Page 262 - DMTH403_ABSTRACT_ALGEBRA
P. 262

Unit 27: Separable Extensions




          For uniqueness, let D  and D  be derivations on L which extend the same derivation on K. Since  Notes
                                 2
                           1
          D (K)  K and D (K)  K, we have D (F)  F and D (F)  F by Lemma. Then D  = D  on F since
                                                                             2
                                        1
                                                                         1
                        2
            1
                                                   2
          F/K is separable, and D  = D  on L since L/F is separable.
                                 2
                             1
          When L/K is an algebraic extension of possibly infinite degree, here is the way separability is
          defined.
          Definition: An algebraic extension L/K is called separable if every finite subextension of L=K is
          separable. Equivalently, L=K is separable when every element of L is separable over K.
          This definition makes no sense if L/K is not an algebraic extension since a non-algebraic extension
          is not the union of its finite subextensions.
          Theorem 1 has a problem in the infinite-degree case: there is no natural trace map. However, the
          conditions in Theorems 2 and 3 both make sense for a general L/K. (In the case of Theorem 2,
          we have to drop the specification of  K  K  L  as a product of copies of  K , and just leave the
          statement  about the  tensor product  having no  non-zero nilpotent  elements.)  It is  left to  the
          reader to check for an infinite algebraic extension L/K that the conditions of Theorems 2 and 3
          match Definition.
          The conditions in Theorems 2 and 3 both make sense if L/K is not algebraic, so they could each
          potentially be used to define separability of a completely arbitrary field extension. But there is
          a problem: for transcendental (that is, non-algebraic) extensions the conditions in Theorems 2
          and 3 are  no longer equivalent.  Indeed, take  L  = K(u),  with  u transcendental over  K. Then
          K  K  L  =  K(u)  is a field, so the condition in Theorem 2 is satisfied. However, the zero derivation
          on K has more than one extension to K(u): the zero derivation on K(u) and differentiation with
          respect to u on K(u).
          Definition: A commutative ring with no nonzero nilpotent elements is called reduced.

          A domain is reduced, but a more worthwhile example is a product of domains, like F3 × Q[X],
          which is not a domain but is reduced.

          Definition: An arbitrary field extension L/K is called separable when the ring K  K  L  is reduced.

          Using this definition, in characteristic 0 all field extensions are separable. In characteristic p, any
          purely transcendental extension is separable. The condition in Theorem 3, that derivations on
          the base field admit unique extensions to a larger field, characterizes not separable field extensions
          in general, but separable algebraic field extensions.

          A condition equivalent to that in Definition is that  F  K  L  is reduced as F runs over the finite
          extensions of K.


          The condition that  K  K  L  is reduced makes sense not just for field extensions L/K, but for any
          commutative K-algebra. Define an arbitrary commutative K-algebra A to be separable when
          the ring  K  K  A  is reduced. This condition is equivalent to A   F being reduced for every finite
                                                            
          extension field F/K.


                Example: Let A = K[X]/(f(X)) for any non-constant f(X)  K[X]. The polynomial f(X) need
          not be irreducible, so A might not be a field. It is a separable K-algebra precisely when f(X) is a
          separable polynomial in K[X].
          When [A : K] is finite, an analogue of Theorem 1 can be proved: A is a separable K-algebra if and
          only if the trace pairing hx; yi = Tr A/K (xy) from A × A to K is non-degenerate.




                                           LOVELY PROFESSIONAL UNIVERSITY                                  255
   257   258   259   260   261   262   263   264   265   266   267