Turning a factorization into projections
An operator may be difficult to understand on a whole vector space but simple on pieces. The same idea works for modules without choosing a basis. Suppose that is a commutative ring with identity, is a unital -module (so ), and satisfy and . The first condition means that there are with . The second says that for every ; neither nor has to be zero.
Definition 10.1.1
(The part of killed by )
For , put . It is a submodule: , and because is commutative.
The decomposition we want says that is assembled from the parts killed by the two factors.
Theorem 10.1.2
(Two-factor primary decomposition)
If and , then
If , the projections onto and are and .
Proof
(Splitting )
Write any as . The first term lies in , since ; the second lies in , since . If belongs to both submodules, then , so the sum is direct. Thus the two projections are and .
Proof X-ray
Each hypothesis has one job
The annihilation condition is used only to place the two pieces of in the correct kernels. Comaximality is used twice through the same identity : read as it gives existence of the decomposition, and read as it gives uniqueness, because both terms vanish on an element of . No basis, finiteness or division is needed.
The coefficients need not be unique, but the projections are: for example, the projection onto is the unique map that fixes and kills . This also shows that the construction is natural: every -linear map between two such modules preserves both summands and commutes with the projections.
Worked example: an abelian group of order twelve
Worked example 10.1.3
(Splitting with the factors and )
Take as a -module, and . We have . Therefore
where is killed by and is killed by . More generally, every class has components and . The first submodule is and the second is . Their intersection is zero and their direct sum has all twelve elements.
Common mistake
(Reading as the elements of order exactly four)
is not the subgroup of elements of order exactly four: it includes zero and the element of order two. It consists of the elements whose order divides four.
Worked example 10.1.4
(A unit factor)
The degenerate case is allowed: for every , with and . The hypothesis now says , and . The theorem gives , and the projection formulas give and , as they must.
From two factors to many
Theorem 10.1.5
(Decomposition along pairwise comaximal factors)
Suppose are pairwise comaximal, meaning for , and . Then
Proof
(Induction on the number of factors)
Apply the two-factor result to and . Pairwise comaximality implies : multiplying the equations for gives an expression for as an element of plus a multiple of . Hence . The module is killed by , so induction applies to it. Finally, for the part of killed by is exactly , since already forces . This yields the displayed decomposition.
For an integer annihilating an abelian group, this recovers its -primary pieces without assuming the group is finite. The exponent is a valid common bound; it need not be the least exponent that annihilates a piece.
Worked example 10.1.6
(Three factors of )
Let with the pairwise coprime factors . The parts are , and . The integers satisfy
and each is divisible by the product of the other two factors (, and ). Multiplication by therefore kills and , fixes and lands in ; by uniqueness it is the projection onto , and similarly for and . For the components are , and , and indeed .
Why the hypotheses cannot be omitted
The two hypotheses play different roles, and each can fail on its own.
Counterexample mode
Dropping coprimality or the annihilation condition
Let and take . Although , both are , so their sum is not and is not direct. Here . Conversely, if , on , the factors are coprime but and both kernels are zero. Repeated factors must first be grouped into a single power, as in for , before the theorem is applied.
Exercises with explanations
Quick check
(Split using the projections above.)
Solution · Answer
and ; their sum is .
Quick check
(Suppose a linear operator on a vector space satisfies and the field has characteristic different from two. Show .)
Solution · Answer
View as an -module with acting by ; and are comaximal because their difference is the invertible constant , and annihilates .
Quick check
(What fails in the decomposition above in characteristic two?)
Solution · Answer
The factors coincide, so comaximality fails. For on a nonzero space the two kernels are both , and the displayed sum is not direct.
Quick check
(Let be -linear and assume both modules are annihilated by . Prove that .)
Solution · Answer
For , . The same argument applies to , hence respects the decompositions.