A module can contain an invariant submodule without containing an invariant complement. We therefore need a way to describe its pieces that does not begin by asking for a direct-sum decomposition. A composition series does this by repeatedly taking a submodule and a quotient. The Jordan–Hölder theorem says that the resulting list of simple pieces is independent of the choices, apart from their order and replacement by isomorphic modules. It does not say that the pieces can be reassembled in only one way.
This chapter supplies the finite-length foundation used when later notes count composition factors and add their characters. We use the definitions and quotient correspondence from modules, quotients, and homomorphisms. Throughout, is an associative ring with identity, and modules are unital left -modules. Commutativity, finite generation, and a ground field are imposed only where stated.
Simple pieces and composition series
Definition 1.3.1
(Simple modules and composition series)
A module is simple if and its only submodules are and . A composition series of is a finite chain
such that every quotient is simple. These quotients are its composition factors. The zero module has the empty series, with no factors. A module has finite length if it has a composition series.
Strict containment matters: a zero quotient is not simple. So does the word quotient. The factor is generally not a specified submodule of , and a series is not a direct-sum decomposition. The quotient correspondence shows that is simple precisely when there is no submodule strictly between its two endpoints. Thus a composition series is a finite chain in which every step has been refined as far as possible.
For example, take over a field . The chain is a composition series of the regular module. Both factors are the one-dimensional module on which acts as zero. Nevertheless is not the direct sum of these two simple modules: multiplication by is nonzero on , whereas it is zero on .
Why termination needs two chain conditions
Definition 1.3.2
(Ascending and descending chain conditions)
A module is Noetherian if every ascending chain of submodules is eventually constant. It is Artinian if every descending chain of submodules is eventually constant. These are conditions on the module under discussion, not automatic consequences of using a familiar ring.
The ascending condition is equivalent to saying that every submodule is finitely generated. Indeed, if a submodule has no finite generating set, start with zero and repeatedly adjoin an element outside the submodule generated so far. This gives an ascending chain that never stops. Conversely, the union of an ascending chain is a submodule. If that union is generated by finitely many elements, all generators lie in one member of the chain; every later member then equals the union.
In particular, a Noetherian module is finitely generated by applying this assertion to the whole module. The converse need not hold over an arbitrary ring: finite generation of alone does not assert finite generation of all its submodules.
Theorem 1.3.3
(Finite length and chain conditions)
A module has finite length if and only if it is both Noetherian and Artinian. If it has a composition series with factors, every finite strictly increasing chain of its submodules has at most strict steps. Consequently all its composition series have the same number of factors. This number is the length , with .
A bound for every chain
Proof
(Proof of the chain bound)
We first prove the asserted bound by induction on the number of factors in one fixed composition series. For , the module is zero. For , let . It has a composition series with factors, and is simple. For a chain of submodules of , consider the two chains
The first can change strictly at most times by induction; repeated terms can be removed when applying that statement. The second can change strictly at most once, because a simple module has only zero and itself as submodules.
Every strict step in the original chain must change at least one of these two chains. To check this, suppose and both the intersection and image remain unchanged. Given , equality of images supplies with . Then , so . Hence . There are therefore at most strict steps.
An infinite ascending or descending chain that did not stabilize would supply finite chains with arbitrarily many strict steps. Reversing the order of a finite descending chain makes it increasing. Both possibilities contradict the bound, so is Noetherian and Artinian. If another composition series has factors, this bound gives ; applying the bound using the other series gives . Thus the lengths agree, before we have proved uniqueness of the factors themselves.
Constructing a series from the two conditions
Proof
(Building a series from maximal submodules)
Assume now that has both chain conditions. Every nonzero submodule has a maximal proper submodule: otherwise, starting at zero, one could keep enlarging a proper submodule of , contradicting the ascending condition inherited from . Choose such a maximal submodule of , then a maximal proper submodule of that submodule if it is nonzero, and continue. The descending condition prevents this process from continuing indefinitely. It must reach zero, since a nonzero last term would still have a maximal proper submodule. Reversing the resulting chain gives a composition series, because maximality makes each quotient simple. This proves the converse.
Submodules inherit both chain conditions. Quotients inherit them by lifting chains through the quotient correspondence. Hence a submodule and a quotient of a finite-length module again have finite length. In particular, a proper submodule of a module of length has length at most : append to a composition series of and apply the chain bound. This is the precise smaller-size fact needed in the next proof.
Counterexample mode
One chain condition does not imply the other
The -module is Noetherian, since every subgroup is cyclic. But never stabilizes. Thus even a cyclic module over a Noetherian ring need not have finite length.
For the reverse direction, fix a prime and put . Write , a cyclic subgroup of order , with . Every element of order generates . A subgroup with elements of unbounded orders therefore contains every and equals . A subgroup with bounded orders lies in a finite cyclic , so is some . This describes all subgroups. A descending chain either stays at , or enters a finite subgroup and then stabilizes. Thus is Artinian. But shows that it is not Noetherian and has no finite length.
Why the simple factors are independent of choices
Theorem 1.3.4
(Jordan–Hölder for finite-length modules)
For any two composition series of a finite-length -module , there is a bijection between their factors that pairs isomorphic simple modules. Thus, for each isomorphism class of simple modules , the number of factors isomorphic to is well defined. It is called the composition multiplicity of in .
Proof
(Induction through the intersection of two maximal submodules)
We prove this by induction on . The zero module has no factors. A module of length one is simple and has the single factor . For the induction step, let and be the penultimate submodules in two composition series. They are maximal proper submodules, because and are simple.
If , the induction hypothesis compares the two series inside , whose length is smaller than . Appending their common final factor completes this case.
Suppose , and put . Neither maximal submodule can contain the other: such containment would force equality or make one equal to . Therefore properly contains , so maximality gives . The map
has kernel and is onto, since . The first isomorphism theorem, and then the same argument with the roles reversed, give
In particular, both quotients on the left are simple. Choose a composition series of , which exists because is a submodule of a finite-length module. Append to obtain a composition series of . By induction inside , its factors agree with those from the first original series below . Accordingly, the first original series has the factors of , followed, as a list of isomorphism classes, by
Appending instead gives a composition series of . Induction inside shows that the second original series has the same factors of , followed by and . Interchanging these last two factors gives the required pairing. Every invocation of induction concerned a proper submodule of , so no uniqueness assertion has been used at its own size. The proof is complete.
Proof X-ray
The intersection repairs two incompatible first choices
Different maximal submodules need not be isomorphic. The proof does not try to identify them. Their intersection gives a common smaller module, and the two quotients above exchange the final simple factors. Induction compares the factors inside the smaller modules. This is why the theorem promises the same factors with multiplicity and permits a different order; it supplies no preferred chain and no invariant complements.
Additivity along a short exact sequence
Theorem 1.3.5
(Length and multiplicity additivity)
For a short exact sequence , the module has finite length if and only if both and do. In that case, for every simple ,
No splitting assumption is required.
Proof
(Concatenating series through the quotient map)
If has finite length, so do its submodule and quotient , as already proved. Conversely, identify with its image, choose composition series and , and concatenate
Surjectivity of makes the lifted containments strict. For , the map from to induced by is onto, with kernel . Its quotient is therefore simple and isomorphic to . The concatenation is a composition series of , with precisely the factors of and . Counting them proves both formulas; Jordan–Hölder ensures that these counts are intrinsic.
This makes composition multiplicities useful in representation theory. If a character construction is additive on short exact sequences, repeated use of the sequence for each step of a composition series expresses the character as a sum of simple characters with coefficients . Such additivity does not require the representation to be semisimple. Recovering the coefficients from character values is a separate question of independence of the particular simple characters.
Worked example 1.3.6
(Two series with differently ordered factors)
In the -module , write for the image of multiplication by . Both and are composition series. The subgroup orders in the first are , giving factors of orders . In the second they are , giving factors of orders . A group of prime order is a simple -module. Thus both series have two factors isomorphic to and one isomorphic to , even though their orders differ. This is exactly the freedom that Jordan–Hölder allows.
What finite vector-space dimension does and does not imply
Let be a unital algebra over a field , and let be an -module whose underlying -vector space has finite dimension, with compatible scalar action. Every -submodule is a -subspace, so strict containment raises dimension. Thus both chain conditions hold and has finite length. Dimension additivity along its series gives
Only the finitely many simple classes occurring in contribute. Equality in the inequality holds exactly when every occurring simple factor has -dimension one. There is no need to assume that itself is finite-dimensional for this argument.
Worked example 1.3.7
(A simple module can have dimension two)
Let act on through . Denote this operator by , so . A nonzero proper submodule would have to be a -invariant real line. On that line would act as multiplication by some real , and would force . This is impossible, so is simple and , although .
If we forget the action of and retain only the -module structure, every line is a submodule and the length becomes two. Length depends on the scalar ring and its action, not just on the underlying additive group.
Worked example 1.3.8
(The factors can agree while the modules differ)
Return to and its simple module . Both and have composition multiplicity two for and zero for every other simple. Both have length two. They are not isomorphic, because an -linear isomorphism would intertwine multiplication by , which has rank one on and rank zero on .
The sequence cannot split. A section would send to an element with . Linearity would require , but . The simple factors record what is joined together; the nonsplit sequence records information about how it is joined.
Misconceptions to repair
Common mistake
(A simple-quotient chain need not terminate)
A chain with simple successive quotients does not automatically terminate: finiteness is part of the definition of a composition series. Finite generation alone also does not guarantee termination, as demonstrates. On the other hand, finite-dimensionality of a representation is sufficient even if its coefficient algebra is infinite-dimensional.
Common mistake
(Composition factors do not determine the module)
The multiset of composition factors classifies the direct sum of those simple factors, often called a semisimplification, up to isomorphism. It need not classify the original module, and the construction gives no canonical isomorphism from that direct sum to the module. Length counts all simple quotients in a composition series. Loewy length instead counts radical layers, which may each contain several simple summands. For example, a direct sum of ten simple modules has composition length ten and Loewy length one.
Exercises
Quick check
(Regard as a -module. Construct a composition series using the subgroups generated by , in that order. Determine every nonzero composition multiplicity and explain why another series cannot change them.)
Read off the subgroup orders along the chain.
Solution
Writing for the subgroup generated by the class of , the series is . Its subgroup orders are . Successive quotients have prime orders , so they are the simple -modules twice and three times. Thus the length is five and these are its only nonzero multiplicities. Jordan–Hölder preserves their isomorphism classes with these multiplicities, not their displayed order.
Quick check
(Let be a finite-length module and linear. Prove separately that injectivity implies surjectivity and that surjectivity implies injectivity. Explain the failure of the first assertion for , and of the second for the module above.)
Apply length additivity to the sequences built from and .
Solution
Apply length additivity to and . If the kernel is zero, the image has the same length as , forcing the quotient to have length zero and hence to be zero. If the image is , the first sequence forces the kernel to have length zero. Multiplication by two on is injective and not onto. Multiplication by on is onto, since one can increase the denominator by a factor of , but it kills the nonzero subgroup .
Quick check
(Let be the real operator module in the worked example “A simple module can have dimension two”. Determine the length and simple multiplicities of as an -module and as an -module. Why does real dimension alone not give the first length?)
Use additivity together with the simplicity of .
Solution
As an -module the length is two and , by additivity and simplicity of . As an -module it has length four, with four copies of the one-dimensional simple module . The operator action disallows invariant real lines in each , so a real dimension of four does not imply four operator-module factors.
Quick check
(If is any finite filtration and every quotient has finite length, prove that has finite length and that its composition multiplicities are the sums of the quotient multiplicities. Repeated terms are allowed here.)
Induct on with the sequence .
Solution
Induct on , applying the short exact sequence . The zero starting module has finite length; each step gives finite length and adds the quotient multiplicities. A repeated term contributes the zero quotient, with length zero and no factors. This proves the claim without assuming that the filtration itself is a composition series.
Quick check
(Over , find the composition length of . Can these multiplicities distinguish from ? Give a linear-action invariant that does distinguish them.)
Compare the rank of multiplication by .
Solution
Additivity gives length , with . Those multiplicities agree with , so they do not distinguish the modules. Multiplication by has rank three on , one from each regular summand, and rank zero on . An -module isomorphism would preserve that rank, so no such isomorphism exists.