1: \documentstyle[twoside,12pt]{article}
2: %\documentstyle[twoside,11pt]{article}
3: %\documentstyle[twoside,10pt]{article}
4:
5: %\begin{document}
6:
7: \input{mac}
8:
9: %special newcommands for this article
10: \newcommand{\Middle}{{\mbox{Middle}\,}}
11: \newcommand{\End}{{\mbox{End}\,}}
12: \newcommand{\GL}{{\mbox{GL}\,}}
13: \newcommand{\diam}{{\mbox{diam}\,}}
14: \newcommand{\tr}{{\mbox{tr}\,}}
15:
16: %Topmatter
17: \title{How to Axiomatize School Geometry}
18: %Author info
19: \author{Eliahu Levy\\
20: %\address{
21: Department of Mathematics\\
22: Technion -- Israel Institute of Technology,
23: Haifa 32000, Israel\\
24: email: eliahu@techunix.technion.ac.il}
25:
26: %\keywords{}
27: %\subjclass{}
28:
29: \date{}
30:
31: %End topmatter
32:
33: \maketitle
34: %\begin{abstract}
35: %\end{abstract}
36:
37: %\tableofcontents
38:
39: %\setcounter{section}{-1}
40:
41: This is an attempt to present axioms for Euclidean Geometry, aiming at the
42: following goals:
43: \begin{itemize}
44: \item To work with ``geometrical'' notions. Thus we would not merely
45: identify points in the plane with pairs of real numbers, which means that a particular
46: coordinate system is given special status.
47: \item To be appropriate to the way geometry is done in higher mathematics
48: (including physics and engineering). This means that the algebraic nature
49: of much of geometry need not be hidden.
50: \item To respond to the desire that one would confidently accept
51: empirically/intuitively
52: that the axioms are valid in our physical everyday world (or rather in the
53: usual idealization that ``geometry'' is). This seems to disfavor
54: taking the Theorem of Pythagoras as an axiom.
55: \item To have accessible the usual rigor of ``pure'' mathematics, and to
56: make the axioms satisfying by the standards of the latter. In particular,
57: not to take as an axiom something that can be naturally proved. Note that some
58: ``topological'' notions, necessary for the rigor of the presented axioms, can
59: be readily kept silent with an ``unsophisticated'' audience
60: (such as school).
61: \end{itemize}
62: \medskip
63:
64: The style in the sequel is intended for those accustomed to mathematical
65: writings, in order to make the mathematical contents clear. Of course, in case
66: an approach in this spirit can be practiced in school the style of
67: presentation must be quite different.
68:
69: %\newpage
70:
71: \section{The Axioms: Plane Geometry}
72:
73: Primitive notions:
74:
75: \begin{itemize}
76: \item A set $P$ (The {\EM Plane}), whose elements are called {\EM points}.
77: For the sophisticated -- the Plane is assumed a Hausdorff
78: topological space. We shall let $x,y,z,\;a,b,c$ etc.\ vary over
79: points.
80: \item A relation among 3 points, indeed a commutative ``algebraic'' operation:
81: $z$ is the middle between $x$ and $y$ (to be written $z=\Middle(x,y)$).
82: \item An equivalence relation on $P\times P$: two pairs of points have the same
83: distance.
84: \end{itemize}
85: \bigskip
86:
87: We shall have altogether three axioms.
88: \medskip
89:
90: By $\bR$ we denote, as usual, the set of {\EM real numbers}
91: (for the unsophisticated - just the set of {\EM numbers}, representable, say,
92: as possibly unending decimal fractions).
93:
94: {\Ax \label{Ax:Co}
95: {\EM (Axiom of Coordinates)}. There is a bijection
96: (coordinates system) between $\bR^2$ and the Plane (which is a homeomorphism)
97: such that any mapping given in the coordinates
98: by $x\mapsto a+x$ or $x\mapsto a-x$ (here $a\in\bR^2\,,x\in\bR^2$) ``preserves
99: the geometry'': it is an ``isomorphism'' with respect to the middle operation
100: and maps any pair of points to a pair with the same distance.
101: }
102: \bigskip
103:
104: Note that we did not take the coordinates as a primitive notion: there may
105: be many such bijections, the axiom saying that there is at least one.
106: \medskip
107:
108: The empirical/intuitive evidence for this axiom is plain: one encounters such
109: coordinates daily (with the ``rough'' everyday correlate of the idealized
110: ``set of points''). Maybe it is more intuitive to postulate the stronger
111: requirements that the reflections $(x_1,x_2)\mapsto(\alpha\pm x_1,x_2)$ and
112: $(x_1,x_2)\mapsto(x_1,\alpha\pm x_2)$\quad($\alpha\in\bR$) preserve the
113: geometry, the existence of enough reflections being implicit even in Euclid.
114: \medskip
115:
116: As a simple consequence we may prove
117:
118: {\Thm \label{T} In any coordinates system satisfying Axiom \ref{Ax:Co}
119: the middle operation $\Middle(a,b)$ corresponds to the ``algebraic middle''
120: $\dfrac{a+b}2\quad a,b\in\bR^2$.}
121:
122: \begin{Prf}
123: The map $x\mapsto(a+b)-x$ maps $a\to b$, $b\to a$, and preserves the
124: geometry. Hence it fixes $\Middle(a,b)$. But its only fixed point is
125: $\dfrac{a+b}2$.
126: \end{Prf}
127: \medskip
128:
129: Define, for integer $n\ge2$, an {\EM $n$-ruler} as a
130: sequence of points $(a_i)_{0\le i\le n}$ such that for any $0<i<n$\quad
131: $a_i=\Middle(a_{i-1},a_{i+1})$
132: \medskip
133: \NOT{
134: In fact, we need only $2$-rulers and the following Fact might be stated
135: only for $n=2$.
136: Introducing $n$-rulers for general $n$ serves only for didactic purposes.
137: \medskip}
138:
139: Theorem \ref{T} implies that for any coordinates system as in Axiom
140: \ref{Ax:Co}, a ruler is just an ``algebraic ruler'', i.e.\ a sequence
141: with constant difference. We may deduce:
142: \medskip
143:
144: {\Fact For any integers $n\ge k>l\ge 0$ and points $a,b$ there is
145: a unique $n$-ruler $(c_i)_{0\le i\le n}$ with $a=c_k$ and $b=c_l$.
146: }
147: \medskip
148:
149: If we define, with respect to some coordinates system, an
150: ``algebraic {\EM straight line}'' as usual (as a set $L\subset\bR^2$ of the form
151: $L=\{a+\lambda c\,|\,\lambda\in\bR\}$ where $a,c\in\bR^2,\,c\ne0$) then the
152: straight line joining $a,b\in P$ contains the ``rational line'' joining
153: $a$ and $b$, i.e.\ the set of all points obtained by constructing $n$-rulers
154: according to the Fact, and is its closure. Thus the notion of straight line
155: is independent of the coordinates (note that we needed the topology
156: here, and that just $2$-rulers would have sufficed).
157: \medskip
158:
159: A quadrangle $(a,b,c,d)$ is an ``algebraic parallelogram'' with respect to
160: some coordinates system if $a-b=d-c$. But this is equivalent to $(a,c)$
161: and $(b,d)$ having the same ``algebraic middle'', i.e.\ to
162: $\Middle(a,c)=\Middle(b,d)$. Thus the notion of parallelogram is again
163: ``geometrical'' -- independent of the coordinates system.
164: This allows us to define the {\EM vectors} geometrically as ``differences of
165: pairs of points'', that is, say, as equivalence classes of pairs of points
166: by the equivalence relation defined by parallelograms. (Thus a point minus
167: a point is a vector, and a point plus a vector is a point).
168: Any coordinates system lets us identify the vectors with $\bR^2$, thus making them into a
169: 2-dimensional $\bR$-vector space, and one easily shows that the vector
170: operations can be defined ``geometrically'' -- independent of the coordinates.
171: (For multiplication by general real numbers we again need the topology).
172: Denote the 2-dimensional space of vectors by $V$. By $\End(V)$ we will mean
173: the space of linear self-maps of $V$.
174: \bigskip
175:
176: In so far we had little to do with the primitive equivalence relation of two
177: pairs of points having the same distance. Now we come to it. By the
178: requirements from coordinates in Axiom \ref{Ax:Co} any two pairs with the
179: same vector difference have the same distance, thus we get an equivalence
180: relation between vectors: {\EM having the same length},
181: and moreover $v$ and $-v$ always have the same length.
182: \medskip
183:
184: Define an {\EM isometry} as an invertible linear self-map $\cU\in\End(V)$
185: mapping each vector into a vector with same length. The set of isometries
186: is a group. By the above, $-1$ belongs to this group. (Here and in the sequel
187: we identify a scalar operator with the scalar).
188: \medskip
189:
190: The two remaining axioms deal with isometries. They have a markedly algebraic
191: flavor, which seems justifiable in view of the above.
192: \medskip
193:
194: {\Ax \label{Ax:Ist}
195: {\EM (Axiom of Isotropy)}. The group of isometries is transitive on a set of
196: all vectors of the same length, and is also transitive on the set of
197: $1$-dimensional subspaces of $V$. (That is: for any two vectors of the same
198: length, or two $1$-dimensional subspaces $\exists$ an isometry mapping one
199: to the other).
200: }
201: \medskip
202:
203: Instead of the first half of Axiom \ref{Ax:Ist}, one could take the group
204: of isometries as a primitive notion\NOT{(postulating that they are
205: (homeomorphisms) that preserves the Middle),} and define vectors to have the
206: same length iff an isometry maps one to the other.
207: \medskip
208:
209: {\Ax \label{Ax:Bd}
210: {\EM (Axiom of Boundedness)}. The group of isometries is bounded (as a subset
211: of the $4$-dimensional $\bR$-vector space $\End(V)$).
212: }
213: \bigskip
214:
215: Axiom \ref{Ax:Ist} is related to the empirical/intuitive possibility of motions
216: (rotations etc.), which is often expressed by congruence axioms.
217: Axiom \ref{Ax:Bd} postulates that the circle is bounded in a coordinate
218: system, in spite of the latter extending to infinity in the idealization
219: which is ``geometry''.
220: \medskip
221:
222: Now we shall be able to use the following theorem from algebra/analysis
223: to obtain that there is a positive-definite quadratic form $Q$ on $V$ such
224: that vectors $u,v\in V$ have the same length iff $Q(u)=Q(v)$
225: (thus we have the Theorem of Pythagoras). The resort to such theorem
226: here seems natural from our point of view. Unfortunately, proving it
227: requires some mathematical sophistication.
228:
229: {\Thm \label {Th:Q}
230: For any bounded group $G\subset\GL(V)$, where $V$ is a 2-dimensional
231: $\bR$-vector space, there exists a $G$-invariant positive-definite quadratic
232: form $Q$.}
233: \bigskip
234:
235: We give three proofs, differing in the tools used.
236:
237: \begin{Prff}{\EM Proof 1}
238: $\BAR{G}$ is a compact group, thus admits a normalized Haar measure $\mu$.
239: Take any positive-definite
240: quadratic form $Q_0$ and take as $Q$ the average
241: $$Q(v)=\int_{g\in\BAR{G}} Q_0(gv)\,d\mu(g).$$
242: \end{Prff}
243:
244: This proof works for any finite-dimensional $V$ over $\bR$.
245:
246: \begin{Prff}{\EM Proof 2}
247: This again works for any finite-dimensional $V$.
248: \medskip
249:
250: Let $W$ be the $\bR$-vector space of quadratic forms on $V$, and $W_+$ the
251: set of the positive-definite ones (this set is an open convex cone). $G$
252: acts on $W$ in the canonical way: $(gQ)(v):=Q(g^{-1}v),\; Q\in W$, and
253: leaves $W_+$ invariant.
254:
255: Choose a norm $\|\|_0$ on $W$, say the maximum of the absolute values of the
256: matrix entries with respect to a basis of $V$. Replace $\|\|_0$ by the
257: $G$-invariant norm
258: $$\|Q\|:=\sup_{g\in G}\|gQ\|_0.$$
259: We know that there is a fixed integer $N>0$ such that any subset of $W$ with
260: $\|\|$-diameter $\le d$ can be covered by at most $N$ sets of $\|\|$-diameter
261: $\le\frac d2$. if $K\subset W$ is bounded non-empty $G$-invariant convex,
262: say the convex hull of the orbit of some $Q$, and $\diam(K)\le d$,
263: then we have a finite set $F\subset K$, of at most $N$ elements, such that
264: $\forall Q\in K\,\exists Q'\in F\;\|Q-Q'\|\le\frac d2$. This holds, in
265: particular, for any $Q$ of the form $gQ_0,\;g\in G,\;Q_0:=\dfrac{\sum F}{\# F}$.
266: Thus $gQ_0$ has distance $\le\frac d2$ from some $Q'\in F$ and distance
267: $\le d$ from the other members of $F$. This implies $\|gQ_0-Q_0\|<\gamma d$
268: where $\gamma:=\dfrac{2N-1}{2N}<1$. Since $\|\|$ is $G$-invariant, we have
269: that the orbit of $Q_0$, hence its convex hull, has diameter $\le\gamma d$.
270:
271: So we know that any bounded non-empty $G$-invariant convex set $\subset W$
272: with diameter $\le d$ has a non-empty $G$-invariant convex subset of diameter
273: $\le\gamma d$. Repeating the process we get an infinite sequence of nested
274: sets which converges to a $G$-invariant $Q_I\in W$. If we ensure that for any
275: $Q\in K$ $Q(v)\ge\alpha\|v\|^2$ for some fixed norm $\|\|$ on $V$ and some
276: fixed $\alpha>0$, then $Q_I$ will be positive-definite.
277: \end{Prff}
278:
279: \begin{Prff}{\EM Proof 3}
280: This is a purely algebraic proof, using $\dim(V)=2$.
281:
282: There is only a 1-dimensional space of antisymmetric forms on $V$; that is,
283: a choice of such non-zero form, which we make and denote by
284: $u\wedge v\;\;u,v\in V$, is possible and unique up to a scalar multiple.
285: (This follows from $b_1\wedge b_1=0,\;b_2\wedge b_2=0,\;
286: b_2\wedge b_1=-b_1\wedge b_2$ which any such form must satisfy for a basis
287: $(b_1,b_2)$, while these formulas indeed give a non-zero antisymmetric form.)
288:
289: The determinant and trace of the matrix of an $\cA\in\End(V)$ are independent
290: of the basis (since different bases give similar matrices), hence we may
291: speak of $\det(\cA)$ and $\tr(\cA)$. The characteristic polynomial of $\cA$
292: is
293: $$x^2-\tr(\cA)\cdot x+\det(\cA),\leqno{(1)}$$
294: its real roots are the real eigenvalues of $\cA$, and plugging $\cA$ in it
295: gives $0$, by the Cayley-Hamilton Theorem.
296:
297: For a traceless $\cA\in\End(V)$ (i.e.\ with $\tr(\cA)=0$),
298: we obtain from the Cayley-Hamilton
299: Theorem that $\cA^2$ is a scalar, equal to $-\det(\cA)$, and of course to
300: $\frac12 \tr(\cA^2)$. This scalar gives a quadratic form on the
301: $3$-dimensional $\bR$-vector space $\{\cA\in\End(v)\,|\,\tr(\cA)=0\}$,
302: where the corresponding symmetric bilinear form is
303: $$\LA\cA,\cB\RA:=\frac12\tr(\cA\cB)=\frac12\tr(\cB\cA).$$
304: Checking the orthogonal basis
305: $\LP\begin{array}{cc}1&0\\0&-1\end{array}\RP,
306: \LP\begin{array}{cc}0&1\\1&0\end{array}\RP,
307: \LP\begin{array}{cc}0&1\\-1&0\end{array}\RP$
308: shows that the signature is $(+,+,-)$. Thus there cannot be two orthogonal
309: elements with non-positive value of the quadratic form.
310:
311: If $G\subset\GL(V)$ is a bounded group, then the image of $G$ by $\det$ is a
312: bounded subgroup of $\bR^\times(=\bR\setminus\{0\})$, therefore
313: $\det(\cA)=\pm1$ for $\cA\in G$. Also, if $\cA\in G$ has a real eigenvalue
314: $\lambda$, then $\cA^n\in G$ for integer $n$ and has the eigenvalue
315: $\lambda^n$, and these must be bounded, therefore $\lambda=\pm1$.
316:
317: Hence if $\cA\in G$ has determinant $1$, $(1)$ cannot have
318: a real root different from $\pm1$, which implies $|\tr(\cA)|\le2$. For such
319: $\cA$, $(1)$ can be written as (recall that $\det(\cA)=1$):
320: $$\LP x-\frac12\tr(\cA)\RP^2=-\LP1-\LP\frac12\tr(\cA)\RP^2\RP\le0.\leqno{(2)}$$
321: By Cayley-Hamilton, such $\cA$ can be written as $\frac12\tr(\cA)+\cJ$
322: where $\cJ$ is traceless with $\cJ^2=-\det(\cJ)$ non-positive, equal to
323: $-\LP1-\LP\frac12\tr(\cA)\RP^2\RP$.
324: Also, the product of $\frac12\tr(\cA)\pm\cJ$ is $1$, thus
325: $\cA^{-1}=\frac12\tr(\cA)-\cJ$.
326:
327: If we had $\cJ^2=0$ without $\cJ=0$, then $\frac12\tr(\cA)=\pm1$, thus
328: $\pm\cA=1+\cJ_1$, $\cJ_1^2=0$, $\cJ_1\ne0$ and $(\pm\cA)^n=1+n\cJ_1$
329: contradicting the boundedness of $G$. Hence if $\cA$ is not $\pm1$ then
330: $\cJ^2<0$ and $|\tr(\cA)|<2$.
331:
332: We claim that if $\cA_1=\tau+\cJ_1$ and $\cA_2=\tau+\cJ_2$, $\tau$ scalar,
333: are elements of $G$ with determinant $1$ and the same trace, $\cJ_1$ and
334: $\cJ_2$ being traceless with the non-positive square $-(1-\tau^2)$,
335: then $\cJ_2=\pm\cJ_1$, that is $\cA_2=\cA_1^{\pm1}$.
336: Indeed, in the above quadratic form on the space of traceless elements of
337: $\End(V)$, given by the scalar square, $\cJ_1$ and $\cJ_2$ have the same
338: non-positive square and thus are the sum and difference of the orthogonal
339: $\frac12(\cJ_1\pm\cJ_2)$. These cannot both have negative square, the
340: signature being $(+,+,-)$, and none can have square $0$ if $\cJ_1\ne\pm\cJ_2$.
341: Hence in the latter case they have squares of strictly different signs which
342: implies
343: $$|\LA\cJ_1,\cJ_2\RA|=|\frac12\tr(\cJ_1\cJ_2)|>|\LA\cJ_i,\cJ_i\RA|=1-\tau^2.$$
344: Returning to $\cA_1$ and $\cA_2$ this gives either $\cA_1\cA_2$ or
345: $\cA_1\cA_2^{-1}$ is an element of $G$ which has half-trace greater than $1$
346: which we saw above is impossible.
347:
348: Suppose that we have picked an $\cA\in G$ with determinant $1$ and $\cA$ not
349: the scalar $\pm1$. We have
350: $\cA=\frac12\tr(\cA)+\cJ$, $\cA^{-1}=\frac12\tr(\cA)-\cJ$, $\cJ$ is traceless
351: and $\cJ^2<0$. Consider the symmetric bilinear form on $V$
352: $$B(u,v):=\frac12\LQ(\cA u)\wedge v+(\cA v)\wedge u\RQ=(\cJ u)\wedge v.
353: \leqno{(3)}$$
354: If $b_1$ is a non-zero vector and $b_2=\cJ b_1$, then $b_2$ cannot be
355: $=\lambda b_1$, $\lambda\in\bR$ because that would imply
356: $\cJ^2 b_1=\cJ b_2=\lambda \cJ b_1=\lambda b_2=\lambda^2 b_1$.
357: Therefore $(b_1,b_2)$ is a basis, and $B(b_1,b_1)=b_2\wedge b_1\ne0$.
358: We have $B(b_1,b_2)=0$ and
359: $B(b_2,b_2)=(\cJ^2 b_1)\wedge(\cJ b_1)=-(\cJ^2)B(b_1,b_1)$.
360: So we conclude that $B$ or $-B$ must be positive-definite.
361:
362: We claim that $B$ is invariant under $G$. Indeed, for $\cB\in G$:
363: \begin{eqnarray*}
364: B(\cB u,\cB v)&=&\frac12\LQ(\cA\cB u)\wedge(\cB v)+(cA\cB v)\wedge(\cB u)\RQ=
365: \\&=&\det(\cB)\frac12\LQ(\cB^{-1}\cA\cB u)\wedge v+
366: (\cB^{-1}\cA\cB v)\wedge u\RQ
367: \end{eqnarray*}
368: But $\cB^{-1}\cA\cB$ is an element of $G$ with determinant $1$ and the same
369: trace as $\cA$. By the above, it is equal to either $\cA$ or $\cA^{-1}$ and
370: we find that $B(\cB u,\cB v)$ is one of $\pm B$. Since both are
371: positive-definite or negative-definite, they are equal.
372:
373: The theorem is hence proved except when all members of $G$ with determinant
374: $1$ are scalars. If that is the case, then if there are no elements in $G$
375: with determinant $-1$ we are done. In any case, for any $\cJ\in G$ with
376: $\cJ^2=-1$ we can, as above, construct a basis $(b_1,b_2),\;b_2=\cJ b_1$
377: and the matrix of $\cJ$ in this basis is
378: $\LP\begin{array}{cc}0&1\\-1&0\end{array}\RP$ with determinant $1$. Hence
379: such a $\cJ$ is excluded in our case, and all $\cA\in G$ with $\det(\cA)=-1$
380: must satisfy $\cA^2=1$. Then for any $v\in V$
381: $v=\frac12(v+\cA v)+\frac12(v-\cA v)$ is a sum of eigenvectors of $\cA$
382: with eigenvalues $1$ and $-1$, respectively. Since $\det(\cA)=-1$ we
383: have a $1$-dimensional space of each, i.e.\ $\cA$ has matrix
384: $\LP\begin{array}{cc}1&0\\0&-1\end{array}\RP$ in some basis. Any other member
385: of $G$ with determinant $-1$ is a multiple of this $\cA$ by a member of $G$
386: with determinant $1$, hence the only possible members of $G$ are
387: $\pm1$, $\pm\cA$ and one easily finds many positive-definite quadratic forms
388: invariant with respect to these.
389: \end{Prff}
390: \medskip
391:
392: One may remark, following Bourbaki, that after one has the quadratic form on
393: $V$ that determines equality of length, and thus a corresponding symmetric
394: bilinear form $\LA\,\RA$, one may define a
395: $\cJ\in\End V$ by $\LA u,v\RA=(\cJ u)\wedge v$
396: (see the third proof above) and prove that $\cJ^2$ is a negative scalar,
397: hence by normalizing the $\wedge$ one may have $\cJ^2=-1$ (which determines
398: $\cJ$ up to sign). This turns $V$ in a canonical way into a $1$-dimensional
399: complex vector space (didactically, the complex numbers may be defined by
400: our ``geometric'' $\cJ$ -- in this approach every plane (say, in $3$-space)
401: has, strictly speaking, its own ``complex numbers''.) Using this complex
402: structure to do plane geometry is very fruitful. For example, angles (with
403: their trigonometry) can be easily treated in a rigorous way.
404:
405: \section{Space Geometry}
406:
407: To axiomatize the Euclidean geometry of $n$-space, one may start from a set
408: {\EM Space} of points with exactly analogous primitive notions, replace
409: $\bR^2$ by $\bR^n$ in the Axiom of Coordinates, and postulate the Axioms of
410: Isotropy and of Boundedness for every sub-$2$-plane of Space
411: (or alternatively for Space itself). In passing from the existence of an
412: equality-of-length -- determining quadratic form on every plane to the
413: existence of one such form for Space, one may use the well-known
414:
415: {\Thm
416: Let $V$ be an (not necessarily finite-dimensional) $\bR$-vector space,
417: $\dim(V)\ge2$, and let $U\subset V$. If for every $2$-dimensional subspace
418: (=plane) $V'\subset V$ there exists a positive-definite quadratic form $Q'$ on $V'$
419: such that $U\cap V'=\{v\in V'\,|\,Q'(v)=1\}$ then there exists a
420: positive-definite quadratic form $Q$ on the whole $V$ such that
421: $U=\{v\in V\,|\,Q(v)=1\}$.
422: }
423: \begin{Prf}
424: It is clear that the $Q'$ are unique for each plane, and that they agree
425: on intersections. Hence they define a function $Q$ on $V$, positive on
426: $V\setminus{0}$ and it remains to prove that $Q$ is quadratic, i.e.\ comes
427: from a symmetric bilinear form on $V$. That form must be:
428: $$\LA u,v\RA=\frac12(Q(u+v)-Q(u)-Q(v))\quad u,v\in V$$
429: and we have to prove that this is bilinear. Since we have
430: $\LA\lambda u,v\RA=\lambda\LA u,v\RA$, $Q$ being quadratic on the plane
431: containing $u$ and $v$, it remains to prove biadditivity.
432: As we clearly have $\LA u,0\RA=0$,
433: biadditivity will follow if we prove that $\LA u,v\RA+\LA u,w\RA$ depends
434: only on $u$ and $v+w$. This obtains from the following calculation
435: (where one uses the parallelogram
436: equality $2Q(a)+2Q(b)=Q(a+b)+Q(a-b),\;\;a,b\in V$, which holds since $Q$
437: is quadratic in the plane containing $a$ and $b$):
438: \begin{eqnarray*}
439: &&2\LP\LA u,v\RA+\LA u,w\RA\RP=Q(u+v)-Q(u)-Q(v)+Q(u+w)-Q(u)-Q(w)=\\
440: &&=\frac12\LP Q(2u+v+w)+Q(v-w)\RP-2Q(u)-\frac12\LP Q(v+w)+Q(v-w)\RP=\\
441: &&=\frac12\LP Q(2u+v+w)-Q(v+w)\RP-2Q(u)
442: \end{eqnarray*}
443: \end{Prf}
444:
445: \end{document}
446: