Page 296 - 35Linear Algebra
P. 296

296                                                                Kernel, Range, Nullity, Rank


                            Theorem 16.2.5 (Dimension Formula). Let L: V → W be a linear trans-
                                                                                1
                            formation, with V a finite-dimensional vector space . Then:
                                                  dim V   = dim ker V + dim L(V )
                                                          = null L + rank L.

                            Proof. Pick a basis for V :

                                                       {v 1 , . . . , v p , u 1 , . . . , u q },

                            where v 1 , . . . , v p is also a basis for ker L. This can always be done, for exam-
                            ple, by finding a basis for the kernel of L and then extending to a basis for V .
                            Then p = null L and p + q = dim V . Then we need to show that q = rank L.
                            To accomplish this, we show that {L(u 1 ), . . . , L(u q )} is a basis for L(V ).
                               To see that {L(u 1 ), . . . , L(u q )} spans L(V ), consider any vector w in L(V ).
                                                             j
                                                          i
                            Then we can find constants c , d such that:
                                                                 p
                                                                        1
                                                                                     q
                                                    1
                                         w = L(c v 1 + · · · + c v p + d u 1 + · · · + d u q )
                                                 1
                                                                            1
                                                                                            q
                                                                 p
                                             = c L(v 1 ) + · · · + c L(v p ) + d L(u 1 ) + · · · + d L(u q )
                                                                  q
                                                  1
                                             = d L(u 1 ) + · · · + d L(u q ) since L(v i ) = 0,
                                  ⇒ L(V ) = span{L(u 1 ), . . . , L(u q )}.
                               Now we show that {L(u 1 ), . . . , L(u q )} is linearly independent. We argue
                                                                              j
                            by contradiction. Suppose there exist constants d (not all zero) such that
                                                           1
                                                                            q
                                                   0 = d L(u 1 ) + · · · + d L(u q )
                                                              1
                                                                           q
                                                      = L(d u 1 + · · · + d u q ).
                                                                                            q
                                                                               1
                                            j
                            But since the u are linearly independent, then d u 1 + · · · + d u q 6= 0, and
                                             q
                                1
                                                                                   1
                                                                                                q
                            so d u 1 + · · · + d u q is in the kernel of L. But then d u 1 + · · · + d u q must
                            be in the span of {v 1 , . . . , v p }, since this was a basis for the kernel. This
                            contradicts the assumption that {v 1 , . . . , v p , u 1 , . . . , u q } was a basis for V , so
                            we are done.
                              1
                               The formula still makes sense for infinite dimensional vector spaces, such as the space
                            of all polynomials, but the notion of a basis for an infinite dimensional space is more
                            sticky than in the finite-dimensional case. Furthermore, the dimension formula for infinite
                            dimensional vector spaces isn’t useful for computing the rank of a linear transformation,
                            since an equation like ∞ = ∞ + x cannot be solved for x. As such, the proof presented
                            assumes a finite basis for V .
                                                      296
   291   292   293   294   295   296   297   298   299   300   301