1: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2: %% Trim Size: 9.75in x 6.5in
3: %% Text Area: 8in (include Runningheads) x 5in
4: %% ws-jktr.tex : 10-4-06
5: %% Tex file to use with ws-jktr.cls written in Latex2E.
6: %% The content, structure, format and layout of this style file is the
7: %% property of World Scientific Publishing Co. Pte. Ltd.
8: %% Copyright 1995, 2002 by World Scientific Publishing Co.
9: %% All rights are reserved.
10: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
11: %
12: %\documentclass[11pt]{article}
13: \documentclass[envcountsame]{llncs}[12/12/07]
14: \usepackage{psfrag}
15: \def\rhd{\bullet}
16: \usepackage{picins}
17: \def\abs#1{\left \vert #1 \right \vert}
18: \def\Frac#1#2{{\displaystyle{\frac{#1}{#2}}}}
19: \def\[#1\]{\begin{equation}#1\end{equation}}
20: \def\$#1\${\begin{eqnarray}#1\end{eqnarray}}
21: \def\phi{\varphi}
22: \def\bref#1{{\bf\ref{#1}}}
23: \def\mod#1{\;[\hbox{\rm mod}\; #1]}
24: \def\p{\vspace{-5.17ex} \hspace{6.5ex}}
25: \def\cC{{\cal C}}
26: \newtheorem{theo}{Theorem}
27: \newtheorem{lem*}{Lemma}
28: \def\pn{\medskip\par\noindent}
29: \def\eps{\varepsilon}
30: \def\gmod{\, \hbox{\bf mod}\, }
31: \def\Lim{\mathop{\hbox{lim}}\limits}
32: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
33: \def\abs#1{\left \vert #1 \right \vert}
34: \def\Frac#1#2{{\displaystyle{\frac{#1}{#2}}}}
35: \def\[#1\]{\begin{equation}#1\end{equation}}
36: \def\$#1\${\begin{eqnarray}#1\end{eqnarray}}
37: \def\[#1\]{\begin{eqnarray} #1 \end{eqnarray}}
38: \def\phi{\varphi}
39: \def\p{\vspace{-5.17ex} \hspace{6.5ex}}
40: \def\RR{{\bf R}}
41: \def\bn{\begin{enumerate}}
42: \def\en{\end{enumerate}}
43: %--Parties entieres
44: \def\ent#1#2{{\textstyle{{#1} \overwithdelims[] {#2}}}}
45: \def\Ent#1#2{{\displaystyle{{#1} \overwithdelims[] {#2}}}}
46: \def\pent#1#2{\pe{\frac{#1}{#2}}}
47: \def\Pent#1#2{\pe{\Frac{#1}{#2}}}
48: \def\pe#1{{\left \lbrack #1 \right \rbrack}} % partie entiere
49: \def\Ch#1#2{{\displaystyle{{#1} \choose {#2}}}}
50: \def\comb#1#2{{\textstyle{{#1} \choose {#2}}}}
51: %%%%%%%%%%%%%%%%%%%%{\`u}{\`u}{\`u}
52: \def\Sum{\mathop{\sum}\limits}
53: \def\Oplus{\mathop{\oplus}\limits}
54: \def\Ssum{\mathop{\displaystyle{\sum}}\limits}
55: \def\Cap{\mathop{\cap}\limits}
56: \def\Cup{\mathop{\cup}\limits}
57: \def\Prod{\mathop{\prod}\limits}
58: \def\ssum{\mathop{\sum}\nolimits}
59: \def\pprod{\mathop{\prod}\nolimits}
60: \def\summ#1{\sum_{#1=0}^{\infty}}
61: \def\equi{\mathop{\displaystyle{\simeq}}\limits}
62: %%%%%%%%%%%%%%%%%%{\`u}
63: \def\cadv{c'est-{\`a}-dire, }
64: \def\Pf{{\bf Proof. }}
65: \def\EPf{\hbox{}\hfill$\Box$\vspace{.5cm}\medskip\par\noindent}
66: \def\tiret{--- }
67: \def\pn{\medskip\par\noindent}
68: \def\vect{\hbox{\rm vect }}
69: \def\eps{\varepsilon}
70: \def\NN{{\rm I\!N}} %natuerliche Zahlen
71: \def\RR{{\rm I\!R}} %reelle Zahlen
72: \def\RR{{\bf R}}
73: %\def\NN{\bbbn}
74: %\def\RR{\bbbr}
75: \begin{document}
76: \pagestyle{myheadings}
77: \markboth{P. -V. Koseleff, D. Pecker}{{\em A polynomial parametrization of torus knots}Submitted}
78: \title{A polynomial parametrization of torus knots}
79: \author{P. -V. Koseleff \and D. Pecker}
80: \institute{UPMC Paris 6, 4, place Jussieu, F-75252 Paris Cedex 05, \\
81: %\and P. -V. Koseleff \at Institut de Math{\'e}matiques de Jussieu,\\
82: \email{\tt\{koseleff,pecker\}@math.jussieu.fr}}
83: \maketitle
84: \begin{abstract}
85: For every odd integer $N$ we give an explicit construction of a polynomial
86: curve $\cC(t) = (x(t), y (t))$, where $\deg x = 3$,
87: $\deg y = N + 1 + 2\pent N4$ that has exactly $N$
88: crossing points
89: $\cC(t_i)= \cC(s_i)$ whose parameters satisfy $s_1 < \cdots < s_{N} < t_1
90: < \cdots < t_{N}$. Our proof makes use of the theory of Stieltjes series and Pad{\'e} approximants.
91: This allows us an explicit polynomial parametrization of the torus knot $K_{2,N}$.
92: \end{abstract}
93: {\bf keywords:} {Polynomial curves, Stieltjes series, Pad{\'e} approximant, torus
94: knots}
95: \section{Introduction}
96: Let $N$ be an odd integer.
97: We look for a parametrized curve $\cC(t)= (x(t),y(t))$ of minimal
98: lexicographic degree such that $\cC$ has exactly $N$ crossing points,
99: corresponding to parameters $(s_i, t_i)$ such that
100: \begin{equation}\label{st}
101: \cC(s_i) = \cC(t_i), \
102: s_1< \cdots < s_N
103: < t_1 <\cdots <t_N.
104: \end{equation}
105: Here we look for curves with $\deg x=3$.
106: As a consequence of B{\'e}zout
107: theorem, we have $\deg y \geq N+1$.
108: \begin{figure}[th]
109: \begin{center}
110: \psfrag{a1}{\hspace{-1.0cm}{\small $(s_1,t_1)$}}
111: \psfrag{a2}{\hspace{-.4cm}{\small$(s_2,t_2)$}}
112: \psfrag{a6}{{}}
113: \psfrag{a7}{\hspace{.4cm}{\small$(s_N,t_N)$}}
114: \centerline{\scalebox{.8}{\includegraphics{dessin2.eps}}}
115: \vspace*{8pt}
116: \caption{Curve of degree $(3, 19)$ in logarithmic scale in $y$}
117: \end{center}
118: \end{figure}
119: We have translated this problem into a problem on real roots of
120: certain real polynomials in one variable.
121: In \cite{KP} we proved that if $N>3$, there is no solution with
122: $\deg y = N+1$. We have computed the first examples and we have shown that
123: the minimal degrees are $\deg y = N + 1 + 2 \pent N4$ for $N=3, 5, 7$.
124: \pn
125: The purpose of this paper is to give an explicit construction at any
126: order of such curves with $\deg y = N + 1 + 2 \pent N4$.
127: \pn
128: In section {\bf \ref{cheby}.}, we first recall some properties of the Chebyshev
129: polynomials. Our construction is based on certain relations in the space
130: spanned by some of these polynomials.
131: \pn
132: The explicit construction is given in section {\bf \ref{pc}}. It involves
133: some particular polynomial basis whose existence is proved in section
134: {\bf \ref{pa}.}, using Stieltjes series theory
135: and Pad{\'e} approximation theory (see \cite{BG}).
136: \pn
137: In section {\bf \ref{t2t6}.}, we will show that the algebraic relation
138: between $\cos 2\theta$ and $\cos 6\theta$ may be
139: seen as a Stieltjes series, namely some algebraic hypergeometric
140: function. We will recall some properties of these
141: functions and their approximations by rational functions in section
142: {\bf \ref{pa}.}, the so-called Pad{\'e} approximants.
143: \pn
144: The polynomial curves whose existence are proved are of interest for an
145: explicit polynomial parametrizations of the $(2,N)$-type torus knot $K_{2,N}$ (see
146: \cite{Ad,KP,Mu,RS}). In section {\bf \ref{tk}.}, we give an explicit
147: parametrizations for the knots $K_{2,N}$.
148: They are symmetric with respect to the $y$-axis and of smaller degrees
149: than those already known.
150: \section{Some properties of the Chebyshev polynomials}\label{cheby}
151: \begin{definition}[Monic Chebyshev polynomials]\label{tn}\\
152: If $t= 2 \cos \theta$, let $T_n(t)= 2 \cos ( n \theta )$ and
153: $V_n (t) = \Frac{\sin ((n+1) \theta)}{\sin \theta }$.
154: \end{definition}
155: $T_n$ and $V_n$ are both monic and have degree
156: $n$. It is convenient for our problem to consider them as
157: basis of $\RR[t]$.
158: \pn
159: Looking for a polynomial curve $\cC(t)= (x(t),y(t))$ where $\deg x= 3$,
160: one can suppose that
161: $$
162: x(t) = T_3(t), \
163: y(t) = T_m(t) + a_{m-1}T_{m-1}(t) + \cdots + a_1 T_1(t).
164: $$
165: In \cite{KP} (Lemma {\bf A}) we have shown that
166: if $s \not = t$ are real numbers such that $T_3(s)= T_3(t),$ then
167: for any integer $k$ we have
168: \begin{equation}\label{tv}
169: \Frac{T_k (t) - T_k(s)}{t - s} =\Frac{2}{\sqrt 3} \sin \Frac{k
170: \pi}3 \, V_{k-1}(s+t) = \eps_k V_{k-1}(s+t).
171: \end{equation}
172: We proved the following:
173: \begin{proposition}\label{curve1}
174: Let $\eps_k = \Frac 2{\sqrt 3} \sin \Frac{k\pi}3 = V_{k-1}(1)$ and
175: \begin{equation}\label{R}
176: R_m = \eps_m V_{m-1} + \eps_{m-1} a_{m-1} V_{m-2} + \cdots + \eps_1
177: a_1 V_0.
178: \end{equation}
179: %If $R_m$ has $n$ real roots
180: %$u_1 = 2 \cos \alpha_1 < \cdots < u_N = 2 \cos \alpha_N$, then
181: %$$\cC(t) = (T_3(t), \ T_m(t) + a_{m-1}T_{m-1}(t) + \cdots + a_1 T_1(t))$$
182: %has $N$ crossing points corresponding to parameters
183: %$$s_i = 2 \cos(\alpha_i + \pi/3), \, t_i = 2 \cos(\alpha_i - \pi/3).$$
184: \tiret
185: If $R_m$ has exactly $N$ distinct roots $-1<u_1<\cdots<u_N<1$
186: and no other in $[-2,2]$, then
187: $$\cC(t) = (T_3(t), \ T_m(t) + a_{m-1}T_{m-1}(t) + \cdots + a_1 T_1(t))$$
188: has exactly $N$ crossing points. \\
189: \tiret
190: Let $u_i = 2 \cos \alpha_i$, then
191: \begin{equation}\label{sta}
192: s_i = 2 \cos(\alpha_i + \pi/3), \, t_i = 2 \cos(\alpha_i - \pi/3)
193: \end{equation}
194: are the parameters of the crossing points and satisfy
195: $$s_1 < \cdots < s_N<t_1<\cdots<t_N.$$
196: \end{proposition}
197: %\subsection{Adapted polynomials}
198: We look for polynomials $R_m$ in $\RR[t]$ having $N$ roots that are linear
199: combinations of the $V_k$, where $k$ is not equal to $2 \mod 3$.
200: We will consider separately $E \subset \RR[t]$ spanned by $V_{6k+1}$ and
201: $V_{6k+3}$ and $\tilde E$ spanned by the $V_{6k}$ and $V_{6k+4}$. We
202: first describe these vectorial spaces as direct sums:
203: \begin{lemma}\label{EE}
204: $
205: E = T_1 \cdot \RR[T_6] \oplus T_1\cdot T_2 \cdot \RR[T_6], \quad
206: \tilde E = 1 \oplus T_3 \cdot E.
207: $
208: \end{lemma}
209: \begin{proof}
210: \tiret
211: From $\sin(x+y)-\sin(x-y) = 2 \cos(x) \sin(y)$
212: we deduce that for every integers $n$ and $p$, we have
213: $$
214: V_{n+p} - V_{n-p} = V_{p-1} T_{n+1} .
215: $$
216: We thus deduce that $V_1 =T_1=t$, $V_3 = T_1 \cdot T_2$ and
217: $$%\label{wn}
218: V_{6k+1} - V_{6k-3} = T_1 \cdot T_{6k} , \
219: V_{6k+3}-V_{6k-5} = V_3 \cdot T_{6k} = T_1 \cdot T_2 \cdot T_{6k}.
220: $$
221: From $T_{6k} = T_{k}(T_6)$, we
222: deduce by induction that
223: $$E = T_1 \cdot \RR[T_6] \oplus T_1 \cdot T_2\cdot \RR[T_6].$$
224: \tiret From $\sin(x+y)+\sin(x-y)=
225: 2\cos(y)\sin(x)$, we get
226: $$
227: V_{n+3}+V_{n-3} = T_3 V_{n}
228: $$
229: so
230: $$%\label{twn}
231: V_{6k+6}+V_{6k} = T_3 V_{6k+3}, \,
232: V_{6k+4}+V_{6k-2} = T_3 V_{6k+1}.
233: $$
234: As $V_0 = 1$ and $V_{-2}=-1$, we thus deduce by induction that $\tilde E = 1 \oplus T_3 \cdot E$.
235: \qed%
236: \end{proof}
237: \begin{definition}
238: Let us define for $k \geq 0$,
239: \begin{eqnarray*}
240: \tilde W_{2k} = V_{6k},\,
241: W_{2k} &=& V_{6k+1},\, W_{2k+1} = V_{6k+3}, \,
242: \tilde W_{2k+1} = V_{6k+4}.
243: \end{eqnarray*}
244: \end{definition}
245: We have $\deg W_n =2n+2\pent n2+1$ and
246: $E = \vect(W_k, \, k\geq 0)$.\\
247: We have $\deg \tilde W_n =2n+2\pent{n+1}2$ and
248: $\tilde E = \vect(\tilde W_k, \ k\geq 0)$.
249: \pn
250: Using the Pad{\'e} approximation theory, we will prove in section {\bf
251: \ref{pa}} (p. \pageref{cn})
252: \begin{theorem}\label{2n}
253: There exists a sequence of odd polynomials $C_n$ in $E$ such that
254: \begin{eqnarray*}
255: \vect (W_0, \ldots, W_n) =\vect (C_0, \ldots, C_n),\,
256: C_n = t^{2n+1}F_n, \, F_n(0)=1.
257: \end{eqnarray*}
258: Furthermore $F_n(t)>0$ when $t \in [-2,2]$.
259: \end{theorem}
260: We find, up to some multiplicative constant,
261: \begin{eqnarray*}
262: C_{{0}}&=&t = W_0,\\
263: C_{{1}}&=&{t}^{3}=W_{{1}}+2\,W_{{0}},\\
264: C_{{2}}&=&{t}^{5} \left( {t}^{2}-6 \right)=W_{{2}}-10\,W_{{1}}-16\,W_{{0}},\\
265: C_{{3}}&=&{t}^{7} \left( {t}^{2}-9/2 \right)=W_{{3}}+7/2\,W_{{2}}-15\,W_{{1}}-21\,W_{{0}},\\
266: C_{{4}}&=&{t}^{9} \left( {t}^
267: {4}-12\,{t}^{2}+33 \right)=W_{{4}}-22\,W_{{3}}-56\,W_{{2}}+176\,W_{{1}}+231\,W_{{0}},\\
268: C_{{5}}&=&{t}^{11} \left( {t}^{4}-{\frac {
269: 102}{11}}\,{t}^{2}+{\frac {234}{11}} \right)=
270: W_{{5}}+{\frac {52}{11}}\,W_{{4}}-40\,W_{{3}}-
271: {\frac {910}{11}}\,W_{{2}}+208\,W_{{1}}+260\,W_{{0}}.
272: \end{eqnarray*}
273: We deduce from theorem {\bf \ref{2n}} and lemma {\bf \ref{EE}} the
274: following useful result for the construction of the height function
275: $z(t)$ of our knots (see section {\bf \ref{tk}.}).
276: \begin{corollary}\label{p2n}
277: The sequence
278: $\tilde C_0=1, \, \tilde C_n = -\Frac 13 T_3 C_{n-1}$
279: of even polynomials in $\tilde E$ satisfies:
280: $$
281: \vect(\tilde W_0, \ldots, \tilde W_n) =\vect(\tilde C_0,
282: \ldots, \tilde C_n), \,
283: \tilde C_n = t^{2n} \tilde F_n ,\, \tilde F_n (0) = 1.
284: $$
285: Furthermore $\tilde F_n(t)>0$ when $t\in [-2,2]$.
286: \end{corollary}
287: \section{Construction of the prescribed curves}\label{pc}
288: We will construct polynomials $R_m$ in $E$ with $N=2n+1$ real roots
289: in $[-1,1]$ and no other roots in $[-2,2]$. They will be chosen as a
290: slight deformation of $C_n$.
291: Let us first show properties of the polynomials $C_n$.
292: \begin{lemma}\label{ap}
293: Let $0 < u_1 < \cdots < u_n <1$ be real numbers.
294: For $\eps$ being small enough,
295: \bn
296: \item there exists a unique $(a_0, \ldots, a_{n-1})$ such that
297: $\{0,\pm \eps u_1, \ldots, \pm \eps u_n\}$ are roots in $[-1,1]$ of
298: $$
299: A_n(\eps) = C_n + a_{n-1} C_{n-1} + \cdots + a_0 C_0.
300: $$
301: \item $\{0,\pm \eps u_1, \ldots, \pm \eps u_n\}$ are the only real
302: roots in $[-2,2]$ of $A_n(\eps)$.
303: \en
304: \end{lemma}
305: \begin{proof}
306: Looking for $A_n(\eps)$ with roots $0$ and $\pm \eps u_i$ is equivalent to
307: the linear system
308: $$
309: \left (
310: \begin{array}{cccc}
311: C_{0}(\eps u_1) & C_{1}(\eps u_1) & \cdots & C_{n-1}(\eps u_1) \\
312: C_{0}(\eps u_2) & & \cdots & C_{n-1}(\eps u_2)\\
313: \vdots & & & \vdots\\
314: C_{0}(\eps u_n) & C_{1}(\eps u_n) & \cdots &C_{n-1}(\eps u_n)
315: \end{array}
316: \right )
317: \left (
318: \begin{array}{c}
319: a_0 \\ a_1 \\ \vdots \\ a_{n-1}
320: \end{array}
321: \right )
322: =
323: -
324: \left (
325: \begin{array}{c}
326: C_n(\eps u_1) \\ C_n(\eps u_2) \\ \vdots \\ C_n(\eps u_n)
327: \end{array}
328: \right )
329: $$
330: whose determinant is
331: $
332: \eps^{(1+3+\cdots+2n-1)}
333: \left |
334: \begin{array}{cccc}
335: u_1 F_{0}(\eps u_1) & u_1^{3} F_{1}(\eps u_1) &
336: \cdots & u_1^{2n-1}F_{n-1}(\eps u_1) \\
337: u_2 F_{0}(\eps u_2) & & \cdots & u_2^{2n-1}F_{n-1}(\eps u_2)\\
338: \vdots & & & \vdots\\
339: u_n F_{0}(\eps u_n) & u_n^{3} F_{1}(\eps u_n) & \cdots &
340: u_n^{2n-1} F_{n-1}(\eps u_n)
341: \end{array}
342: \right |.
343: $\\
344: It is equivalent to the classical
345: Vandermonde-type determinant when $\eps \rightarrow 0$:
346: $$
347: \eps^{n^2}
348: \left |
349: \begin{array}{cccc}
350: u_1 & u_1^{3} & \cdots & u_1^{2n-1} \\
351: u_2 & & \cdots & u_2^{2n-1}\\
352: \vdots & & & \vdots\\
353: u_n & u_n^{3} & \cdots & u_n^{2n-1}
354: \end{array}
355: \right |
356: = \eps^{n^2} u_1 \cdots u_n \prod_{1 \leq i < j \leq n} (u_j^2 -
357: u_i^2) \not = 0.
358: $$
359: Therefore, this system has a unique
360: solution.
361: \pn
362: \tiret Using Cramer formulas, we get
363: $
364: a_k(\eps) = \Frac{G_k(\eps)}{G_{n}(\eps)}
365: $
366: where
367: \begin{eqnarray*}
368: G_k (\eps) = \det (C_{i_j} (\eps u_i))_{1 \leq i,j \leq n}
369: \end{eqnarray*}
370: and $i_1 < i_2 < \cdots < i_n \in \{0,\ldots,n\} - \{k\}$.
371: We get
372: \begin{eqnarray*}
373: G_k(\eps) &=&
374: \eps^{2(i_1 + \cdots + i_n)+n}
375: \left |
376: \begin{array}{cccc}
377: u_1^{2i_1+1} F_{i_1}(\eps u_1) & u_1^{2i_2+1} F_{i_2}(\eps u_1) &
378: \cdots & u_1^{2i_n+1}F_{i_n}(\eps u_1) \\
379: u_2^{2i_1+1} F_{i_1}(\eps u_2) & & \cdots & u_2^{2i_n+1}F_{i_n}(\eps u_2)\\
380: \vdots & & & \vdots\\
381: u_n^{2i_1+1} F_{i_1}(\eps u_n) & u_n^{2i_2+1} F_{i_2}(\eps u_n) & \cdots &
382: u_n^{2i_n+1} F_{i_n}(\eps u_n)
383: \end{array}
384: \right |\\
385: &\equi_{\eps \to 0}&
386: \eps^{(n+1)^2-(2k+1)}
387: \left |
388: \begin{array}{cccc}
389: u_1^{2i_1+1} & u_1^{2i_2+1} & \cdots & u_1^{2i_n+1} \\
390: u_2^{2i_1+1} & & \cdots & u_2^{2i_n+1}\\
391: \vdots & & & \vdots\\
392: u_n^{2i_1+1} & u_n^{2i_2+1} & \cdots & u_n^{2i_n+1}
393: \end{array}
394: \right |.
395: \end{eqnarray*}
396: \tiret
397: We thus deduce that $a_k(\eps) = {\cal O}(\eps^{2(n-k)})$ and therefore
398: $$\lim_{\eps \to 0} A_n(\eps) = C_n = t^{2n+1} F_n.$$
399: Let $A_n (\eps)= t\prod_{i=1}^n (t^2-\eps^2 u_i^2) D_n(\eps)$.
400: We deduce that $\Lim_{\eps \to 0} D_n = F_n$.
401: Let $\eps$ be small enough, we get $D_n(t)>0$ for $t \in[-2,2]$
402: because of the compactness of $[-2,2]$.
403: \qed%
404: \end{proof}
405: \begin{proposition}\label{curve2}
406: Let $N=2n+1$ be an odd integer. There exists a curve $\cC(t)=(x(t),y(t))$,
407: where $\deg x = 3$ and $\deg y = N+2 \pent N4 +1$, such that $\cC$ has
408: exactly $N$ crossing points
409: corresponding to parameters $(s_i, t_i)$ such that
410: \begin{equation}
411: \cC(s_i) = \cC(t_i), \
412: s_1< \cdots < s_N
413: < t_1 <\cdots <t_N. \nonumber
414: \end{equation}
415: \end{proposition}
416: \begin{proof}
417: Let $N=2n+1$. Let us choose $\eps$ and $0<u_1<\cdots<u_n$, such that
418: there exists a polynomial $A_n(\eps) \in \vect(W_0, \ldots, W_n)$
419: having exactly $N$ distinct roots $\{0, \pm \eps u_1, \ldots, \pm \eps
420: u_n\}$ in $[-2,2].$ It has degree $\deg W_n = 2n + 2 \ent n2 +1 =
421: N+2\pent N4 = m-1$. We have
422: $$
423: A_n = W_n + \cdots + a_0 = V_{m-1} + \eps_{m-1} a_{m-1} V_{m-2}
424: +\cdots + a_2 V_1.
425: $$
426: Using proposition {\bf \ref{curve1}}, the curve
427: $$
428: x(t) = T_3(t) , \ y(t) = \eps_m T_m(t) + a_{m-1}T_{m-1}(t) + \cdots + a_2 T_2(t)
429: $$
430: has the required properties.
431: \qed%
432: \end{proof}
433: \begin{example}[$N=9$]
434: We chose
435: $$y(t) =
436: -{\Frac {27}{10}}\,T_{{14}}
437: +10\,T_{{12}}
438: -23\,T_{{10}}
439: +42\,T_{{8}}
440: -64\,T_{{6}}
441: +85\,T_{{4}}
442: -100\,T_{{2}}
443: +112.$$
444: The roots of $Q$ are $u_0 = 0, \pm u_1 = .355, \pm u_2 =
445: .584, \pm u_3 = .785, \pm u_4= 1.073$.
446: We obtain a polynomial parametrization of degree $(3,14)$.
447: Note that we choose $u_4>1$ for a nicer picture (see
448: fig. \ref{fig9}). Note that the parameters of the crossing points
449: satisfy $s_1 < \cdots < s_9 < t_1 < \cdots <t_9$.
450: \begin{figure}[th]
451: \begin{center}
452: \begin{tabular}{ccc}
453: {\scalebox{.4}{\includegraphics{curve9b-1.eps}}}
454: &\quad&
455: {\scalebox{.4}{\includegraphics{curve9b-2.eps}}}\\
456: Bottom view & & Zoom on the bottom view\\
457: {\scalebox{.4}{\includegraphics{curve9b-3.eps}}}
458: &\quad&
459: {\scalebox{.4}{\includegraphics{curve9b-4.eps}}}\\
460: logarithmic scale & & $\Frac{x(s)-x(t)}{s-t}=0, \
461: \Frac{y(s)-y(t)}{s-t}=0, s<t$
462: \end{tabular}
463: \end{center}
464: \vspace{8pt}
465: \caption{$N=9$. Curve of degree $(3,14)$\label{fig9}}
466: \end{figure}
467: \end{example}
468: \section{Construction of the torus knots}\label{tk}
469: If $N=2n+1$ is odd, the torus knot $K_{2,N}$ of type $(2,N)$ is the
470: boundary of a Moebius band twisted $N$ times (see \cite{Ad,Mu} and
471: fig. (\ref{fig1})).
472: \begin{figure}[th]
473: \centerline{
474: {\scalebox{.8}{\includegraphics{moeb32.eps}}} \quad
475: {\scalebox{.8}{\includegraphics{moeb52.eps}}} \quad
476: {\scalebox{.8}{\includegraphics{moeb72.eps}}}}
477: \vspace{10pt}
478: \centerline{
479: {\scalebox{.8}{\includegraphics{moeb32-n.eps}}} \quad
480: {\scalebox{.8}{\includegraphics{moeb52-n.eps}}} \quad
481: {\scalebox{.8}{\includegraphics{moeb72-n.eps}}}}
482: \vspace*{8pt}
483: \caption{$K_{2,N}$, $N=3,5,7$.\label{fig1}}
484: \end{figure}
485: The purpose of this section is to give an explicit construction of a
486: polynomial curve $\cC(t) = (x(t), y(t), z(t))$ that
487: is equivalent (in the one point compactification $ {\bf S}^3$ of the
488: space $ {\RR}^3$) to the torus knot $K_{2,N}$.
489:
490: Vassiliev (see \cite{Va}) proved that any non-compact knot type can be
491: obtained from a polynomial embedding
492: $ \ t \mapsto ( f(t), g(t), h(t) ), t \in \RR$,
493: using the Weierstrass approximation theorem.
494:
495: Shastri \cite{Sh} gave a detailed proof of this theorem,
496: and a simple polynomial parametrizations of the trefoil and of the
497: figure eight knot.
498: A. Ranjan and R. Shukla \cite{RS} have found small degree
499: parametrizations for $K_{2,N}, N$ odd. They proved that these knots
500: can be attained from polynomials of degrees $ (3, 2N-2, 2N-1).$
501:
502: In \cite{KP}, we proved that it is not possible to attain the torus
503: knot $K_{2,N}$ with polynomial of degrees $(3,N+1,m)$ when $N>3$. We
504: gave explicit parametrization of degrees $(3,N+ 2\pent
505: N4+1,N+2\pent{N+1}4)$ for $N=3,5,7,9$ and showed that they were of
506: minimal lexicographic degree for $N\leq 7$.
507: \pn
508: A sufficient condition is to construct a parametrized curve
509: $\cC(t)=(x(t),y(t),z(t))$
510: such that $(x(t),y(t))$ has exactly $N=2n+1$ crossing points corresponding to
511: parameters
512: $$s_1 < \cdots < s_N < t_1 < \cdots < t_N$$
513: and such that
514: $$x(t_i)=x(s_i), \ y(t_i)=y(s_i),
515: \, (-1)^i(z(t_i)-z(s_i))>0, \ i = 1, \ldots , N.$$
516: We look first for minimal degree in $x$.
517: $x(t)$ must be nonmonotonic and therefore has degree at least 2. In case when
518: $\deg x=2$, we would have constant $t_i + s_i$ and not the condition
519: (\ref{st}). We will give a construction for $\deg x=3$.
520: \begin{proposition}\label{kn}
521: For any odd integer $N=2n+1$, there exists
522: a curve $\cC (t) = (x(t),y(t),z(t))$ of degree
523: $(3,N+ 2\pent N4+1,N+2\pent{N+1}4)$
524: such that the curve $(x(t),y(t))$ has exactly $N$ crossing points
525: $$x(t_i)=x(s_i), \ y(t_i)=y(s_i), \,
526: s_1< \cdots < s_N < t_1 <\cdots <t_N
527: $$
528: and
529: $$(-1)^i (z(t_i)-z(s_i)) >0 , \ i = 1, \ldots , N.$$
530: \end{proposition}
531: \begin{proof}
532: \tiret
533: Following the construction of section {\bf \ref{pc}.},
534: we first choose $\eps$ to be small enough and
535: $0 < c_1 < \cdots < c_n < \eps < 1/2$, such that
536: $$
537: \left \vert
538: \begin{array}{cccc}
539: C_{0}(c_1) & C_{1}(c_1) & \cdots & C_{n-1}(c_1) \\
540: C_{0}(c_2) & & \cdots & C_{n-1}(c_2)\\
541: \vdots & & & \vdots\\
542: C_{0}(c_n) & C_{1}( c_n) & \cdots &C_{n-1}(c_n)
543: \end{array}
544: \right \vert \not = 0.
545: $$
546: Consider
547: \begin{eqnarray*}
548: u_{n+1} = 0, \, u_i = 2\cos \alpha_i = - c_{n+1-i},\,
549: u_{n+1+i} = 2\cos \alpha_{n+1+i} = c_i,
550: \, i = 1,\ldots, n.
551: \end{eqnarray*}
552: We thus have $-1< u_1 <\ldots < u_N <1$.
553: Let
554: $$s_i = 2 \cos(\alpha_i + \pi/3), \, t_i = 2 \cos(\alpha_i - \pi/3),
555: \, i =1, \ldots N.$$
556: %\pn
557: \tiret
558: Using the proposition (\ref{curve2}), there is a polynomial
559: $$
560: y(t) = T_m(t) + a_{m-1}T_{m-1}(t) + \cdots + a_1 T_1(t),
561: $$
562: of degree $m = N + 2 \pent N4+1$ such that
563: $$x(t_i)=x(s_i), \ y(t_i)=y(s_i), \, i =1, \ldots N.$$
564: %\pn
565: \tiret As for lemma {\bf \ref{ap}}, there exists a unique
566: $(b_0, \ldots, b_{n})$
567: such that
568: $$
569: B_n = b_n \tilde C_n + b_{n-1} \tilde C_{n-1} + \cdots + b_{0} \tilde C_0
570: $$
571: satisfies $B_n(u_i) = (-1)^i, \, i=1,\ldots N$. Namely,
572: $(b_0, \ldots, b_{n})$ is the solution of the system
573: $$
574: b_n \tilde C_n (u_i) + b_{n-1} \tilde C_{n-1}(u_i) + \cdots + b_{0}
575: \tilde C_0(u_i) = (-1)^i, \, i=1,\ldots, N.
576: $$
577: Because $u_i = -u_{N+1-i}$ and $\tilde C_k$ are even polynomials, the
578: system is equivalent to
579: $$
580: \left (
581: \begin{array}{cccc}
582: \tilde C_{0}(u_{n+1}) & \tilde C_{1}(u_{n+1}) & \cdots & \tilde C_{n}(u_{n+1}) \\
583: \tilde C_{0}(u_{n+2}) & & \cdots & \tilde C_{n}(u_{n+2})\\
584: \vdots & & & \vdots\\
585: \tilde C_{0}(u_N) & \tilde C_{1}(u_N) & \cdots &\tilde C_{n}(u_N)
586: \end{array}
587: \right )
588: \left (
589: \begin{array}{c}
590: b_0 \\ b_1 \\ \vdots \\ b_{n}
591: \end{array}
592: \right )
593: =
594: \left (
595: \begin{array}{c}
596: (-1)^{n+1} \\ (-1)^{n+2}\\ \vdots \\ (-1)^{N}
597: \end{array}
598: \right ).
599: $$
600: From $\tilde C_0 = 1$, $\tilde C_n = -\Frac 13 T_3 C_{n-1}$ and
601: $C_k(u_{n+1})=0$, we deduce that the determinant of the previous system is
602: $$
603: \pm\Frac{1}{3^n} T_3(c_1) \cdots T_3(c_n)
604: \left \vert
605: \begin{array}{cccc}
606: C_{0}( c_1) & C_{1}(c_1) & \cdots & C_{n-1}( c_1) \\
607: C_{0}( c_2) & & \cdots & C_{n-1}(c_2)\\
608: \vdots & & & \vdots\\
609: C_{0}(c_n) & C_{1}( c_n) & \cdots &C_{n-1}(c_n)
610: \end{array}
611: \right \vert \not = 0.
612: $$
613: $B_n$ is a linear combination of $(\tilde W_n,\ldots,
614: \tilde W_0)$ and it has degree $m'=N+2\pent{N+1}4$:
615: $$
616: B_n = b'_{m'} V_{m'} + \cdots + b'_0 V_0.
617: $$
618: Consider now $z(t) = \eps_{m'+1} b'_{m'} T_{m'+1} + \cdots + \eps_1 b'_0
619: T_1$, we have using eq. (\ref{tv}):
620: $$
621: \Frac{z(t_i) -z(s_i)}{t_i -s_i}=B_n(u_i) = (-1)^i.
622: $$
623: Because $t_i > s_i$ we deduce that $z(t_i)-z(s_i)$ has alternate signs.
624: \qed%
625: \end{proof}
626: \section{$T_2$ as a power series of $T_6+2$}\label{t2t6}
627: Looking for identities in the vectorial space $\RR[T_6] + T_2 \cdot
628: \RR[T_6]$, we show first some relation between $T_2$ and $T_6$.
629: \begin{lemma}\label{t1t3}
630: For $t \in [-1,1]$, we have
631: $$
632: T_2+2 =4\sin^2 \left (\Frac 13 \arcsin \sqrt{\Frac{T_6+2}4} \right )
633: % = \phi\left (\Frac{T_6+2}4 \right )
634: .
635: $$
636: \end{lemma}
637: \begin{proof}
638: Let $t \in [-1,1]$ and $x \in [\pi/3,2\pi/3]$ such that $t=2\cos x$. \\
639: %We have $\Frac 12 T_3 = \cos(3x) \in [-1,1]$.
640: We get $3x-\pi \in [0,\pi]$ and $\cos (3x -\pi)=-\Frac 12 T_3$ so
641: $$x=
642: \Frac{\pi}3+\Frac 13 \arccos \left (-\Frac{T_3}2\right )
643: = \Frac{\pi}3 + \Frac 13 \left (\Frac{\pi}2 + \arcsin \Frac{T_3}2 \right)
644: = \Frac{\pi}2 + \Frac 13 \arcsin \Frac{T_3}2.$$
645: We thus have
646: $$
647: T_1 = 2\cos \left (\Frac{\pi}2 +\Frac 13 \arcsin \Frac{T_3}2 \right
648: ) = - 2\sin\left (\Frac 13 \arcsin \Frac{T_3}2 \right ).
649: $$
650: We thus deduce the lemma from $T_2 = T_1^2 -2$ and $T_6 = T_3^2 -2$.
651: \qed%
652: \end{proof}
653: \begin{lemma}
654: Let $\phi(u) = 4\sin^2 \left (\Frac 13 \arcsin \sqrt{u} \right )$.
655: For $u \in [0,1]$, we have
656: $$
657: %\Frac 49 \sum_{n \geq 1} \Frac{(3n-2)!}{n! (2n-1)!}
658: %\left (\Frac{4u}{27}\right )^n =
659: \phi(u) =
660: \sum_{n \geq 1} \phi_n u^n
661: \hbox{ where }
662: \phi_1 = \Frac 49, \,
663: \phi_{n+1} = \Frac 29 \Frac{(3n+1)(3n-1)}{(n+1)(2n+1)}\phi_n.
664: $$
665: \end{lemma}
666: %\def\Dron#1#2{{\Frac{\partial#1}{\partial#2}}}
667: %\def\dron#1#2{{\frac{\partial#1}{\partial#2}}}
668: %\def\dd#1#2{{\frac{\rm d}\,#1}{{\rm d}\,#2}}}
669: %\def\Dd#1#2{{\Frac{\rm d}\,#1}{{\rm d}\,#2}}}
670: \begin{proof}
671: %Let $u= \Frac 14 T_3^2$ and $v=T_1^2 = t^2.$
672: We have
673: $\phi(u)
674: % = 4 \sin^2 \left ( \Frac 13 \arcsin \sqrt u \right )
675: = 2 - 2 \cos \left ( \Frac 23 \arcsin \sqrt u \right )
676: .$
677: We deduce that
678: \[\label{sysd}
679: \left \{
680: \begin{array}{rcl}
681: \phi(u)
682: &=& - 2 A+2\\[5pt]
683: \Frac {d}{d{u}}\phi(u)
684: &=& \Frac 23 B\\[5pt]
685: \Frac{d^{2}}{d{u}^{2}}\phi(u) &=&
686: - \Frac 29 \Frac 1{(u-u^2)} A+
687: \Frac 13 \Frac{(2u-1)}{(u-u^2)} B
688: \end{array}
689: \right .
690: \]
691: where
692: $A=
693: \cos \left (
694: \Frac 23 \arcsin \sqrt u \right )$ and
695: $B=\Frac{\sin \left (\Frac 23 \arcsin \sqrt u \right )}{\sqrt{u-u^2}}$.
696: \pn
697: Eliminating $A$ and $B$ from system (\ref{sysd}), we find that
698: \[\label{ode}
699: -4+2\,\phi \left( u \right) + 9 \left (1-2\,u \right) {\frac {d}{du}}\phi
700: \left( u \right) + 18\left( u-u^2 \right )
701: {\frac {d^{2}}{d{u}^{2}}}
702: \phi \left( u \right) = 0.
703: \]
704: $\phi$ has a power series expansion and we get from (\ref{ode})
705: $$
706: \phi_0 = 0 , \, \phi_1 = \Frac 49, \,
707: \phi_{n+1} = \Frac 29 \Frac{(3n+1)(3n-1)}{(n+1)(2n+1)}\phi_n.
708: $$
709: \qed%
710: \end{proof}
711: \begin{remark}
712: There is no need to know explicitely $\phi$ with the lemma
713: {\bf \ref{t1t3}}. One can see from $4u=v(v-3)^2$
714: that $\phi$ is an algebraic
715: function. It is therefore the solution of a differential equation we
716: can find using Euclid algorithm. Recursion formula for the $\phi_n$
717: and the differential equation can be easely obtained using
718: the {\sc Maple} package {\tt gfun} (see \cite{SZ}).
719: \end{remark}
720: \begin{definition}
721: Let $\Delta$ defined by $\Delta f_n = f_{n+1}-f_{n}$. We say that
722: $f_n$ is totally monotone when
723: for every integer $k$ and every $n \geq 1$, we have
724: $$(-1)^k \Delta^k f_n > 0.$$
725: \end{definition}
726: \begin{example}\label{ex1}
727: \tiret Let $f_n=\exp(-n)$. We get $(-1)^k \Delta^k f_n = f_n (1-1/e)^k$.
728: \pn
729: \tiret Let $f_n = \Frac 1n$. We get $(-1)^k \Delta^k f_n = f_n \Frac
730: 1{{n+k \choose k}}.$
731: \pn They are both totally monotone.
732: \end{example}
733:
734: \def\NN{{\rm I\!N}} %natuerliche Zahlen
735: \begin{proposition}
736: $\phi_n$ is totally monotone.
737: \end{proposition}
738: \begin{proof}
739: We will show that $$(-1)^k \Delta^k \phi_n = \phi_n \Frac{P_k(n)}{(n+1)
740: \cdots (n+k) \cdot (2n+1)\cdots (2n+2k-1)}>0.$$
741: \tiret
742: We get
743: $$\Delta \phi_n = \phi_{n+1} - \phi_n = \phi_n \left (
744: \Frac{2
745: (3n-1)(3n+1)}{9(n+1)(2n+1)} - 1\right ) =
746: - \phi_n \Frac{3 n+11/9}{(n+1)(2n+1)}. $$
747: Suppose now that
748: $(-1)^k \Delta^k\phi_n = \phi_n \Frac{P_k(n)}{(n+1)
749: \cdots (n+k) \cdot (2n+1)\cdots (2n+2k-1)}.$
750: We thus deduce
751: \begin{eqnarray*}
752: (-1)^{k+1} \Delta^{k+1}\phi_n &=&
753: -\Delta \left [
754: \phi_n \Frac{P_k(n)}{(n+1)
755: \cdots (n+k) \cdot (2n+1)\cdots (2n+2k-1)}
756: \right ]\nonumber\\
757: &=&
758: \phi_n \Frac{P_k(n)}{(n+1) \cdots (n+k) \cdot (2n+1)\cdots (2n+2k-1)}
759: - \nonumber\\
760: &&\quad
761: \phi_{n+1} \Frac{P_k(n+1)}
762: {(n+2) \cdots (n+k+1) \cdot (2n+3)\cdots (2n+2k+1)}\nonumber\\
763: &=&
764: \phi_n \Frac{(n+k+1)(2n+2k+1) P_k(n) - 2(n^2-1/9) P_{k}(n+1)}
765: {(n+1) \cdots (n+k+1) \cdot (2n+1)\cdots (2n+2(k+1)-1)}.
766: \end{eqnarray*}
767: We thus obtain
768: $$
769: (-1)^k \Delta^k \phi_n = \phi_n \Frac{P_k(n)}{(n+1)
770: \cdots (n+k) \cdot (2n+1)\cdots (2n+2k-1)},
771: $$
772: where $P_0 = 1$ and
773: $$
774: P_{k+1} (n) = {(n+k+1)(2n+2k+1) P_k(n) - 2(n^2-1/9) P_{k}(n+1)}.
775: $$
776: %\pn
777: \tiret
778: We will show now by induction that $P_k = a_k X^k + \cdots + a_0$
779: where $a_k >0$. Suppose it is true for a given $k$, we thus deduce that
780: \[
781: P_{k+1} &=& (X+k+1)(2X+2k+1) (a_k X^k + a_{k-1} X^{k-1} + \cdots ) -
782: \nonumber\\
783: &&\quad \quad
784: 2(X^2-1/9) (a_k X^k + (a_{k-1} + k a_k) X^{k-1} + \cdots ) \nonumber\\
785: &=& 2 a_k X^{k+2} + \left ((4k+3) a_k + 2 a_{k-1}) \right ) X^{k+1} + \cdots
786: - \nonumber\\
787: &&\quad \quad
788: \left [ 2 a_k X^{k+2} + (2 a_{k-1} + 2 k a_k) X^{k+1} + \cdots \right
789: ] \nonumber\\
790: &=&
791: (2k+3) a_k X^{k+1} + \cdots .
792: \]
793: $P_k$ is a polynomial of degree $k$ whose leading coefficient is
794: $1 \cdot 3 \cdots (2k+1)$.
795: \def\cP{{\cal P}}
796: \pn
797: \tiret
798: Let us prove now by induction the following
799: $$
800: (-1)^i P_k(-i) > 0, \, i = 0, \ldots, k.
801: $$
802: This is true for $k=0$.
803: \pn
804: Suppose now it is true for $P_k$. Intermediate values theorem says that
805: $P_k$ has exactly $k$ real roots in $]-k,0[$, so
806: $P_k (x) >0$ when $x\geq 0$ or when $x+k\leq 0$.
807: \pn
808: Let us compute
809: $$
810: P_{k+1}(0) = (k+1) (2k+1) P_{k}(0) + 2/9 P_k(1) > 0
811: $$
812: For $i = 1, \ldots, k$~:
813: $$
814: (-1)^i P_{k+1}(-i) = (k-i+1)(2(k-i)+1) (-1)^i P_k (i) + 2 (i^2 - 2/9)
815: (-1)^{i-1} P_k(-(i-1)) > 0.
816: $$
817: For $i = - (k+1)$ we get
818: $$
819: (-1)^{k+1} P_{k+1}(-(k+1)) = 0 - 2 ((k+1)^2 - 2/9) (-1)^{k+1} P_k(-k) > 0
820: $$
821: We thus deduce that $(-1)^i P_{k+1}(-i) >0$ for $i = 0,
822: \ldots , k+1$.
823: \pn
824: \tiret
825: We thus deduce that $P_k$ has exactly $k$ roots in $]-k,0[$ so
826: $P_k(n)$ is nonnegative for any integer $n$.
827: \qed%
828: \end{proof}
829: \begin{definition}\label{stieltjes}
830: $f(z) = \sum_{n\geq 1} f_n z^n$ is a Stieltjes series
831: if for every $n \geq 1$ and $m \geq 0$, one has
832: $$
833: \left |
834: \begin{array}{cccc}
835: f_n & f_{n+1} & \cdots & f_{n+m}\\
836: f_{n+1}& f_{n+2} & \cdots & f_{n+m+1}\\
837: \vdots & & & \vdots\\
838: f_{n+m} & f_{n+m+1} & \ldots & f_{n+2m}
839: \end{array}
840: \right |>0.
841: $$
842: \end{definition}
843: \begin{remark}
844: This last condition is related to the problem of Hamburger moments. It
845: is the Stieltjes condition.
846: The totally monotonicity is related to the
847: Hausdorff condition (see \cite{Ha}).
848: \end{remark}
849: \def\d{\hbox{\rm d}\,}
850: The Hausdorff condition and the Stieltjes condition are equivalent
851: if the series is not a rational function (see \cite{BG}, p. 194 and the proof of
852: Sch{\"o}nberg, \cite{Wa}, p. 267 or \cite{Sc}).
853: We thus deduce that
854: \begin{theorem}
855: $\phi(z)= \sum_{n\geq 1} \phi_n z^n$ is a Stieltjes series.
856: \end{theorem}
857: \begin{proof}
858: $\phi(u)$ is an algebraic function that satisfies
859: $4u=\phi(\phi-3)^2$. Suppose that $\phi = p/q$ where $p(u)$ and $q(u)$
860: are relatively prime polynomials in $u$, then we would have
861: $4uq^3-p^3+6p^2q-9pq^2 =0$ and $p$ would divide $u$ and $q$
862: would divide $1$. We would have $\phi(u) = \lambda u$ and it is not
863: the case. Thus $\phi$ is not a rational function and is therefore a Stieltjes function.
864: \qed%
865: \end{proof}
866: \begin{remark}
867: In example (\ref{ex1}), the sequence $\exp(-n)$ is totally monotonic. But
868: $\sum_n \exp(-n) z^n = \Frac 1{1-e\cdot z}$ is a rational function and the
869: condition (\ref{stieltjes}) does not hold.
870: \end{remark}
871: \begin{remark}
872: $\phi(u)= 2-2 F(1/3,-1/3,1/2;u)$ where $F(a,b,c;z)$ is the hypergeometric
873: function. It results from eq. (\ref{ode}) that is known as the
874: hypergeometric equation (\cite{BG,Wa})
875: $$
876: \left( u-u^2 \right ) {\frac {d^{2}}{d{u}^{2}}} f\left( u \right)
877: + \left (c -(1+a+b)\,u \right) {\frac {d}{du}}f \left( u \right)
878: - ab \,f \left( u \right)
879: = 0.
880: $$
881: for $\phi - 2 = -2f$, $a=-b=\Frac 13, c=\Frac 12$.
882: \end{remark}
883: \section{Pad{\'e} approximation}\label{pa}
884: Rational approximations of Stieltjes series have remarkable
885: properties. Let us remind the following construction of Pad{\'e} approximants:
886: \begin{theorem}[Pad{\'e} approximant]\label{pade-t}
887: Let $f(x) = \sum_{k \geq 1} f_k x^k$ be a Stieltjes series and consider two
888: integers $m\leq n$.
889: There is a unique solution $(P_n, Q_m) \in \RR_n[x] \times
890: \RR_{m}[x]$, such that
891: \begin{equation}
892: Q_m(0)=1,\, P_n - f Q_{m} = 0 \mod{x^{n+m+1}}.\label{pade}
893: \end{equation}
894: Furthermore we have $\deg P_n=n$ and $\deg Q_m=m$.
895: \end{theorem}
896: \begin{proof}
897: Let us write
898: $$
899: P_n = p_0 + \cdots + p_n x^n, \, Q_m = q_0 + q_1 x + \cdots + q_m x^n.
900: $$
901: Eq. (\ref{pade}) gives
902: \begin{eqnarray}
903: &&\left \{
904: \begin{array}{rcl}
905: p_0 &=& f_0, \\
906: p_1 &=& f_0 q_1+f_1 q_0\\
907: & \vdots & \\
908: p_n & = & f_{n-m} q_m +f_{n-m+1} q_{m-1} + \cdots + f_{n}q_0,
909: \end{array}\right . \label{nm}\\
910: &&
911: \left \{
912: \begin{array}{rcl}
913: 0 &= & f_{n-m+1} q_{m} + f_{n-m+2} q_{m-1} + \cdots + f_{n+1} q_0 \\
914: 0 &= & f_{n-m+2} q_{m} + f_{n-m+3} q_{m-1} + \cdots + f_{n+2} q_0 \\
915: &\vdots& \\
916: 0 &=& f_{n} q_m + f_{n+1} q_{m-1} + \cdots + f_{m+n} q_0.
917: \end{array}\right .\label{mm}
918: \end{eqnarray}
919: The last $m \times m$ system (\ref{mm}) is
920: \begin{eqnarray}
921: \left (
922: \begin{array}{cccc}
923: f_{n-m+1} & f_{n-m+2} & \cdots & f_{n}\\
924: f_{n-m+2}& f_{n-m+3} & \cdots & f_{n+1}\\
925: \vdots & & & \vdots\\
926: f_{n} & f_{n+1} & \ldots & f_{m+n-1}
927: \end{array}
928: \right )
929: \left (
930: \begin{array}{c}
931: q_m \\
932: q_{m-1}\\
933: \vdots\\
934: q_1
935: \end{array}
936: \right )
937: =
938: -q_0 \left (
939: \begin{array}{c}
940: f_{n+1} \\
941: f_{n+2}\\
942: \vdots\\
943: f_{m+n}
944: \end{array}
945: \right )
946: \label{sys}
947: \end{eqnarray}
948: and therefore has a unique solution because $f$ is a Stieltjes series
949: and $q_0=1$.
950: The first system (\ref{nm}) is then solved
951: for $p_0, \ldots, p_n$.
952: \pn
953: \tiret
954: System (\ref{mm}) may be also written
955: \begin{eqnarray*}
956: \left (
957: \begin{array}{cccc}
958: f_{n-m+2} & f_{n-m+3} & \cdots & f_{n+1}\\
959: f_{n-m+3}& f_{n-m+4} & \cdots & f_{n+2}\\
960: \vdots & & & \vdots\\
961: f_{n+1} & f_{n+2} & \ldots & f_{m+n}
962: \end{array}
963: \right )
964: \left (
965: \begin{array}{c}
966: q_{m-1} \\
967: q_{m-2}\\
968: \vdots\\
969: q_0
970: \end{array}
971: \right )
972: =
973: -q_m \left (
974: \begin{array}{c}
975: f_{n-m+1} \\
976: f_{n-m+2}\\
977: \vdots\\
978: f_{n}
979: \end{array}
980: \right ).
981: \end{eqnarray*}
982: We thus deduce that if $q_m=0$ then $Q_m=0$ and $Q_m(0)=0$.
983: \pn
984: \tiret
985: With the last equation of (\ref{nm})
986: and (\ref{mm}), we have
987: \begin{eqnarray*}
988: \left (
989: \begin{array}{cccc}
990: f_{n-m} & f_{n-m+1} & \cdots & f_{n}\\
991: f_{n-m+1}& f_{n-m+2} & \cdots & f_{n+1}\\
992: \vdots & & & \vdots\\
993: f_{n} & f_{n+1} & \ldots & f_{m+n}
994: \end{array}
995: \right )
996: \left (
997: \begin{array}{c}
998: q_m \\
999: q_{m-1}\\
1000: \vdots\\
1001: q_0
1002: \end{array}
1003: \right )
1004: =
1005: \left (
1006: \begin{array}{c}
1007: p_n \\
1008: 0\\
1009: \vdots\\
1010: 0
1011: \end{array}
1012: \right ) .
1013: \end{eqnarray*}
1014: We thus deduce that $p_n \not =0$.
1015: \qed%
1016: \end{proof}
1017: \begin{remark}
1018: The system (\ref{sys}) shows that if $Q_m(0)=0$, then $Q_m=0$.
1019: \end{remark}
1020: \begin{definition}
1021: We say that
1022: $f^{[n/m]} = P_n/Q_m$ is the Pad{\'e} approximant of order $(n,m)$ of $f$.
1023: \end{definition}
1024: We will make use of a very useful theorem concerning Stieltjes
1025: series.
1026: \begin{theorem} \label{bmt}
1027: Let $f(x)$ be a Stieltjes series with radius of convergence $R$ and let
1028: us denote by $f^{[n/m]}$ its Pad{\'e} approximant $P_n/Q_m$. Then
1029: \begin{enumerate}
1030: \item \label{poles}
1031: $Q_m$ has exactly $m$ real roots in $]R, + \infty [.$
1032: \item Let $f^{[n/m]}(x) = \sum_{k \geq 1} f_k^{[n/m]} x^k$. We have
1033: \begin{enumerate}
1034: \item for $1 \leq k \leq n+m$, $0 < f_k^{[n/m]} = f_k$. \label{2a}
1035: \item $0 \leq f_{n+m+1}^{[n/m]} < f_{n+m+1}$.\label{2b}
1036: \item for $k \geq n+m+1$, $0 \leq f_k^{[n/m]} \leq f_k$.\label{2c}
1037: \end{enumerate}
1038: \end{enumerate}
1039: \end{theorem}
1040: \begin{proof}
1041: The assertion (\ref{poles}) is proved in \cite{BG}, p. 220. Note that
1042: the authors use the function $f(-z)$.
1043: Assertion (\ref{2a}) is a consequence of the Pad{\'e} approximation
1044: definition. Assertion (\ref{2c}) is proved in \cite{BG}, p. 212. Note
1045: that the authors have shown that $0\leq f_k^{[n/m]} \leq f_k$.
1046: Suppose now that $f_{n+m+1}^{[n/m]} = f_{n+m+1}$. From theorem
1047: {\bf \ref{pade-t}}, we would have $\deg P_n + \deg Q_m = n+m+1$ and this is
1048: not the case. We thus have $f_{n+m+1}^{[n/m]} < f_{n+m+1}$.
1049: \qed%
1050: \end{proof}
1051: We thus deduce
1052: \begin{corollary}\label{cnm}
1053: Let $m\leq n$. There are polynomials $P_n \in \RR_n[u]$,
1054: $Q_m \in \RR_m[u]$ and $F_{n,m} \in \RR[v]$ such that
1055: $$
1056: Q_m(u)v-P_n(u) = v^{n+m+1} F_{n,m}(v),
1057: $$
1058: where $F_{n,m}(0)=1$.
1059: Furthermore, we have $F_{n,m}(v)>0$ when $v \in [0,1]$,
1060: $\deg P_n=n$, $\deg Q_m = m$ and $Q_m(u)>0$ for $u \in [0,1]$.
1061: \end{corollary}
1062: \begin{proof}
1063: $\phi$ is a Stieltjes series and because
1064: $\Frac{\phi_{n+1}}{\phi_n} \equi_{n \to \infty} 1 -\Frac 3{2n}$, we
1065: deduce that its radius of convergence $R$ is 1 and that
1066: $\sum_{n \geq 1} \phi_n = \phi(1) = 1$.
1067: Let $\phi^{[n/m]} = {P_n}/{Q_m}$ be the Pad{\'e} approximant of
1068: $\phi$, we deduce that
1069: $$
1070: \phi(u) - \phi^{[n/m]}(u) = \sum_{k\geq n+m+1} (\phi_k - \phi_k^{[n/m]})u^k
1071: = u^{n+m+1} \psi_{n,m}(u), \ 0 \le u \leq 1.
1072: $$
1073: We have $\psi_{n,m}(u)>0$ for $u \in [0,1])$, from theorem
1074: {\bf\ref{bmt}}, (\ref{2b}).
1075: From $Q_m(0)=1$ and theorem {\bf \ref{bmt}}, (\ref{poles}) we get
1076: $Q_m(u)>0$ for $u \in [0,1]$, and
1077: $$
1078: vQ_m(u)-P_n(u)=u^{n+m+1} \psi_{n,m}(u) Q_m(u) >0.
1079: $$
1080: On the other hand, as $4 u = v(v-3)^2 \equi_{v\to 0} 9v$, we deduce that
1081: $vQ_m(u)-P_n(u)$ is a polynomial in $v$ with $0$ as root of order
1082: $n+m+1$. We deduce that
1083: $$
1084: vQ_m(u)-P_n(u) = v^{n+m+1} F_{n,m}(v),$$
1085: where $F_{n,m}$ is a polynomial.
1086: \qed%
1087: \end{proof}
1088: We deduce
1089: \begin{proposition}\label{cn}
1090: There exists a family $C_n$ in $\vect (W_0, \ldots, W_n)$,
1091: such that
1092: $$
1093: C_n = t^{2n+1} F_n, \ F_n(0)= 1.
1094: $$
1095: Furthermore, $\deg C_n = 2n+2\pent n2+1$ and $F_n(t)>0$ for $t \in [-2,2]$.
1096: \end{proposition}
1097: \begin{proof}
1098: Let us consider
1099: $$
1100: C_{k,l} (t) = v Q_l(u) - P_k(u)= v^{k+l+1}F_{k,l}(v)
1101: $$
1102: given by corollary {\bf \ref{cnm}}. Note that $Q_l(u)>0$ for $u \in [0,1]$.
1103: \pn
1104: \tiret
1105: If $t \in [-1,1]$, we have $u,v \in [0,1]$ and the announced result by
1106: corollary {\bf \ref{cnm}}.
1107: \pn
1108: \tiret
1109: We have $u([1,2])= u([-2,-1])=u([0,1])=[0,1]$. Let
1110: $\abs t \in [1,2]$. There exists $t_1 \in ]0,1]$, such that
1111: $u=u(t)=u(t_1)=u_1$ and we have $v=v(t)=t^2\geq t_1^2=v(t_1)=v_1$.
1112: We deduce
1113: \begin{eqnarray*}
1114: C_{k,l}(t) =
1115: v Q_l(u) - P_k(u) &=&
1116: v Q_l(u_1) - P_k(u_1)
1117: \\
1118: &\geq&
1119: v_1 Q_l(u_1) - P_k(u_1)>0.
1120: \end{eqnarray*}
1121: In conclusion, for $t\in [-2,2]$, we have $F_{k,l}(t)>0$.
1122: \pn
1123: \tiret $Q_l(u) \in (T_2+2)\RR_l[T_6+2]$ and $P_k \in
1124: \RR_k[T_6+2]$. We thus deduce that $C_{k,l} \in \RR[T_6] \oplus T_2 \RR[T_6]$.
1125: Note that $\deg C_{k,l} = \max(6k+2,6l)$.
1126: \pn
1127: \tiret
1128: If $n=2k+1$, let $C_n = t\cdot C_{k,k}$.
1129: If $n=2k$, let $C_n = t\cdot C_{k,k-1}$.
1130: $C_n$ has degree $2n+2\pent n2+1$ and therefore $C_n \in
1131: \vect(W_0,\ldots,W_n)$.
1132: \qed%
1133: \end{proof}
1134: \begin{remark}
1135: We have proved the existence of $C_n$. This is an upper-triangular basis of $E$
1136: with respect to the $W_i$. It is unique and it can be computed by
1137: simple LU-decomposition of the matrix whose lines are the $W_i$.
1138: \end{remark}
1139: \section{Conclusion}
1140: We have shown in this paper the existence of plane polynomial curves
1141: of degree $(3,N+ 2\pent N4+1)$ having the required properties. We
1142: think that they are of minimal lexicographic degrees (it is true for
1143: $N=3,5,7,9$). This question is related to the following
1144: question: where are the real zeros of polynomials in
1145: $\vect(V_k, k \not = 2 \mod 3)$?
1146: We guess that such polynomials cannot have too many zeroes in
1147: $[-1,1]$. It would give a lower bound for the degrees
1148: of the torus knots approximation by polynomial curves.
1149: \pn We have not given explicit formulas for our polynomials.
1150: We have just shown that they can be found by solving some
1151: explicit linear system. In a near future, we hope we will be able to
1152: give explicit function of the degree $N$.
1153: \begin{thebibliography}{ZZ99}
1154: \bibitem[Ad]{Ad}
1155: Adams, C. C.,
1156: {\it The knot book : an elementary introduction to the mathematical theory
1157: of knots},
1158: Providence, R.I. , American Mathematical Society, 2004.
1159: \bibitem[BG]{BG}
1160: Baker, G., Graves-Morris, P., {\it Pad{\'e} Approximants}, second edition,
1161: Cambridge University Press, 1996.
1162: %\bibitem[Fi]{Fi}
1163: %G. Fischer, {\it Plane Algebraic Curves}, A.M.S. Student Mathematical
1164: %Library Vol 15, 2001.
1165: \bibitem[Ha]{Ha}
1166: Hausdorff, F. {\it Summationsmethoden und Momentfolgen. I},
1167: Math. Z. 9, (1921), 74-109.
1168: \bibitem[Ka]{Ka}
1169: L. Kaufmann, {\it On knots,} Princeton University Press, 1987.
1170: \bibitem[KP]{KP}
1171: P. -V. Koseleff, D. Pecker, {\it On polynomial Torus Knots,}
1172: Journal of Knot Theory and its Ramifications, to appear. \goodbreak
1173: {\tt http://fr.arxiv.org/abs/math/0610663}
1174: \bibitem[Mu]{Mu}
1175: K. Murasugi,
1176: {\it Knot theory and its applications}, Boston, Birkh{\"a}user, 341p., 1996.
1177: \bibitem[RS]{RS}
1178: A. Ranjan and R. Shukla, {\it On polynomial Representation of Torus Knots,}
1179: Journal of knot theory and its ramifications, Vol. 5 (2) (1996) 279-294.
1180: \bibitem[Sc]{Sc}
1181: I. J. Schoenberg, {\it On finite and infinite completely monotonic
1182: sequences}, Bull. Amer. Math. Soc., Vol. 38, Number 2 (1932), 72-76.
1183: \bibitem[Sh]{Sh}
1184: A.R. Shastri, {\it Polynomial Representation of knots, }
1185: T{\^o}hoku Math. J. {\bf 44 } (1992), 11-17.
1186: \bibitem[SZ]{SZ}
1187: Salvy, B., Zimmermann, P. {\it GFUN: A Maple Package for the
1188: Manipulation of Generating and Holonomic Functions in One
1189: Variable}. ACM Transactions on Mathematical Software. (1994):
1190: 163-177. Vol. 20. No. 2.
1191: \bibitem[Va]{Va}
1192: V. A. Vassiliev, {\it Cohomology of Knot Spaces,} Theory of singularities
1193: and its Applications, Advances Soviet Maths Vol 1, (1990)
1194: \bibitem[Wa]{Wa}
1195: Wall, H. S., {\it Analytic theory of Continued Fractions}, Chelsea, 1973.
1196: %\bibitem[Wy]{Wy}
1197: %P. Wynn, {\it On the convergence and Stability of the Epsilon Algorithm,}
1198: %SIAM Journal on Numerical Analysis, Vol. 3, No. 1, (1966), pp. 91-122
1199: \end{thebibliography}
1200:
1201: \end{document}
1202:
1203: We have also shown that
1204: if $s \not = t$ are real numbers such that $T_3(s)= T_3(t),$ then
1205: there exists $\alpha \in \RR$, such that
1206: $$
1207: s = 2 \cos(\alpha + \pi/3), \, t = 2 \cos(\alpha - \pi/3), \, s+t =
1208: 2\cos \alpha.
1209: $$
1210:
1211:
1212:
1213: \begin{eqnarray}
1214: s_i = - 2 \cos(\alpha_{k+1-i} + 2 \pi/3), \
1215: t_i = - 2 \cos(\alpha_{k+1-i} - 2 \pi/3),&&\ i = 1,\ldots,k \\
1216: s_0 = - \sqrt 3, \ t_0 = \sqrt 3 &&\\
1217: s_i = 2 \cos(\alpha_{i-k} + \pi/3), \
1218: t_i = 2 \cos(\alpha_{i-k} - \pi/3),&&\ i = k+2,\ldots,n
1219: \end{eqnarray}
1220:
1221: We look for $z(t)$ as a linear combination of even degree polynomials
1222: $V_{6k}$ and $V_{6k+4}$ and we ask for
1223: $$\Frac{z(t_i)-z(s_i)}{t_i-s_i} = (-1)^i, \ i = 1, \ldots , N.$$
1224:
1225:
1226: We have $\deg W_{2k} = 6 k$ and $\deg W_{2k+1} = 6 k +2$, that
1227: is $\deg W_n = 2n + 2 \pent n2$.
1228:
1229: Furthermore we have
1230: $$
1231: P_{n,m}(v) = \Frac{v Q_m(u)-P_n(u)}{v^{n+m+1}}
1232: \mathop{\rightarrow}_{v \to 0}
1233: %\psi_{n,m}(0) Q_m(0) \lim_{v \to 0} \Frac{u^{n+m+1}}{v^{n+m+1}} =
1234: \left (\Frac{9}{4}\right )^{n+m+1} \psi_{n,m}(0)>0,
1235: $$
1236: so $P_{n,m}(v)>0$ when $v \in [0,1]$ because $u= v(v-3)^2/4 \in [0,1]$.
1237: We just consider now
1238: $$
1239: A_n:= -P_{n,m}(0)^{-1} P_n, \ B_m := P_{n,m}(0)^{-1} Q_m
1240: $$
1241: and we get the announced result:
1242: $$
1243: A_n(u) + v B_m(u) = v^{n+m+1} F_{n,m}(v).
1244: $$
1245: %\noindent%
1246: %\tiret Suppose now that
1247: %$$
1248: %A_n(u)+v B_m(u) = v^{n+m+1} F_{n,m}(v) >0, \ F_{n,m}(0)=1.
1249: %$$
1250: %Because $u \equi_{v\to 0} \Frac 94 v$, we deduce that
1251: %$$
1252: %A_n(u)+ v B_m(u) = 0 \mod{u^{n+m+1}}.
1253: %$$
1254: %Using the Eq (\ref{pade}), we deduce that $B_m(0)\not =0$ unless
1255: %$A_n=B_m=0$.
1256: %$-\Frac{A_n}{B_m}$ is the Pad{\'e} approximant $\phi^{[n/m]}$ and is
1257: %uniquely determinate. We also deduce that
1258: %$\deg A_n = n$ and $\deg Q_m=m$.
1259:
1260:
1261: