|
hbghlyj
Posted at 2022-8-12 01:28:45
Last edited by hbghlyj at 2022-8-13 06:46:00青青子衿 发表于 2014-6-8 12:39
回复 11# isee
(美)戴维·盖尔(David Gale)编著;朱惠霖译.蚁迹寻踪及其他数学探索:上海教育出版社,2001年:第69页
帕施公理与西尔维斯特问题
变分方法的一个特征是,它经常导出非常简短的证明,这方面的一个惊人例子是西尔维斯特(J. J. Sylvester)于1893年提出的著名问题:设$S$是平面上的一个有限点集,且任何经过其中两个点的直线都一定经过其中另一个点,证明这些点都在一条直线上。不论是西尔维斯特还是他的同时代人都没能找到一个证明,过了将近50年,才由加莱(Gallai)发表了第一个证明,但相当复杂。下面这个简短的证明现已广为人知,它是由凯利(L. M. Kelly)于1948年发现的(见《美国数学月刊》(Amer. Math. Monthly)55,p.26-28)。假设这些具有西尔维斯特所述性质的点不共线,每条经过其中两点的直线$L$和不在这条直线上的一个点$p$都组成一个线点对$(L,p)$,在所有这些线点对中选取一个使得从$p$到$L$的距离$d$为最小的。令$q$为从$p$向$L$所引垂线的垂足。于是(变分)根据假设,在$L$上至少存在三个点$a,b$和$c$。因此其中两个点,比方说$a$和$b$,将以$a,b,q$的顺序位于$q$的同一侧($c$可在任何一侧),如图1,但这样从$b$到直线$ap$的距离$d'$就小于$d$,这就产生了一个矛盾。
图1:凯利的证明 凯利的证明确实简短——但是,这里有考克斯特(H. S. M. Coxeter)的说法(见《几何学导引》(Introduction to Geometry),Wiley,1961,p.181):“这件关于共线性的事[西尔维斯特的问题]显然属于序几何。[确实,在复数域或有限域上这个结果不成立!你可以轻易地在环面上发现一个九点的反例。]凯利的欧氏几何式证明涉及外在的距离概念:这好比用一把长柄锤子去砸一个杏仁。真正恰当的坚果钳由下述证明所提供。”
考克斯特的可爱的证明(很高兴这个证明用的也是变分方法)依赖于帕施(Pasch)公理。这条公理以它最简单的形式断言:一条直线不可能只与一个二角形的一条边相遇。(理解这条公理的一个方法是,把它看作若尔当(Jordan)曲线定理的一个非常初等的特例。如果这条直线通过一条边进入这个三角形,那么它必定要穿过另一条边以回到外面来。)这个证明的图与凯利证明的图非常相似,不过这次我们选取的是任意的点$p$,并找出一条从它出发的射线$R$,$R$上没有$S$中其他点,但至少与一条连接$S$中点的直线相交,每条这样的直线都与$R$相交于某点,于是我们可从中选取一条直线$L$,它与$R$的交点$q$距$p$最近(当然,不是在距离的意义下,而是把它们看作$R$上的一个序集。也就是说,在$p$和$q$之间没有其他交点),现在(变分)$L$上一定有两个点$a$和$b$,它们位于$q$的同一侧,我们证明直线$ap$上不可能有$S$中的另一个点,有两种情况。
图2:运用帕施公理的证明 情况1 另一个点$y$位于$a$和$p$之间。于是(如图2)不论$c$在哪里,将帕施公理应用于三角形$apq$,直线$cy$将与$R$相交于一个比$q$更近于$p$的点。
情况2 另一个点$x$或$z$不在$a$和$p$之间,于是,同前面一样,不是$bx$就是$bz$,将与$R$相交于一个比$q$更近于$p$的点,(见图3)。
图3:运用帕施公理对另一种情况的证明
A Problem of Collinear Points
Author(s): H. S. M. Coxeter
Source: The American Mathematical Monthly , Jan., 1948, Vol. 55, No. 1 (Jan., 1948), pp. 26-28
2305324 text_filtered.pdf
(422.7 KB, Downloads: 29)
按照这帖方法处理PDF(删除下载的IP):
- D:\gs9.54.0\bin\gswin32.exe -o "D:\2305324 text_filtered.pdf" -sDEVICE=pdfwrite -dFirstPage=2 -dLastPage=4 -dFILTERTEXT "D:\2305324.pdf"
Copy the CodeSylvester himself was doubtless aware of the negative answer in the case of the complex projective plane, where the nine points given by
$$
x^{3}+y^{3}+z^{3}-x y z=0
$$
(which are the points of inflexion of any one of the cubic curves $x^{3}+y^{3}+z^{3}$ $=c x y z)$ lie by threes on twelve lines in such a way that every two of the points belong to such a set of three. Some years later, Veblen and Bussey described the finite affine geometries, one of which, called $E G(2,3)$, consists of the nine points $(x, y)$ whose coördinates are residues modulo 3 , and the twelve lines
$$
X x+Y y+Z \equiv 0\pmod 3
$$
which join them. Since any two points determine a line, and every line contains three points, all the points in this geometry form a set of the desired kind (with $n=9)$.
But the question as applied to the ordinary real plane remained unanswered for forty years, and began to seem as intractable as the four-color map problem. Finally, Grünwald established the affirmative answer: such points must all be collinear. Robert Steinberg transformed Grünwald's affine proof into a projective one (this Monthly, vol. 51 (1944), p. 169), and L. M. Kelly discovered a still shorter (and quite unrelated) Euclidean proof (see below).
It seems to me that parallelism and distance are essentially foreign to this problem, which is concerned only with incidence and order. Thus I would not regard the proofs by Grünwald and Kelly as strictly synthetic. Steinberg's projective proof becomes even more elementary when we transform it into a "descriptive" proof, using the single primitive entity point and the single primitive relation between, in terms of which lines and serial order can be defined (see Veblen, Transactions of the American Mathematical Society, vol. 5 (1904), pp. 353-371).
If the $n$ points are not all collinear, some three of them must form a triangle $A B C$. Lines joining $A$ to all other points of the set meet the line $B C$ in $B$ and $C$ and possibly other points. Let $P^{\prime}$ be a point of $B C$ not among these. Then the line $A P^{\prime}$ contains no point of the set except $A$. Joins of pairs of points belonging to the set meet $A P^{\prime}$ in $A, P^{\prime}$, and possibly other points. Let $P$ be the first one of these, going from $A$ towards $P^{\prime}$. ($P$ may possibly coincide with $P^{\prime}$.) Then no join of two points of the set can pass through a point between $A$ and $P$. We shall obtain a contradiction by showing that this "empty" segment $A P$ is not really empty after all.
By its definition, the point $P$ lies on a line containing at least three points of the set, say $Q, R, S$, named so that the segment $P R$ contains $Q$ but not $S$. (We do not care whether $S$ lies beyond $R$ or beyond $P$.) Since $A$ and $R$ belong to the set, the line $A R$ contains a third point of the set, say $O$. Two cases now arise. If $O$ lies between $A$ and $R$, the line $S O$ must intersect the segment $A P$ (by Pasch's Theorem). Similarly, if $O$ lies outside the segment $A R$, the line $Q O$ must intersect the segment $A P$. In either case we have the desired contradiction. Hence the triangle $A B C$ cannot exist, and the $n$ points must in fact be collinear.
Alternative proof of collinearity, by L. M. Kelly. Assume the contrary. If the points be labeled $p_{1}, p_{2}, \cdots, p_{n}$, then there exists a point $p_{i}$ and a line $p_{j} p_{k}$ such that the perpendicular distance from $p_{i}$ to the line $p_{j} p_{k}$ is the shortest non-zero perpendicular distance from any of the points to any of the lines defined by pairs of the points. Let $q$ be the foot of the perpendicular from $p_{i}$ to the line $p_{j} p_{k}$. On the line $p_{j} p_{k}$ there is at least one more point of the set, $p_{l}$. At least two of $p_{j}, p_{k}, p_{l}$ must lie on one side of $q$. Suppose for definiteness that $p_{j} p_{k} q$ is the order $\left(p_{k}\right.$ may coincide with $\left.q\right)$. Then the distance from $p_{k}$ to the line $p_{j} p_{i}$ provides us with a contradiction since it is shorter than the assumed minimum $p_{i} q$. 又见cut-the-knot
Steinberg's Solution
Given the set $Π$ of noncollinear points, consider the set of lines $Σ$ that pass through at least two points of $Π$. Such lines are said to be connecting. Among the connecting lines, those that pass through exactly two points of $Π$ are called ordinary. We consider the configuration in the projective plane.
Let $p$ be any point of $Π$. If $p$ lies on an ordinary line we are done, so we may assume that $p$ lies on no ordinary line. Let $t$ be a line (in the plane) through $p$ but not through any other point of $Π$. The lines in $Σ$ not through $p$ meet $t$, in points $x_1,x_2,…,x_k$ say, named in cyclic order so that one of the two segments determined by $p$ and $x_1$ contains none of the points $x_2,…,x_k$ within it:Let $s$ be a line of $Σ$ through $x_1$. Then $s$ must be ordinary!
(还没抄完...待续) |
|