Motivation
Writing a module as a direct sum is useful only if we understand the pieces and the extent to which they are determined. An indecomposable module cannot split further, but that definition alone does not ensure different decomposition attempts produce the same pieces. The uniqueness theorem needs a mechanism connecting two apparently unrelated decompositions.
The mechanism is local behavior inside the endomorphism ring of an indecomposable finite-dimensional module. Every endomorphism there is either invertible or nilpotent, and the noninvertible maps form an ideal. Comparing two decompositions expresses an identity map as a sum of composites; one composite must then be invertible. It identifies a matching summand, and an actual exchange argument permits cancellation.
Let be a unital -algebra, possibly infinite-dimensional, and consider only finite-dimensional left -modules. Their -endomorphisms are -linear, so kernels and images stabilize by vector-space dimension. The proof also applies to the earlier finite-dimensional-algebra examples and to finite-dimensional -modules. No classification of all indecomposables for an arbitrary algebra is promised.
Definitions
Definition 12.1.1
(Indecomposables and a local endomorphism ring)
A nonzero module is indecomposable if implies or . Its endomorphism ring has addition of maps, multiplication by composition, and identity . A not-necessarily-commutative ring is called local here when its nonunits form a proper two-sided ideal; its quotient by that ideal is a division ring. This condition implies that this ideal is the unique maximal left ideal and the unique maximal right ideal.
A finite indecomposable decomposition is an isomorphism from to a finite direct sum of nonzero indecomposables. The uniqueness below concerns their isomorphism types and multiplicities. It does not specify preferred submodules, preferred bases or a canonical isomorphism between two displayed sums. The zero module has the empty decomposition.
Theorems
Theorem 12.1.2
(Fitting's decomposition and the local endomorphism test)
For any endomorphism of a finite-dimensional -module , there is a positive integer such that
The summands are -submodules. If is nonzero, it is indecomposable if and only if is local. In the indecomposable case every nonunit endomorphism is nilpotent.
The second theorem uses the first only through the locality of for each indecomposable summand.
Theorem 12.1.3
(Krull–Schmidt uniqueness and cancellation)
Every finite-dimensional -module has a finite indecomposable decomposition. If
are two such decompositions, then and there is a permutation matching each to an isomorphic . Consequently, for finite-dimensional modules , an isomorphism implies .
Concept lensStructural
Decompositions as coordinates in a free monoid
Write for the isomorphism class of and set . The classes of finite-dimensional -modules then form a commutative monoid with identity . The theorem says exactly that this monoid is free on the classes of indecomposables: each has unique coordinates, the multiplicities of the indecomposable types , and . Cancellation becomes subtraction of natural-number coordinates.
Proof sketch or proof idea
Stabilized kernels and images give an honest direct sum
Proof
(Fitting's decomposition)
Choose so large that the increasing kernel chain and decreasing image chain of have both stabilized; this happens because the dimensions are bounded by . In particular . If lies in the intersection of image and kernel, then , so . The intersection is zero. Rank–nullity gives , hence the direct sum. Both parts are submodules because is -linear.
If is indecomposable, one summand is zero. If the kernel is zero, is injective and therefore invertible by finite dimension. Its inverse is -linear, since applying to the required linearity identity proves it. If the image is zero, . Thus a noninvertible is nilpotent. Its complement is invertible, with inverse the finite geometric sum .
Nonunits really form an ideal
Proof
(Locality of the endomorphism ring of an indecomposable)
Let for an indecomposable . If is a nonunit and any endomorphism, neither nor can be invertible: invertibility of the first makes injective, and invertibility of the second makes onto, either of which would make invertible in finite dimension. Thus multiplication on either side preserves nonunits.
They are closed under addition as well. Suppose are nonunits but is a unit. Then and are nonunits summing to . The first is nilpotent, so must be a unit; that complement is , a contradiction. Negatives preserve noninvertibility. Hence the nonunits form a proper two-sided ideal, and every nonzero coset in the quotient has an inverse represented by a unit. This proves locality rather than merely asserting it from the invertible-or-nilpotent dichotomy.
Conversely, a local ring has no idempotents other than and . For an idempotent , either or is a unit, since both cannot lie in the nonunit ideal and sum to . If is a unit, gives ; if is a unit, gives . A nontrivial decomposition of would give a nontrivial idempotent projection in its endomorphism ring. A local endomorphism ring therefore forces indecomposability.
Find one common isomorphism type inside two decompositions
Proof
(Existence, and one matching summand)
Existence follows by repeatedly splitting a decomposable nonzero summand. Each split replaces it by modules of strictly smaller positive dimensions, and the total number of nonzero summands never exceeds . Thus the process terminates with indecomposables.
For uniqueness, regard both displayed sums as internal decompositions of the same . Include into , project onto each and call that composite . Let be inclusion followed by projection onto . The direct-sum identities give
The endomorphism ring of is local, so not every summand on the right can be a nonunit. Choose with invertible. Then has a left inverse , so its nonzero image is a direct summand of . Indecomposability of forces the complementary summand to vanish. Thus is an isomorphism .
Proof X-ray
An identity written as a sum identifies a matching summand
The compared maps pass through the one at a time. They need not individually be projections or inclusions onto a common submodule. Locality says that a sum of nonunits cannot equal the identity. One invertible composite supplies a split injection into an indecomposable , forcing it to fill that entire summand. This is where the matching is proved; it is not inferred from equality of dimensions.
Exchange the complement before canceling
Proof
(The exchange step and induction)
Relabel the matching as , and write for the sum of the other . The projection restricts to the isomorphism on . Consequently meets trivially, since is the kernel of that projection, and every differs from the element of by something in . Therefore .
On the other hand , where is the sum of the remaining . Taking the quotient by this same actual submodule identifies . Apply induction on dimension to these smaller modules. This proves all remaining matches. The exchange step supplies a valid cancellation argument instead of assuming the cancellation conclusion while trying to prove uniqueness.
Finally, decompose into indecomposables. Uniqueness for gives equal multiplicities of each type. Subtract the finite multiplicities contributed by to obtain matching decompositions of and , proving cancellation.
Proof X-ray
Why the exchange avoids assuming cancellation
After matching , one might simply cancel these summands, but that is the cancellation statement still being proved. The exchange step instead shows that the actual submodule has as a complement in , because the projection onto restricts to an isomorphism on . Then and are two complements of the same submodule, and every complement of is isomorphic to . Finite dimension enters twice: locality produces the match, and the induction runs on .
Worked examples
Worked example 12.1.4
(A single primary polynomial block is indecomposable)
For a monic irreducible in , consider with as a -module. Every endomorphism is determined by the image of the cyclic generator , and every choice in defines multiplication by that element. Thus . Its units are exactly the classes not divisible by , by the polynomial Bézout identity. Its nonunits form the ideal , so the endomorphism ring is local and the module is indecomposable.
The earlier PID structure theorem decomposes any finite-dimensional -module into these blocks. Krull–Schmidt explains uniqueness of their module isomorphism types. Over an algebraically closed field, gives the familiar Jordan blocks. This argument does not require a basis calculation for every possible invariant subspace of a block.
Worked example 12.1.5
(One linear map classifies the upper-triangular algebra's modules)
Let be the upper triangular matrix algebra, with diagonal idempotents and . A module is the vector-space sum , where . The action of is a map , zero on ; diagonal entries act as the corresponding scalars. These data determine the entire module action.
Choose and . Pairing a basis of with its image gives one copy of the two-dimensional indecomposable for each unit of rank. The remaining basis vectors give the simple modules and . Thus
The three displayed types are indecomposable: the simples are immediate, and has simple top as proved with projective covers. For , and , the multiplicities are two, two and one. This is a full classification for this particular algebra, rather than a claim about arbitrary finite-dimensional algebras. Different choices of complements give isomorphic decompositions, consistent with uniqueness up to isomorphism.
Worked example 12.1.6
(Changing the ground field can split an indecomposable)
Let the generator of act on by rotation through . Its minimal polynomial is , irreducible over , so the real representation is a cyclic -module with that quotient. Its endomorphism ring is , a division ring. It is therefore indecomposable, indeed irreducible, over .
After extending scalars to , the generator has distinct eigenvalues and . Its representation splits as the sum of their two one-dimensional eigenspaces. Krull–Schmidt has not failed: it compares decompositions in a fixed module category with a fixed coefficient field. Scalar extension has changed that category and its endomorphism rings.
Common mistakes
Counterexample mode
Indecomposable infinite modules need not have local endomorphism rings
The -module is indecomposable: any two nonzero subgroups and contain , so they cannot be complementary summands. Its endomorphism ring is , which is not local: and are both nonunits although their sum is . Multiplication by is neither invertible nor nilpotent. Its descending image chain never stabilizes. Finite-dimensional stabilization was an actual hypothesis, not an optional convenience.
Common mistake
(Equal dimensions, or indecomposability, are not stronger than they look)
Equal dimensions do not make modules isomorphic, and an indecomposable need not be simple: for is indecomposable by the local-ring argument, yet it has the proper submodule . Likewise, uniqueness of summand types does not select canonical invariant subspaces or bases.
Common mistake
(Nilpotent nonunits are special to the indecomposable case)
That every nonunit is nilpotent holds in the endomorphism ring of an indecomposable here, not in general. In the matrix ring the nilpotent matrix units and sum to the invertible matrix , so the nonunits do not form an ideal; correspondingly is decomposable.
Counterexample mode
Cancellation fails for infinite-dimensional modules
Claim: implies for all modules. Take , , and , the vector space with basis . The linear map sending the basis vector of to and each to carries a basis of bijectively onto a basis of , so , although . The repair is the hypothesis in the theorem: with finite-dimensional, its multiplicities are finite and can be subtracted.
Quick check
Quick check
(If a nonzero finite-dimensional module has a nontrivial idempotent endomorphism, can it be indecomposable?)
Look at the image and kernel of the idempotent.
Solution · Answer
No. For an idempotent , with both summands nonzero. Conversely a nontrivial direct sum provides just such a projection.
Exercises
Quick check
(Give a direct endomorphism-ring proof that is indecomposable, identifying all invertible and nilpotent endomorphisms.)
Every endomorphism is multiplication by a residue polynomial.
Solution · Answer
An endomorphism is multiplication by a residue polynomial . It is invertible exactly when the constant term of is nonzero, using a finite geometric-series inverse after extracting that scalar. If the constant term is zero, belongs to , so . The nonunits form the ideal , giving locality and indecomposability. This identifies nilpotent nonunits, including the zero map.
Quick check
(Prove that for a nonzero finite-dimensional the module is decomposable, and exhibit a nontrivial idempotent in its endomorphism ring. Why is that ring not local even if is local?)
Project onto the first copy.
Solution · Answer
Projection onto the first copy is idempotent, nonzero and not the identity; its complementary projection is also nonzero. A local ring has no such idempotent by the theorem's elementary argument. Thus this endomorphism ring is not local; decomposing the source changes the conclusion even if each individual copy has a local endomorphism ring.
Quick check
(A nilpotent operator on a six-dimensional space has for . Determine its indecomposable -block sizes and justify the calculation.)
A block of size contributes to .
Solution · Answer
A block of size contributes to . Successive increments are therefore the numbers of blocks of size at least . Here they are . Subtracting consecutive increments gives one block each of sizes one, two and three. Their total dimension is six, and their kernel counts reproduce the data. The PID decomposition gives existence of these blocks; Krull–Schmidt gives the uniqueness of their multiset.
Quick check
(Explain why cancellation in the theorem applies to any finite-dimensional , even if is decomposable, and why it does not assert a preferred isomorphism .)
Count multiplicities of indecomposable types.
Solution · Answer
Decompose into finitely many indecomposables and compare multiplicities in the two enlarged decompositions. Each contributes the same finite count on both sides and can be subtracted. Reassembling the remaining matched types gives an isomorphism , but comparisons between isomorphic summands and their internal decompositions involved choices. The argument proves existence and multiplicity uniqueness, not a distinguished map.
Solutions and related notes
Projective covers provide a special family of indecomposables indexed by simple tops; this theorem applies to all finite-dimensional modules. Classifying the indecomposable types for a general algebra, modular blocks and representation type requires further work beyond uniqueness of decompositions.