Page 339 - A Course in Linear Algebra with Applications
P. 339

9.2:  Quadratic  Forms                  323


        Corollary    9.2.3
                    2              2
        Let  q  =  ax  +  2bxy  +  cy  be  a  quadratic  form  in  x  and  y.
        Then:
             (a)  q  is  positive  definite  if  and  only  if  ac  >  b 2  and  a  >  0;
             (b)  q  is  negative  definite  if  and  only  if ac  >  b 2  and  a  <  0;
                                                     2
             (c)  q  is  indefinite  if  and  only  if  ac  <  b .
        Example     9.2.3
                                  2
        Let  q  =  -2x 2  +  xy  -  3y .  Here  we  have  a  =  - 2 ,  b  =  1/2,
        c  =  —3.  Since  ac  —  b 2  >  0  and  a  <  0,  the  quadratic  form  is
        negative  definite,  by  9.2.3.
             The  status  of  a quadratic  form  in three  or  more  variables
        can  be  determined  by  using  9.2.2.

        Example     9.2.4
        Let  q =  —2x 2  —  y 2  —  2z 2  + 6xz  be  a  quadratic  form  in  x,  y,  z.
        The  matrix  of the  form  is


                                  1-2      0    3
                            A=       0   - 1    0
                                 \   3      0 - 2


        which  has  eigenvalues  —5, —1,1.  Hence  q is  indefinite.
             Next  we  record  a  very  different  criterion  for  a  matrix  to
        be  positive  definite.  While  it  is  not  a  practical  test,  it  has  a
        very  striking  form.
        Theorem     9.2.4
        Let  A  be a real symmetric  matrix.  Then  A  is  positive  definite
                             T
        if  and  only  if  A  — B B  for  some  invertible  real matrix  B.
        Proof
                                    T
        Suppose   first  that  A  =  B B  with  B  an  invertible  matrix.
                                         T
        Then   the  quadratic  form  q =  X AX  can  be  rewritten  as

                                                            2
                                           T
                         T T
                    q =  X B BX     =  (BX) BX     =  \\BX\\ .
   334   335   336   337   338   339   340   341   342   343   344