2 A matricial approach
To study geometric objects in Euclidean spaces, we shall often use positive
definite and positive semidefinite matrices, or the corresponding quadratic
forms. Their detailed properties will be given in the Appendix. The following
basic theorem will enable us to mutually intertwine the geometric objects and
matrices.
Theorem 1.1.1Letp1,...,pnbe an ordered system of vectors in some
Euclideanr-dimensional (but not(r−1)-dimensional) vector space. Then
the Gram matrixG(p1,...,pn)=[gik],wheregik=Σpi,pkffis positive
semidefinite of rankr.
Conversely, letA=[aik]be a positive semidefiniten×nmatrix of rankr.
Then there exists in anym-dimensional Euclidean vector space, form≥r,n
vectorsp1,...,pnsuch that
Σpi,pkff=aik,for alli, k=1,...,n.
In addition, every linear dependence relation among the vectorsp1,
p2,...,pnimpliesthe samelinear dependence relation among the rows (and
thus also columns) of the Gram matrixG(p1,...,pn), and vice-versa.
The proof is in the Appendix, TheoremsA.1.44andA.1.45.
Another important theorem concerns so-called biorthogonal bases in the
Euclidean vector spaceEn. The proof will also be given in the Appendix,
TheoremA.1.47.
Theorem 1.1.2Letp1,...,pnbe an ordered system of linearly independent
vectors inEn. Then there exists a unique system of vectorsq1,...,qninEn
such that for alli, k=1,...,n
Σpi,qkff=δik
(δikis the Kronecker delta, meaning zero ifiΓ=kand one ifi=k). The
vectorsq1,...,qnare again linearly independent and the Gram matrices of
the systemsp1,...,pnandq1,...,qnare inverse to each other.
In other words, ifG(p),G(q)are the Gram matrices of the vectorspi,qj,
then the matrix
ff
G(p)I
IG (q)
Γ
has rankn.
Remark 1.1.3The basesp1,...,pnandq1,...,qnare calledbiorthogonal
basesinEn.
We shall be using, at least in the first chapter, the usual orthonormal
coordinate system inEn, which assigns to every point ann-tuple (usually
real, but in some cases even complex) of coordinates. We also recall that the