real projective space is orientable when $m{+}p$ is odd, and thus the
784: topological degree of $\Wr\!_\R$ is not defined when $m{+}p$ is odd.
785: Eremenko and Gabrielov get around this by computing the degree of the restriction of the
786: Wronski map to open cells of $\Gr_\R$ and $\R\P^{mp}$, which is a proper map.
787: They also show that it is the degree of a lift of the
788: Wronski map to oriented double covers of both spaces.
789: The degree bears a resemblance to the Welschinger invariant
790: as it has the form $|\sum \pm 1|$, the sum over all real points in
791: $\Wr_\R^{-1}(\Phi)$, for $\Phi$ a regular value of the Wronski map.
792: This resemblance is no accident.
793: Solomon~\cite{Sol} showed how to orient a moduli space of rational curves with marked
794: points so that the Welschinger invariant is indeed the degree of a map.
795: \medskip
796:
797: While both of these examples of geometric problems possessing a lower bound
798: on their numbers of real solutions are quite interesting, they are
799: rather special.
800: The existence of lower bounds for more general problems or for more general
801: systems of polynomials would be quite important in applications, as these lower
802: bounds guarantee the existence of real solutions.
803:
804: With Soprunova, we~\cite{SS} set out to develop a theory of lower bounds
805: for sparse polynomial systems, using the approach
806: of Eremenko and Gabrielov via topological degree.
807: This is a first step toward practical applications of these ideas.
808: Chapter~\ref{Ch:lower} will elaborate this theory.
809: Here is an outline:
810: %
811: \begin{enumerate}
812: \item[({\it i})] Realize the solutions to a system of polynomials as the fibers
813: of a map from a toric variety.
814: \item[({\it ii})] Give a condition which implies that the degree of this map (or a
815: lift to double covers) exists.
816: \item[({\it iii})] Develop a method to compute the degree in some (admittedly
817: special) cases.
818: \item[({\it iv})] Give a nice family of examples to which this theory applies.
819: \item[({\it v})] Use the sagbi degeneration of a Grassmannian to a toric variety~\cite[Ch.~11]{GBCP} and
820: the systems of ({\it iv}) to recover the result of Eremenko and Gabrielov.
821: \end{enumerate}
822:
823: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
824: \begin{ex}\label{Ex1:P2P2}
825: We close this Chapter with one example from this theory.
826: Let $\Blue{w},\Blue{x},\Brown{y},\Brown{z}$ be indeterminates, and consider a
827: sparse polynomial of the form
828: %
829: \begin{eqnarray}
830: & c_4\, \Blue{wx}\Brown{yz}\hspace{0.2em}&\nonumber\\
831: &+\,c_3(\Blue{wx}\Brown{z} + \Blue{x}\Brown{yz})\hspace{0.9em}&\nonumber\\
832: &+\,c_2(\Blue{wx} + \Blue{x}\Brown{z} + \Brown{yz})\hspace{1em}&\label{E1:P2P2}\\
833: &+\,c_1(\Blue{x}+\Brown{z})\hspace{0.74em}&\nonumber\\
834: &\hspace{.25em}\,+\,c_0\,,&\nonumber
835: \end{eqnarray}
836: where the coefficients $c_0,\dotsc,c_4$ are real numbers.\medskip
837:
838: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
839: \begin{thm}\label{T1:P2P2}
840: A system of four equations involving polynomials of the
841: form~\eqref{E1:P2P2} has six solutions, at least \Blue{two} of which are real.
842: \end{thm}
843: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
844:
845: We make some remarks to illustrate the ingredients of this theory.
846: First, the monomials in the sparse system~\eqref{E1:P2P2} are the
847: integer points in the \Blue{{\sl order polytope}} of the poset $P$,
848: \[
849: \begin{picture}(90,50)(-35,0)
850: \put(-40,20){$P\ :=$}
851: \put(14,3){\includegraphics{figures/6/P2P2.eps}}
852: \put(0,40){$\Blue{x}$} \put(56,40){$\Brown{z}$}
853: \put(0, 0){$\Blue{w}$} \put(56, 0){$\Brown{y}$}
854: \put(65,20){.}
855: \end{picture}
856: \]
857: That is, each monomial corresponds to an order ideal of $P$ (a subset which is
858: closed upwards).
859: The number of complex roots is the number of linear extensions of the
860: poset $P$.
861: There are six, as each is a permutation of the word $\Blue{wx}\Brown{yz}$ where
862: $\Blue{w}$ precedes $\Blue{x}$ and $\Brown{y}$ precedes $\Brown{z}$.
863:
864: One result ({\it ii}) gives conditions on the Newton polytope which imply that
865: any polynomial system with that Newton polytope has a
866: lower bound, and order polytopes satisfy these conditions.
867: Another result ({\it iv}) computes that lower bound for certain families of
868: polynomials with support an order polytope.
869: Polynomials in these families have the form~\eqref{E1:P2P2} in that
870: monomials with the same total degree have the same coefficient.
871: For such polynomials, the lower bound is the absolute value of the sum of the signs of the
872: permutations underlying the linear extensions.
873: We list these for $P$.
874: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
875: \begin{center}
876: \begin{tabular}{|r||c|c|c|c|c|c||c|}\hline
877: permutation&$\Blue{wx}\Brown{yz}$
878: &$\Blue{w}\Brown{y}\Blue{x}\Brown{z}$
879: &$\Brown{y}\Blue{wx}\Brown{z}$
880: &$\Blue{w}\Brown{yz}\Blue{x}$
881: &$\Brown{y}\Blue{w}\Brown{z}\Blue{x}$
882: &$\Brown{yz}\Blue{wx}$&sum\\\hline
883: sign & $+$ & $-$ & $+$ & $+$ & $-$ & $+$ & 2 \\\hline
884: \end{tabular}
885: \end{center}
886: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
887: This shows that the lower bound in Theorem~\ref{T1:P2P2} is 2.
888:
889: We record the frequency of the different root counts in each of
890: 10,000,000 instances of this polynomial system, where the coefficients were chosen
891: uniformly from $[-200,200]$.
892: \[
893: \begin{tabular}{|r|c|c|c|c|}\hline
894: number of real roots&0&2&4&6\\\hline
895: frequency &0&9519429&0&480571\\\hline
896: \end{tabular}
897: \]
898: This computation took 19,854 CPU-seconds, or 5:30:54 on a AMD Athlon 64-bit dual core
899: processor 4600+ (2.4 GHz, 512KB cache), using a single core.
900: \QED
901: \end{ex}
902: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
903: %
904: % Result of 10,000,000 instances with c_i \in [-200,200]
905: % (50,000 each iteration) on May 2008 version of miln303
906: %
907: %model name : AMD Athlon(tm) 64 X2 Dual Core Processor 4600+
908: %cpu MHz : 2400.000
909: %cache size : 512 KB
910: %
911: %Freq:=[0, 9519429, 0, 480571]:
912: %Time:=19854:
913: %Iter:= 200:
914: %
915: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
916: The apparent gap in the numbers of real solutions (4 does not seem a possible number of
917: real solutions) is proven for the system of Example~\ref{Ex1:P2P2}.
918: This is the first instance we have seen of this phenomena of gaps in the numbers of real
919: solutions.
920: More are found in~\cite{SS} and~\cite{RSSS}.