Canonical: https://knowlpedia.com/algebra-modules/polynomial-module-from-operator/

# Polynomial module associated with a linear operator

A linear operator T defines an F[X]-module whose action factors through the operator algebra F[T].

Let \(V\) be an \(F\)-[vector space](https://knowlpedia.com/linear-algebra/vector-space/) and \(T:V\to V\) an \(F\)-[linear map](https://knowlpedia.com/linear-algebra/linear-map/). The **polynomial module associated with \(T\)** is the \(F[X]\)-[module](https://knowlpedia.com/algebra-modules/module/) on \(V\) whose action is
\[
\left(\sum_{j=0}^m a_jX^j\right)\cdot v
=\sum_{j=0}^m a_jT^j(v),\qquad T^0=\operatorname{id}_V.
\]

This is the unique action of the [polynomial ring](https://knowlpedia.com/algebra-rings/polynomial-ring/) \(F[X]\) for which constants act by the original scalar multiplication and \(X\cdot v=T(v)\). Finite dimensionality is not required.

## Evaluation and the operator algebra

Write \(\operatorname{End}_F(V)\) for the ring of \(F\)-linear maps from \(V\) to itself, with pointwise addition and composition as multiplication. For \(p=\sum_i a_iX^i\in F[X]\), define \(p(T)=\sum_i a_iT^i\). The **operator algebra generated by \(T\)** is
\[
F[T]=\{p(T):p\in F[X]\}\subseteq\operatorname{End}_F(V).
\]
Here \(X\) is an indeterminate, whereas \(T\) is an operator. Distinct polynomials can determine the same operator.

Polynomial evaluation defines a surjective unital [ring homomorphism](https://knowlpedia.com/algebra-rings/ring-homomorphism/)
\[
\operatorname{ev}_T:F[X]\longrightarrow F[T],\qquad p\longmapsto p(T).
\]
To prove this, let \(p=\sum_i a_iX^i\) and \(q=\sum_j b_jX^j\). Addition is coefficientwise, so \((p+q)(T)=p(T)+q(T)\). Since \(T\) is \(F\)-linear, it commutes with every scalar operator \(a\operatorname{id}_V\). Hence
\[
p(T)q(T)=\sum_{i,j}a_ib_jT^{i+j}=(pq)(T).
\]
Also \(1(T)=\operatorname{id}_V\) and \(a(T)=a\operatorname{id}_V\) for \(a\in F\). These identities show that \(F[T]\) is a commutative unital \(F\)-subalgebra, even though \(\operatorname{End}_F(V)\) may be noncommutative. Surjectivity follows from the definition of \(F[T]\).

## The kernel and quotient identification

The kernel is the ideal
\[
I_T=\ker(\operatorname{ev}_T)=\{p\in F[X]:p(T)=0\}.
\]
It is an ideal because evaluation is a ring homomorphism: if \(p,q\in I_T\) and \(r\in F[X]\), then \((p-q)(T)=0\) and \((rp)(T)=r(T)p(T)=0\).

There is an isomorphism of \(F\)-algebras
\[
F[X]/I_T\longrightarrow F[T],\qquad p+I_T\longmapsto p(T).
\]
Indeed, \(p+I_T=q+I_T\) if and only if \(p-q\in I_T\), which is equivalent to \(p(T)=q(T)\). This proves both well-definedness and injectivity. Surjectivity and preservation of addition, multiplication, identity, and scalars follow from the corresponding properties of evaluation.

If \(V\) is finite-dimensional and nonzero, then
\[
I_T=(m_T),\qquad F[T]\cong F[X]/(m_T),
\]
where \(m_T\) is the [minimal polynomial](https://knowlpedia.com/linear-algebra/minimal-polynomial/) of \(T\). To see existence, put \(n=\dim_F V\). The \(n^2+1\) operators \(\operatorname{id}_V,T,\ldots,T^{n^2}\) are linearly dependent in the \(n^2\)-dimensional space \(\operatorname{End}_F(V)\). Thus \(I_T\) contains a nonzero polynomial. Choose a monic one \(m_T\) of least degree. No nonzero constant annihilates nonzero \(V\), so this degree is positive. For \(p\in I_T\), polynomial division gives \(p=qm_T+r\), with \(r=0\) or \(\deg r<\deg m_T\). Evaluation gives \(r(T)=0\), so minimality forces \(r=0\). Conversely every multiple of \(m_T\) belongs to \(I_T\). The monic generator is unique.

Finite dimensionality is needed here to guarantee a nonzero annihilating polynomial, not for evaluation or the quotient identification. If \(V=0\), then \(I_T=F[X]\) and \(F[T]\) is the zero ring.

## The two module structures and faithfulness

The \(F[T]\)-action is evaluation of operators on vectors:
\[
F[T]\times V\longrightarrow V,\qquad (A,v)\longmapsto A(v).
\]
For \(A,B\in F[T]\) and \(v,w\in V\), linearity and composition give
\[
A(v+w)=A(v)+A(w),\qquad (A+B)(v)=A(v)+B(v),
\]
\[
(AB)(v)=A(B(v)),\qquad \operatorname{id}_V(v)=v.
\]
These are the module axioms. Restricting scalars along \(\operatorname{ev}_T\) gives the original \(F[X]\)-action \(p\cdot v=p(T)(v)\). Equivalently, the action factors through \(F[X]/I_T\) by \((p+I_T)\cdot v=p(T)(v)\); this is independent of the representative because every element of \(I_T\) acts as zero.

An action is **faithful** if the only scalar that annihilates every vector is zero. The \(F[T]\)-action is faithful: an operator \(A\) satisfying \(A(v)=0\) for every \(v\in V\) is the zero operator. The \(F[X]\)-action has annihilator \(I_T\), so it is faithful exactly when \(I_T=0\). In particular it is not faithful when \(V\) is finite-dimensional and nonzero. Faithfulness does not require every nonzero scalar to act injectively.

## Example: a two-dimensional nilpotent operator

Let \(V=F^2\) with basis \(e_1,e_2\), and define \(T(e_1)=e_2\), \(T(e_2)=0\). Then \(T\ne0\) but \(T^2=0\). For \(p=\sum_j a_jX^j\),
\[
p(T)=a_0\operatorname{id}_V+a_1T,\qquad
p\cdot(ue_1+ve_2)=a_0u e_1+(a_0v+a_1u)e_2.
\]
Applying \(p(T)\) to \(e_1\) shows that \(p(T)=0\) if and only if \(a_0=a_1=0\). Thus \(I_T=(X^2)\), \(m_T=X^2\), and
\[
F[T]=\{a\operatorname{id}_V+bT:a,b\in F\}\cong F[X]/(X^2).
\]
The coefficients \(a,b\) are unique, since \((a\operatorname{id}_V+bT)(e_1)=ae_1+be_2\). Multiplication satisfies
\[
(a\operatorname{id}_V+bT)(c\operatorname{id}_V+dT)
=ac\operatorname{id}_V+(ad+bc)T.
\]

Over \(F[X]\), the nonzero scalar \(X^2\) annihilates all of \(V\), so the action is not faithful. Over \(F[T]\), the scalar \(T^2\) is already zero; a scalar annihilating \(V\) must have \(a=b=0\), so the action is faithful. The nonzero scalar \(T\) still kills \(e_2\), illustrating that faithfulness concerns annihilating every vector.

The vector \(e_1\) generates both modules. More precisely,
\[
F[X]/(X^2)\longrightarrow V,\qquad p+(X^2)\longmapsto p(T)e_1
\]
is an \(F[X]\)-module isomorphism, whereas \(F[T]\to V\), \(A\mapsto A(e_1)\), is an \(F[T]\)-module isomorphism. The first uses polynomial scalars acting through the quotient; the second identifies \(V\) with the regular module of the operator algebra.

## Morphisms and invariant subspaces

For operators \(T\) on \(V\) and \(S\) on \(W\), a linear map \(u:V\to W\) is an \(F[X]\)-module homomorphism exactly when \(uT=Su\).

The condition \(uT=Su\) says that the square below commutes:

```tikz-cd
V \arrow[r, "u"] \arrow[d, "T"'] & W \arrow[d, "S"] \\
V \arrow[r, "u"'] & W
```

A subspace is an \(F[X]\)-submodule exactly when it is \(T\)-invariant.

For finite-dimensional nonzero \(V\), the [minimal polynomial](https://knowlpedia.com/linear-algebra/minimal-polynomial/) generates the annihilator of this module. The construction is the bridge to [rational canonical form from the module structure theorem](https://knowlpedia.com/algebra-modules/rcf-from-structure-theorem/).

## References

1. The polynomial identities above verify existence directly; the images of constants and \(X\) prove uniqueness. The intertwining condition extends from \(X\) to every polynomial by induction and linearity.
