Primitive Idempotents and Minimal Left Ideals

Introduction

An idempotent $e$ splits the regular left module of an algebra into two summands,

$$ A = Ae \oplus A(1-e), $$

and a primitive idempotent is one that cannot be split further: it is not a sum of two nonzero orthogonal idempotents. A minimal left ideal is a nonzero left ideal with no nonzero proper left subideal. The two notions meet in the corner $eAe$, and the main theorem of the article is that, in a semisimple algebra, the following four conditions are equivalent:

$$ e\ \text{primitive} \iff Ae\ \text{minimal left ideal} \iff eA\ \text{minimal right ideal} \iff eAe\ \text{division ring}. $$

The equivalence is the idempotent form of the Wedderburn–Artin theory, and it is the reason a primitive idempotent indexes a simple module.

Remark (the carrier comes first). The object on which the minimality is read is fixed by the idempotent before any minimality is mentioned:

$$ C(e)=Ae=eAe\oplus fAe . $$

The spinor is an element of $C(e)$, written in coordinates over $eAe$. The minimality of $Ae$, and the division property of $eAe$, are consequences of primitivity and semisimplicity, not ingredients of the definition. The coordinate theory is Peirce Coordinates and Companion Nilpotents, and the definition of a spinor on the carrier is Spinors as Peirce Coordinates.

The equivalence is not available without semisimplicity, and the article is careful about the failure. The unconditional statement is one-directional: a minimal one-sided ideal generated by $e$ forces $eAe$ to be a division ring, but the converse fails, and it fails on the two sides independently. The upper triangular algebra $T_2(k)$ shows the asymmetry in its smallest form: $E_{11}$ generates a minimal left ideal and not a minimal right ideal, while $E_{22}$ generates a minimal right ideal and not a minimal left ideal, and both corners are division rings. The dual numbers show the other failure: $e=1$ is primitive, but $eAe=A$ is local and not a division ring, and neither one-sided ideal is minimal.

The article is the idempotent side of the module theory of Simple and Semisimple Modules. The Peirce decomposition is that of Unital Algebras, §Idempotents and the Peirce Decomposition; the complement operation, the complete families of primitive idempotents and the refinement of the Peirce decomposition are Complementary Idempotents and the Peirce Decomposition; the corner $eAe$ and its unit $e$ are the algebra of the corner; the left and right multiplications are Left and Right Multiplication; the matrix case is Matrix Algebras and The Column Ideals of a Matrix Ring; the dual-numbers example is Roots of the Identity, Split Algebras and Nontrivial Idempotents; and the two-sided structure theory is Simple and Semisimple Modules, §Semisimple Algebras and Wedderburn–Artin. The article uses no form, no involution and no topology.

Throughout, $k$ is a field and $A$ is a unital associative $k$-algebra with $1\neq0$, not assumed commutative and not assumed finite-dimensional unless a statement says so. The idempotent is $e$, and $f=1-e$. We write $Ae$ for the left ideal generated by $e$ and $eA$ for the right ideal; the two sides are dual as far as the definitions go, but not as far as the minimality criterion goes, and that asymmetry is one of the subjects of the article. The algebra is semisimple when its left regular module is a direct sum of simple modules, in the sense of Simple and Semisimple Modules.

Primitive Idempotents

Definition

Definition. A nonzero idempotent $e\in A$ is primitive when it is not a sum of two nonzero orthogonal idempotents: there are no idempotents $e_1,e_2$ with

$$ e_1\neq0,\qquad e_2\neq0,\qquad e_1e_2=e_2e_1=0,\qquad e=e_1+e_2 . $$

The zero idempotent is excluded, and $e=1$ is primitive exactly when $A$ has no nontrivial idempotent.

The definition is the idempotent form of indecomposability. A decomposition $e=e_1+e_2$ of the stated kind gives a direct sum $Ae=Ae_1\oplus Ae_2$ of nonzero left ideals, and a direct sum decomposition of $Ae$ gives such a decomposition of $e$. The equivalence is proved below.

The corner criterion

Proposition. Let $e\neq0$ be an idempotent. The following are equivalent.

(a) $e$ is primitive.

(b) The corner $eAe$ has no idempotents other than $0$ and $e$.

(c) The left module $Ae$ is indecomposable: it is not a direct sum of two nonzero left ideals.

(d) The right module $eA$ is indecomposable.

Proof. (a)$\Leftrightarrow$(b): if $q\in eAe$ is an idempotent with $q\neq0,e$, then $e-q$ is idempotent, nonzero, and orthogonal to $q$, and $e=q+(e-q)$; conversely, if $e=e_1+e_2$ with nonzero orthogonal idempotents, then $e_1=ee_1e$ lies in $eAe$ and is neither $0$ nor $e$. (b)$\Leftrightarrow$(c): if $Ae=I\oplus J$ with $I,J\neq0$, then $e=e_I+e_J$ with $e_I,e_J$ the components of $e$; as $e$ is the identity on $Ae$, the two components are idempotent and orthogonal, and they are nonzero because $I,J\neq0$. Conversely $e=e_1+e_2$ gives $Ae=Ae_1\oplus Ae_2$ with $Ae_i\neq0$. The right-hand statements are the same with the products reversed.

Remark. The proposition is the reason the word "primitive" is used for the idempotent and "indecomposable" for the module: they are the same condition, read on the element and on the module it generates. The corner $eAe$ is the ring of endomorphisms of the module in the opposite order, as §The Division Ring Criterion shows.

Examples

Example (matrix algebras). In $M_n(k)$ the primitive idempotents are the rank-one idempotents. A rank-$r$ idempotent $e$ has corner $eAe\cong M_r(k)$; for $r\geq2$ the matrix unit $E_{11}$ of the corner is an idempotent different from $0,e$, so $e$ is not primitive. For $r=1$ the corner is $k$, a division ring, and $e$ is primitive. The left ideal $Ae$ consists of the matrices supported in the first $r$ columns; in a basis in which $e$ is the projection onto the first $r$ coordinates, it is the span of the first $r$ standard columns, of dimension $rn$. It is minimal exactly for $r=1$.

Example (the split algebra). In $A=k\times k$ the primitive idempotents are $e_1=(1,0)$ and $e_2=(0,1)$. Each corner is $k e_i\cong k$, a division ring, and the left ideals $Ae_i=k e_i$ are the two coordinate lines, both minimal. The identity $1=e_1+e_2$ is not primitive, and $A=Ae_1\oplus Ae_2$ is the decomposition that makes $A$ split.

Example (the upper triangular algebra, one side at a time). Let $A=T_2(k)$ be the algebra of upper triangular $2\times2$ matrices. The idempotents $E_{11}$ and $E_{22}$ are primitive: their corners are $kE_{11}$ and $kE_{22}$, division rings. But the one-sided ideals are not symmetric:

$$ AE_{11}=kE_{11}, \qquad E_{11}A=kE_{11}\oplus kE_{12}, $$ $$ AE_{22}=kE_{12}\oplus kE_{22}, \qquad E_{22}A=kE_{22}. $$

So $E_{11}$ generates a minimal left ideal and a nonminimal right ideal, while $E_{22}$ generates a minimal right ideal and a nonminimal left ideal. The algebra is not semisimple, and the example shows that the two sides of the primitive idempotent can separate. The computation is examined again in §Without Semisimplicity.

Example (the dual numbers). Let $A=k[\varepsilon]/(\varepsilon^2)$ with $k$ of characteristic not two. The only idempotents are $0$ and $1$, so $e=1$ is primitive; the corner $eAe=A$ has no nontrivial idempotent. But $A$ is not a division ring, and $Ae=A$ is not a minimal left ideal: $(\varepsilon)$ is a nonzero proper left ideal. The same holds on the right. This is the smallest example in which primitivity does not give minimality, and it is examined in §Without Semisimplicity.

Minimal Left Ideals

Definition

Definition. A nonzero left ideal $L\subseteq A$ is minimal when it contains no left ideal other than $0$ and $L$. Equivalently, $L$ is a simple left $A$-module. A minimal right ideal is defined in the same way on the right.

A minimal left ideal is necessarily of the form $Ae$ for an idempotent $e$ when it is a direct summand of $A$; in a semisimple algebra every left ideal is a direct summand, so every minimal left ideal is generated by an idempotent. Without semisimplicity a minimal left ideal need not be a direct summand and need not be generated by an idempotent; the general theory belongs to Simple and Semisimple Modules.

The division ring criterion

The endomorphism ring of $Ae$ computes the corner, and Schur's lemma turns minimality into division.

Proposition (necessity). Let $e\neq0$ be an idempotent. If $Ae$ is a minimal left ideal, then $eAe$ is a division ring. If $eA$ is a minimal right ideal, then $eAe$ is a division ring.

Proof. The endomorphism ring of the left module $Ae$ is isomorphic to $(eAe)_{\mathrm{opp}}$. Indeed, an $A$-linear map $\varphi:Ae\to Ae$ is determined by $a=\varphi(e)\in eAe$, because for $x\in Ae$ one has $x=xe$ and

$$ \varphi(x)=\varphi(xe)=x\varphi(e)=xa, $$

and conversely every $a\in eAe$ defines the $A$-linear map $\rho_a(x)=xa$. The assignment $\varphi\mapsto a$ is a ring isomorphism onto $(eAe)_{\mathrm{opp}}$, since $\rho_a\rho_b=\rho_{ba}$. If $Ae$ is simple, Schur's lemma says that its endomorphism ring is a division ring; hence $(eAe)_{\mathrm{opp}}$ and therefore $eAe$ are division rings. The right-hand statement is the same computation for the right module $eA$, whose endomorphism ring is $eAe$.

Remark (the converse fails). The converse is not a theorem of the definitions. In $T_2(k)$, the idempotent $E_{22}$ has corner $kE_{22}$, a division ring, but $AE_{22}=kE_{12}\oplus kE_{22}$ contains the proper nonzero left ideal $kE_{12}$, so it is not minimal. Symmetrically, $E_{11}$ has a division corner and a nonminimal right ideal. The converse requires semisimplicity, and it is proved in §The Semisimple Case. The failure is not an accident of the triangular algebra: it is the same phenomenon that makes a local corner not a division ring, read on the one-sided ideal instead of on the corner.

Corollary. If either $Ae$ is a minimal left ideal or $eA$ is a minimal right ideal, then $e$ is primitive.

Proof. The corner $eAe$ is a division ring, hence has no idempotents other than $0,e$; the corner criterion then gives primitivity.

The simple module

When $Ae$ is minimal, it is a simple left $A$-module, and its endomorphism ring is the division ring $(eAe)_{\mathrm{opp}}$. The element $e$ is a cyclic vector: $Ae$ is generated by $e$, and the evaluation map

$$ A\longrightarrow Ae,\qquad a\longmapsto ae, $$

is $A$-linear and surjective, with kernel the left ideal $\{a:ae=0\}$. The simple module is therefore the quotient of $A$ by the left ideal of the elements annihilating $e$, and the idempotent $e$ is a lift of the identity of the quotient. In the matrix algebra, $Ae$ is the column ideal and $e$ is the first basis vector; in the split algebra, $Ae_i$ is the coordinate line and $e_i$ is its unit.

The Semisimple Case

The equivalence

Theorem. Let $A$ be semisimple and let $e\neq0$ be an idempotent. The following are equivalent:

(a) $e$ is primitive;

(b) $Ae$ is a minimal left ideal;

(c) $eAe$ is a division ring;

(d) $eA$ is a minimal right ideal;

(e) $Ae$ is a simple left $A$-module.

Proof. The implications (b)$\Rightarrow$(c) and (d)$\Rightarrow$(c) are the necessity proposition, valid in every algebra; (c)$\Rightarrow$(a) is the corollary of the same proposition, since a division ring has no nontrivial idempotents. It remains to prove (a)$\Rightarrow$(b) and (a)$\Rightarrow$(d) in the semisimple case. Suppose $e$ is primitive and $Ae$ is not minimal. Then $Ae$ contains a nonzero proper left ideal $I$. Since $A$ is semisimple, $Ae$ is semisimple, so $I$ is a direct summand: $Ae=I\oplus J$ with $J\neq0$. Then $e=e_I+e_J$ is a sum of two nonzero orthogonal idempotents, contradicting primitivity. Hence (a)$\Rightarrow$(b). The same argument with the right regular module gives (a)$\Rightarrow$(d). Finally (b)$\Leftrightarrow$(e) is the definition of a simple module, and (b)$\Rightarrow$(a) is the argument that a direct sum decomposition of $Ae$ would split $e$.

Corollary. In a semisimple algebra, the primitive idempotents are exactly the idempotents $e$ for which $Ae$ is a minimal left ideal, and equivalently for which $eA$ is a minimal right ideal; the corner $eAe$ is then a division ring. The three descriptions of a primitive idempotent — no splitting, minimal one-sided ideal, division corner — are equivalent.

Remark. The theorem is the idempotent form of the Wedderburn–Artin theory. In a semisimple algebra $A\cong\prod_{i=1}^r M_{n_i}(D_i)$, the primitive idempotents are the rank-one idempotents inside one matrix factor, the left ideal $Ae$ is the defining module $D_i^{n_i}$ of that factor, and the corner $eAe$ is $D_i$. Two primitive idempotents give isomorphic simple modules exactly when they lie in the same factor; equivalently, when they are Murray–von Neumann equivalent, $e=xy$ and $f=yx$ for some $x,y\in A$. The central idempotents $(1,\dots,1,0,\dots,0)$ are primitive only when the corresponding factor is a division algebra, that is when $n_i=1$.

The index of the simple modules

The primitive idempotents are the indices of the simple left modules. In the semisimple algebra $A\cong\prod_i M_{n_i}(D_i)$, the simple left modules are $S_i=D_i^{n_i}$, one for each factor, and

$$ \boxed{\;Ae\cong S_i,\qquad eAe\cong D_i,\qquad \operatorname{End}_A(Ae)\cong D_{\mathrm{opp},i}\;} $$

for a primitive idempotent $e$ in the $i$-th factor. The right modules are the duals, with $eA$ a minimal right ideal and $eAe$ acting on the other side. The finite-dimensional algebra has finitely many simple modules up to isomorphism, and the primitive idempotents of a complete orthogonal family $e_1,\dots,e_n$ with $\sum_i e_i=1$ give a decomposition

$$ \boxed{\;A=\bigoplus_{i=1}^{n} Ae_i\;} $$

of the regular module into indecomposable projective summands. In a semisimple algebra these summands are simple, and the family is a complete set of primitive idempotents up to equivalence; in a general finite-dimensional algebra the $Ae_i$ are the indecomposable projective modules, and the simple modules are their heads $Ae_i/\operatorname{rad}(Ae_i)$.

Without Semisimplicity

The upper triangular algebra separates the two sides, and the dual numbers separate primitivity from the division corner. Together they show that none of the converses in the semisimple theorem survives on its own.

The two sides separate

Example (the upper triangular algebra). Let $A=T_2(k)$ and let $e=E_{22}$. The corner is $eAe=kE_{22}$, a division ring, so $e$ is primitive. The right ideal

$$ eA=E_{22}T_2(k)=kE_{22} $$

is minimal, but the left ideal

$$ Ae=T_2(k)E_{22}=kE_{12}\oplus kE_{22} $$

contains the nonzero proper left ideal $kE_{12}$, so it is not minimal. The idempotent $E_{11}$ is the mirror image: $AE_{11}=kE_{11}$ is minimal and $E_{11}A=kE_{11}\oplus kE_{12}$ is not. Thus

$$ \boxed{\;eAe\ \text{division}\ \not\Rightarrow\ Ae\ \text{minimal}, \qquad eAe\ \text{division}\ \not\Rightarrow\ eA\ \text{minimal}\;} $$

in general, and the failure occurs on the side opposite to the one that generated the corner. The two primitive idempotents $E_{11}$ and $E_{22}$ are primitive on one side each, and the identity $1=E_{11}+E_{22}$ is primitive on neither.

The computation also shows why the proof of the converse cannot be repaired by a formal argument. In $Ae=kE_{12}\oplus kE_{22}$, the element $E_{12}$ is nonzero and satisfies $E_{12}e=E_{12}$, so it lies in the corner on the right, but $eE_{12}=0$, so it is not in $eAe$; the corner sees only the elements that survive multiplication by $e$ on both sides. The left ideal $kE_{12}$ is a submodule that the corner does not detect, and its existence is exactly the failure of minimality. In a semisimple algebra there is no such submodule, because every submodule is a direct summand and would split $e$.

Primitivity without a division corner

Example (the dual numbers). Let $A=k[\varepsilon]/(\varepsilon^2)$ with $k$ of characteristic not two. The only idempotents are $0$ and $1$, so $e=1$ is primitive and $eAe=A$ has no nontrivial idempotent. The left ideal $Ae=A$ is not minimal, because $(\varepsilon)$ is a nonzero proper left ideal, and $eAe=A$ is not a division ring. Thus

$$ e\ \text{primitive}\ \not\Rightarrow\ eAe\ \text{division} $$

without semisimplicity. The corner $eAe=A$ is a local ring: it has a unique maximal left ideal $(\varepsilon)$ and no nontrivial idempotent. The general statement is that, in a finite-dimensional algebra, $e$ is primitive exactly when $eAe$ is local, and then $Ae$ is an indecomposable projective module, not necessarily a simple one. The simple module is the quotient $Ae/\operatorname{rad}(Ae)$; in the dual numbers it is the one-dimensional quotient $A/(\varepsilon)$.

The logical map

The implications among the five conditions are summarized as follows, with the general one-directional implications:

Hypothesis Conclusion
$Ae$ minimal $eAe$ division
$eA$ minimal $eAe$ division
$eAe$ division $e$ primitive

In a semisimple algebra all five conditions are equivalent, and each implication reverses. Without semisimplicity none of the reverse implications holds: the upper triangular algebra is the counterexample to the converses of the two minimal-ideal implications, and the dual numbers are the counterexample to the converse of the division-corner implication. The correct general notion is the local idempotent, for which $eAe$ is local; the primitive idempotents of the semisimple theory are the local idempotents whose corner is a division ring.

Summary

A nonzero idempotent $e$ is primitive when it is not a sum of two nonzero orthogonal idempotents; equivalently, the corner $eAe$ has no idempotents other than $0,e$, and equivalently the left module $Ae$ (and the right module $eA$) is indecomposable. A minimal one-sided ideal generated by $e$ forces $eAe$ to be a division ring: this is the division ring criterion, proved through the endomorphism ring of the module and Schur's lemma. The converse fails without semisimplicity, and it fails independently on the two sides: in the upper triangular algebra $T_2(k)$, $E_{22}$ has a division corner and a minimal right ideal but a nonminimal left ideal, while $E_{11}$ is the mirror image. The dual numbers show the other failure: $e=1$ is primitive, but $eAe=A$ is local and not a division ring. In a semisimple algebra the five conditions — primitivity, minimal left ideal, minimal right ideal, division corner, simple module — are equivalent, and the primitive idempotents index the simple modules: in $A\cong\prod_i M_{n_i}(D_i)$ they are the rank-one idempotents in one factor, $Ae$ is the simple module $D_i^{n_i}$, and $eAe\cong D_i$. The matrix algebras, the split algebra, the upper triangular algebra and the dual numbers are the worked examples. No form, involution or topology is used.

Summary of Notation

Symbol Meaning
$e,f$ idempotent and its complement $f=1-e$
$Ae$ the left ideal generated by $e$
$eA$ the right ideal generated by $e$
$eAe$ the corner with identity $e$
primitive no decomposition into two nonzero orthogonal idempotents
minimal left ideal nonzero left ideal with no nonzero proper left subideal
$Ae$ simple $Ae$ is a simple left $A$-module
$eAe$ division every nonzero element of the corner is invertible in the corner
semisimple the left regular module is a direct sum of simple modules
local idempotent $eAe$ is a local ring

Further Reading

  • Richard S. Pierce, Associative Algebras (Springer, 1982), for primitive idempotents, minimal left ideals and the Peirce decomposition.
  • Nathan Jacobson, Basic Algebra II (Freeman, 1989), for the Wedderburn–Artin theory, simple modules and the division ring criterion.
  • T. Y. Lam, A First Course in Noncommutative Rings (Springer, 2001), for primitive and local idempotents, the dual numbers and the upper triangular examples.
  • Frank W. Anderson and Kent R. Fuller, Rings and Categories of Modules (Springer, 1992), for the module-theoretic form of the division ring criterion and the projective covers.