← All writingMathematics blog

A Proof of Jennings’ Theorem

A proof of Jennings’ theorem using restricted Lie algebras, with formulas for radical layers and Loewy length and an example of order 27.

Introduction

The goal of this post is to provide a complete proof of a theorem of Jennings in representation theory and discuss some applications thereof. Jennings' theorem provides quite a bit of information on the structure of group algebras of finite pp-groups over a field of characteristic pp.

Most importantly, we will get explicit formulas for the Loewy length of the group algebra and the dimensions of the radical and socle layers, and we will be able to deduce that the radical series coincides with the socle series, in reverse order, in this case.

Computing the group ring kGkG for char⁡k=p>0\chr k=p>0 and GG a finite pp-group is of interest for various reasons. It is a situation where we can deduce more information than the general case; for example, we know that Rad⁡(kG)=IG\Rad(kG)=IG, where IGIG is the augmentation ideal. If GG is cyclic, then we have a complete description of its indecomposable modules, as well as its simple and indecomposable projective modules.

Also, in many cases even if GG is not a pp-group, its structure separates a normal Sylow pp-subgroup from a complement of order prime to pp. This is the case when G=H⋊KG=H\rtimes K, with HH a Sylow pp-subgroup: the group algebra kKkK is semisimple, while kHkH is the modular part of the problem. A finite group with a normal Sylow pp-subgroup admits such a complement.

Background

Having motivated Jennings' theorem, let us give a brief historical overview of its proof. The result was first published by S. A. Jennings in 1941 in his original paper, as a special result of his thesis. Jennings defined the Jennings series of a group and proved the theorem in its original form. Jennings' proof was somewhat complicated. It is notable that in a footnote to the original paper, Jennings pointed out that Zassenhaus had around the same time proved that the Jennings series also arises out of the Lie algebra associated to the group ring.

D. G. Quillen, in 1968, expanded on that idea and gave a much shorter proof using the natural Lie algebra structure which gives rise to the Jennings series, in On the associated graded ring of a group ring. His argument is the starting point for the one we use here, but we follow more closely the presentation of Benson in Representations and Cohomology I. Quillen's proof uses the notion of a pp-restricted Lie algebra, which can be associated to a certain central series of the group.

The key idea is that the group commutator provides a Lie algebra commutator, and there is a natural ppth power operation in the algebra. Restricted Lie algebras satisfy a Poincaré–Birkhoff–Witt (PBW) theorem. For Jennings' theorem, the spanning part of that theorem and a dimension comparison will suffice: we will use them to show that the Jennings series coincides with the series of dimension subgroups.

The post will proceed as follows. In Jennings' theorem we will state and prove Jennings' theorem. Deviating from the presentation in Benson, we will begin with a more general discussion of restricted Lie algebras because they are of interest in their own right, especially in the theory of group schemes in geometry. Then in Examples and applications we prove the important corollary about the socle series, and conclude with an example.

Jennings' theorem

Restricted Lie algebras

For this section, fix a field kk of characteristic p>0p>0. If g\g is a Lie algebra, then for x∈gx\in\g we use the standard notation ad⁡(x) ⁣:g→g\ad(x)\colon\g\to\g for the map y↦[x,y]y\mapsto[x,y]. Our main object of study in this section is a pp-restricted Lie algebra, or just restricted Lie algebra for short. The main reference for this subsection is Strade and Farnsteiner, Modular Lie Algebras and Their Representations, Chapter II, §§1–5.

Definition 1. A pp-restricted Lie algebra over kk is a Lie algebra g\g over kk equipped with a map

g→g,x↦x[p],\g\to\g,\qquad x\mapsto x^{[p]},

called the pp-mapping, satisfying the following properties:

  1. ad⁡(x[p])=ad⁡(x)p\ad(x^{[p]})=\ad(x)^p for all x∈gx\in\g.
  2. (αx)[p]=αpx[p](\alpha x)^{[p]}=\alpha^p x^{[p]} for all x∈gx\in\g and α∈k\alpha\in k.
  3. (x+y)[p]=x[p]+y[p]+∑j=1p−1sj(x,y)(x+y)^{[p]}=x^{[p]}+y^{[p]}+\sum_{j=1}^{p-1}s_j(x,y) for all x,y∈gx,y\in\g.

Here the terms sj(x,y)s_j(x,y) are uniquely determined by the relation

ad⁡(tx+y)p−1(x)=∑j=1p−1j sj(x,y)tj−1,\ad(tx+y)^{p-1}(x)=\sum_{j=1}^{p-1}j\,s_j(x,y)t^{j-1},

where powers of ad⁡\ad are interpreted as composition, and tt is a formal variable. A morphism of restricted Lie algebras is a Lie algebra homomorphism which preserves the map x↦x[p]x\mapsto x^{[p]}.

For example, if p=2p=2 then s1(x,y)=[y,x]s_1(x,y)=[y,x], and if p=3p=3 then the relation gives

t[x,[y,x]]+[y,[y,x]]=2t s2(x,y)+s1(x,y).t[x,[y,x]]+[y,[y,x]]=2t\,s_2(x,y)+s_1(x,y).

Like this one can see how to derive a general formula, but we will not need one.

Example 2. Recall that if AA is an associative algebra, then AA is naturally a Lie algebra with [x,y]=xy−yx[x,y]=xy-yx. The map x↦x[p]=xpx\mapsto x^{[p]}=x^p makes AA into a pp-restricted Lie algebra. For the first property, use the relation

ad⁡(x)m(y)=∑i=0m(−1)m−i(mi)xiyxm−i\ad(x)^m(y)=\sum_{i=0}^m(-1)^{m-i}\binom{m}{i}x^iyx^{m-i}

for any mm. When m=pm=p, the intermediate binomial coefficients vanish, giving ad⁡(x)p(y)=[xp,y]\ad(x)^p(y)=[x^p,y]. The second property is immediate.

For the third, differentiate formally in A[t]A[t]. Since (p−1i)=(−1)i\binom{p-1}{i}=(-1)^i in kk, the same formula gives

ddt(tx+y)p=∑i=0p−1(tx+y)ix(tx+y)p−1−i=ad⁡(tx+y)p−1(x).\begin{aligned} \frac{d}{dt}(tx+y)^p &=\sum_{i=0}^{p-1}(tx+y)^i x(tx+y)^{p-1-i} \\ &=\ad(tx+y)^{p-1}(x). \end{aligned}

The polynomial (tx+y)p−tpxp−yp(tx+y)^p-t^px^p-y^p has zero constant term and degree at most p−1p-1. Comparing derivatives with the defining formula for sjs_j therefore gives

(tx+y)p−tpxp−yp=∑j=1p−1sj(x,y)tj.(tx+y)^p-t^px^p-y^p=\sum_{j=1}^{p-1}s_j(x,y)t^j.

Setting t=1t=1 proves the third property. In particular, gln(k)\mathfrak{gl}_n(k) is a pp-restricted Lie algebra.

Instead of relying on an explicit formula for (x+y)[p](x+y)^{[p]}, we need the following more important property: each sj(x,y)s_j(x,y) is a linear combination of iterated brackets of length pp in xx and yy. This follows directly by expanding ad⁡(tx+y)p−1(x)\ad(tx+y)^{p-1}(x). In particular, the correction terms stay in the Lie subalgebra generated by xx and yy.

We move on to the restricted universal enveloping algebra. If g\g is a restricted Lie algebra, the restricted enveloping algebra u(g)u(\g) should have the universal property that, for AA an associative kk-algebra, restricted Lie algebra homomorphisms g→A\g\to A correspond to algebra homomorphisms u(g)→Au(\g)\to A.

It is easy to see how to construct this: take the standard enveloping algebra U(g)U(\g), let II be the two-sided ideal generated by elements x[p]−xpx^{[p]}-x^p for all x∈gx\in\g, and set

u(g)≔U(g)/I.u(\g)\coloneqq U(\g)/I.

Then the universal property of U(g)U(\g) immediately implies the desired one for u(g)u(\g). The main result of this section is the following restricted PBW theorem; see Strade and Farnsteiner, Chapter II, Theorem 5.1.

Theorem 3. Let g\g be a restricted finite-dimensional Lie algebra, and suppose e1,…,ene_1,\dots,e_n is a basis for g\g.

  1. The collection of elements

    e1j1⋯enjn∈u(g),0≤ji<p,e_1^{j_1}\cdots e_n^{j_n}\in u(\g),\qquad 0\leq j_i<p,
    (1)

    is a basis for u(g)u(\g).

  2. In particular, the canonical map g→u(g)\g\to u(\g) is injective, and dim⁡ku(g)=pn\dim_k u(\g)=p^n.

Remark 4. For the proof of Jennings' theorem, we will not need the full strength of this result. We need only the easier direction that the collection Equation (1) spans u(g)u(\g), so that dim⁡ku(g)≤pn\dim_k u(\g)\leq p^n. In fact, the same spanning argument works directly for an algebra given by the commutator and ppth-power relations, before we have verified all the restricted Lie algebra axioms.

Proof. Observe first that (2) is an immediate consequence of (1): the basis for g\g is a subset of the basis for u(g)u(\g) by taking all jij_i to be 00, except one to be 11. Hence g↪u(g)\g\mono u(\g) is injective. The dimension statement is clear by counting the number of elements in Equation (1).

For (1), it is easy to see that these elements span u(g)u(\g). Indeed, the statement would be trivial if we could reorder the eie_i and if the powers jij_i could be arbitrarily large. The former part is easy to handle because we can use commutators to reorder the basis, with correction terms of smaller word length. To address the powers, note that if j≥pj\geq p then

xj=xpxj−p=x[p]xj−pin u(g),x^j=x^px^{j-p}=x^{[p]}x^{j-p}\qquad\text{in }u(\g),

so we can again reduce word length. Induction on word length, and then on the number of out-of-order pairs, expresses every word using ordered powers strictly less than pp. This is the spanning argument we will use for Jennings' theorem.

For independence, we use the classical PBW theorem in U(g)U(\g). Put

ci=eip−ei[p]∈U(g).c_i=e_i^p-e_i^{[p]}\in U(\g).

These elements are central, since for y∈gy\in\g,

[eip−ei[p],y]=ad⁡(ei)p(y)−ad⁡(ei[p])(y)=0.[e_i^p-e_i^{[p]},y]=\ad(e_i)^p(y)-\ad(e_i^{[p]})(y)=0.

Moreover, the map x↦xp−x[p]x\mapsto x^p-x^{[p]} from g\g to U(g)U(\g) is additive and satisfies (αx)p−(αx)[p]=αp(xp−x[p])(\alpha x)^p-(\alpha x)^{[p]}=\alpha^p(x^p-x^{[p]}): the correction terms in the two pp-power formulas cancel. Thus II is generated by c1,…,cnc_1,\dots,c_n.

In the usual PBW filtration, the elements

c1a1⋯cnane1b1⋯enbn,ai≥0,0≤bi<p,c_1^{a_1}\cdots c_n^{a_n}e_1^{b_1}\cdots e_n^{b_n}, \qquad a_i\geq0,\quad0\leq b_i<p,

have leading monomials e1pa1+b1⋯enpan+bne_1^{pa_1+b_1}\cdots e_n^{pa_n+b_n}. Every PBW monomial occurs exactly once as a leading monomial, by division of its exponents by pp. Consequently these elements form a basis of U(g)U(\g): spanning and independence both follow by induction on the filtration degree.

Since the cic_i are central, multiplication by a cic_i simply increases its corresponding exponent in this basis. The ideal II is therefore spanned by the displayed basis elements for which at least one aia_i is positive. The remaining elements, with every ai=0a_i=0, descend to a basis of U(g)/IU(\g)/I. These are exactly Equation (1). □

For our purposes, this is all we need about restricted Lie algebras, but much more can be said about them. We refer the reader to Chapter II of Strade and Farnsteiner. These objects play a fundamental role in the theory of algebraic groups as well.

The proof of Jennings' theorem

In this section we state and prove Jennings' theorem. Fix a field kk of characteristic p>0p>0, and let GG be a finite pp-group, say of order pnp^n. Recall that in this case we have Rad⁡(kG)=IG\Rad(kG)=IG, and so Rad⁡i(kG)=(IG)i\Rad^i(kG)=(IG)^i; see Webb, Proposition 6.3.3. We consider the following:

Definition 5. The dimension subgroups of GG are

Fr(G)≔{g∈G:g−1∈Rad⁡r(kG)}F_r(G)\coloneqq\{g\in G:g-1\in\Rad^r(kG)\}

for r≥1r\geq1.

Recall that IGIG is spanned by the elements g−1g-1 for g≠1g\neq1, so F1(G)=GF_1(G)=G. These are indeed subgroups because, for g,h∈Gg,h\in G, we have

gh−1=(g−1)+(h−1)+(g−1)(h−1),gh-1=(g-1)+(h-1)+(g-1)(h-1),
(2)

and g−1−1=−g−1(g−1)g^{-1}-1=-g^{-1}(g-1). Moreover, Fr(G)⊴GF_r(G)\trianglelefteq G, because for hh arbitrary we have

h(g−1)h−1=hgh−1−1,h(g-1)h^{-1}=hgh^{-1}-1,

which is in Rad⁡r(kG)\Rad^r(kG) whenever g−1g-1 is, since this is a two-sided ideal. The fact that the Rad⁡r(kG)\Rad^r(kG) form a decreasing sequence implies that {Fr(G)}\{F_r(G)\} is also a decreasing sequence, and if r≥ℓr\geq\ell, the Loewy length, then Fr(G)={1}F_r(G)=\{1\}. So the dimension subgroups form a decreasing filtration of GG.

We also have (g−1)p=gp−1(g-1)^p=g^p-1, so the ppth power of an element from Fr(G)F_r(G) is in Fpr(G)F_{pr}(G). In addition, note that

[g,h]−1=((g−1)(h−1)−(h−1)(g−1))g−1h−1,[g,h]-1=\big((g-1)(h-1)-(h-1)(g-1)\big)g^{-1}h^{-1},
(3)

where we are using the convention [g,h]=ghg−1h−1[g,h]=ghg^{-1}h^{-1}. Again because the radical powers are two-sided ideals, this implies that [Fr(G),Fs(G)]⊂Fr+s(G)[F_r(G),F_s(G)]\subset F_{r+s}(G).

Finally, recall that the Frattini subgroup Φ(G)\Phi(G) is the intersection of the maximal subgroups of GG, and for a finite pp-group we know that Φ(G)=Gp[G,G]\Phi(G)=G^p[G,G]. We claim that

Φ(G)=F2(G).\Phi(G)=F_2(G).

The containments just proved give Gp[G,G]⊂F2(G)G^p[G,G]\subset F_2(G). Conversely, if g∉Φ(G)g\notin\Phi(G), choose a maximal subgroup HH not containing gg. A maximal subgroup of a finite pp-group is normal of index pp, so the quotient map G→G/H≅CpG\to G/H\cong C_p induces a map of group algebras carrying IGIG into I(Cp)I(C_p) and (IG)2(IG)^2 into I(Cp)2I(C_p)^2.

If cc generates CpC_p, then kCp≅k[t]/(tp)kC_p\cong k[t]/(t^p) with t=c−1t=c-1. The image of gg is cjc^j for some 1≤j<p1\leq j<p, and

cj−1≡j(c−1)≢0(modI(Cp)2).c^j-1\equiv j(c-1)\not\equiv0\pmod{I(C_p)^2}.

Thus g−1∉(IG)2g-1\notin(IG)^2, proving F2(G)⊂Φ(G)F_2(G)\subset\Phi(G). The following is what we have proved so far.

Proposition 6. Let GG be a finite pp-group and kk a field of characteristic pp.

  1. The dimension subgroups of GG form a decreasing filtration

    G=F1(G)⊃F2(G)⊃F3(G)⊃⋯G=F_1(G)\supset F_2(G)\supset F_3(G)\supset\cdots

    of subgroups which is trivial from the Loewy length of kGkG onward.

  2. Each Fr(G)F_r(G) is a normal subgroup of GG, and F2(G)=Φ(G)F_2(G)=\Phi(G).

  3. We have the containments

    Fr(G)p⊂Frp(G),[Fr(G),Fs(G)]⊂Fr+s(G)F_r(G)^p\subset F_{rp}(G),\qquad[F_r(G),F_s(G)]\subset F_{r+s}(G)

    for all r,s≥1r,s\geq1.

The properties of the dimension subgroups motivate what is called the Jennings series of GG: it is the fastest descending series

G=Γ1(G)⊃Γ2(G)⊃Γ3(G)⊃⋯G=\Gamma_1(G)\supset\Gamma_2(G)\supset\Gamma_3(G)\supset\cdots

such that

Γr(G)p⊂Γrp(G),[Γr(G),Γs(G)]⊂Γr+s(G)\Gamma_r(G)^p\subset\Gamma_{rp}(G),\qquad [\Gamma_r(G),\Gamma_s(G)]\subset\Gamma_{r+s}(G)

for all r,s≥1r,s\geq1. Here HpH^p denotes the subgroup generated by the ppth powers of elements of HH.

To construct the fastest such series without a recursive ambiguity, take the termwise intersection of all decreasing series satisfying these properties and starting with GG. The power and commutator containments are preserved by intersections, so the resulting series has the required properties and is contained termwise in every other such series. In particular,

Γr(G)⊂Fr(G)\Gamma_r(G)\subset F_r(G)

for all rr, so it is trivial from the Loewy length onward. Moreover, [G,Γr(G)]⊂Γr+1(G)[G,\Gamma_r(G)]\subset\Gamma_{r+1}(G), so it is a central series, and all its terms are normal.

The defining properties say that Γr(G)\Gamma_r(G) must contain the subgroups

[Γs(G),Γt(G)](s+t≥r),Γs(G)p(ps≥r).[\Gamma_s(G),\Gamma_t(G)]\quad(s+t\geq r),\qquad \Gamma_s(G)^p\quad(ps\geq r).

These are useful for computing it. For instance, Γ2(G)\Gamma_2(G) contains [G,G][G,G] and GpG^p, so it contains Φ(G)\Phi(G). Since it is also contained in F2(G)=Φ(G)F_2(G)=\Phi(G), the dimension series and Jennings series agree for r=1,2r=1,2.

Since the series is central, successive quotients are abelian. Moreover,

Γr(G)p⊂Γpr(G)⊂Γr+1(G),\Gamma_r(G)^p\subset\Gamma_{pr}(G)\subset\Gamma_{r+1}(G),

because p≥2p\geq2. Thus the quotients are elementary abelian pp-groups, and as Z\ZZ-modules they are annihilated by pZp\ZZ, hence they are naturally Fp\F_p-vector spaces. In what follows we will set

nr≔dim⁡Fp(Γr(G)/Γr+1(G)),r≥1,n_r\coloneqq\dim_{\F_p}\big(\Gamma_r(G)/\Gamma_{r+1}(G)\big),\qquad r\geq1,

and so pnr=[Γr(G):Γr+1(G)]p^{n_r}=[\Gamma_r(G):\Gamma_{r+1}(G)]. We can also tensor with kk to make this an nrn_r-dimensional vector space over kk. Doing this, define the Jennings algebra of GG over kk to be the graded vector space

Jen∗(G)≔⨁r≥1k⊗Fp(Γr(G)/Γr+1(G)).\Jen_*(G)\coloneqq\bigoplus_{r\geq1}k\underset{\F_p}{\otimes}\big(\Gamma_r(G)/\Gamma_{r+1}(G)\big).

The plan is to apply the machinery from the previous section to this object. We will see that it is a graded restricted Lie algebra. For the Lie bracket, we use the Hall–Witt identity, with the convention zy=yzy−1z^y=yzy^{-1} for conjugation.

Lemma 7 (Hall–Witt). Let HH be a group. With [x,y]=xyx−1y−1[x,y]=xyx^{-1}y^{-1} and zy=yzy−1z^y=yzy^{-1}, we have

[[x,y],zy][[y,z],xz][[z,x],yx]=1\big[[x,y],z^y\big]\big[[y,z],x^z\big]\big[[z,x],y^x\big]=1

for all x,y,z∈Hx,y,z\in H.

Proof. Expand each of the terms using the definitions. Their product is

(xyx−1zx y−1x−1yz−1y−1)⋅(yzy−1xy z−1y−1zx−1z−1)⋅(zxz−1yz x−1z−1xy−1x−1),\begin{aligned} &(xyx^{-1}zx\,y^{-1}x^{-1}yz^{-1}y^{-1}) \\ &\quad\cdot(yzy^{-1}xy\,z^{-1}y^{-1}zx^{-1}z^{-1}) \\ &\quad\cdot(zxz^{-1}yz\,x^{-1}z^{-1}xy^{-1}x^{-1}), \end{aligned}

and successive cancellation gives the identity. □

Now the containments

[Γr+1(G),Γs(G)], [Γr(G),Γs+1(G)]⊂Γr+s+1(G)[\Gamma_{r+1}(G),\Gamma_s(G)],\ [\Gamma_r(G),\Gamma_{s+1}(G)] \subset\Gamma_{r+s+1}(G)

show that the group commutator induces a well-defined bracket from degrees r,sr,s to degree r+sr+s. Extend it kk-bilinearly to Jen∗(G)\Jen_*(G). The identities

[ab,c]=a[b,c]a−1[a,c],[a,bc]=[a,b]b[a,c]b−1[ab,c]=a[b,c]a^{-1}[a,c],\qquad [a,bc]=[a,b]b[a,c]b^{-1}

show additivity on the group quotients, since the conjugation corrections lie one step deeper in the filtration. Alternation follows from [x,x]=1[x,x]=1, and antisymmetry follows by interchanging the two group elements. The Jacobi identity follows from Hall–Witt: all three factors lie in Γr+s+t(G)\Gamma_{r+s+t}(G), and replacing a conjugated argument by the original argument changes its class only in Γr+s+t+1(G)\Gamma_{r+s+t+1}(G). Taking classes there gives the sum of the three iterated brackets equal to zero. Thus Jen∗(G)\Jen_*(G) is a graded Lie algebra over kk.

The expected pp-mapping sends the class of a group element in degree rr to the class of its ppth power in degree prpr. Rather than assume the identities for this operation, we will construct the relevant associative algebra directly and verify the restricted structure through a dimension comparison. This uses exactly the spanning argument in Remark 4.

Choose an Fp\F_p-basis of each Γr(G)/Γr+1(G)\Gamma_r(G)/\Gamma_{r+1}(G) and group representatives for these basis elements. After extending scalars, order the resulting homogeneous basis of Jen∗(G)\Jen_*(G) as

x1,…,xn1,…,xn1+n2,…,xm,m=∑rnr.x_1,\dots,x_{n_1},\dots,x_{n_1+n_2},\dots,x_m, \qquad m=\sum_r n_r.

For each chosen xix_i of degree rr, its representative has ppth power in Γpr(G)\Gamma_{pr}(G). Write xi[p]x_i^{[p]} for the corresponding vector in the degree-prpr component of Jen∗(G)\Jen_*(G), using these fixed representatives for now. Form the associative algebra

Tk(Jen∗(G))(xixj−xjxi−[xi,xj], xip−xi[p])i,j.\frac{T_k(\Jen_*(G))} {\big(x_ix_j-x_jx_i-[x_i,x_j],\ x_i^p-x_i^{[p]}\big)_{i,j}}.
(4)

The relations preserve the grading coming from the degrees of the xix_i. Reordering words with the commutator relations and reducing ppth powers shows that the ordered monomials

x1j1⋯xmjm,0≤ji<p,x_1^{j_1}\cdots x_m^{j_m},\qquad0\leq j_i<p,
(5)

span this algebra. Indeed, each correction replaces a product by an element of the vector space Jen∗(G)\Jen_*(G) and therefore reduces ordinary word length, just as in the spanning proof above. Its dimension is consequently at most pmp^m.

The way we use this construction is by taking the associated graded of kGkG, filtered by the radical:

gr∗kG≔⨁i≥0Rad⁡i(kG)/Rad⁡i+1(kG).\mathrm{gr}_*kG\coloneqq\bigoplus_{i\geq0}\Rad^i(kG)/\Rad^{i+1}(kG).

This has a well-defined multiplication because Rad⁡i(kG)⋅Rad⁡j(kG)⊂Rad⁡i+j(kG)\Rad^i(kG)\cdot\Rad^j(kG)\subset\Rad^{i+j}(kG). Because Γr(G)⊂Fr(G)\Gamma_r(G)\subset F_r(G), the rule g↦g−1g\mapsto g-1 gives maps into Rad⁡r(kG)\Rad^r(kG). It descends to the successive quotients, and Equation (2) shows that the resulting maps are additive. Extending scalars gives a graded kk-linear map

Jen∗(G)⟶gr∗kG.\Jen_*(G)\longrightarrow\mathrm{gr}_*kG.

By Equation (3), this map takes the group bracket to the algebra commutator. The factor g−1h−1g^{-1}h^{-1} in that formula can be replaced by 11 modulo the next radical power. Similarly, (g−1)p=gp−1(g-1)^p=g^p-1 shows that the chosen ppth-power relations hold in the target. Thus we obtain a graded algebra homomorphism ψ\psi from the algebra Equation (4) to gr∗kG\mathrm{gr}_*kG.

Since gr∗kG\mathrm{gr}_*kG is generated by its degree-one component and Γ1(G)=G\Gamma_1(G)=G, the elements g−1g-1 show that ψ\psi is surjective. Notice that

pm=∏rpnr=∏r[Γr(G):Γr+1(G)]=∣G∣.p^m=\prod_rp^{n_r}=\prod_r[\Gamma_r(G):\Gamma_{r+1}(G)]=\abs{G}.

On the other hand,

dim⁡kgr∗kG=∑i≥0(dim⁡kRad⁡i(kG)−dim⁡kRad⁡i+1(kG))=dim⁡kkG=∣G∣.\begin{aligned} \dim_k\mathrm{gr}_*kG &=\sum_{i\geq0}\big(\dim_k\Rad^i(kG)-\dim_k\Rad^{i+1}(kG)\big) \\ &=\dim_k kG=\abs{G}. \end{aligned}

The source of ψ\psi has dimension at most pm=∣G∣p^m=\abs{G}, while ψ\psi surjects onto a vector space of exactly this dimension. Therefore ψ\psi is an isomorphism, and the spanning monomials Equation (5) are a basis. In particular, Jen∗(G)\Jen_*(G) injects into this algebra and hence into gr∗kG\mathrm{gr}_*kG.

This also supplies the restricted structure promised above. The copy of Jen∗(G)\Jen_*(G) in its associative algebra is closed under commutators and under the ppth powers of its basis vectors. The formula in Example 2 then shows that it is closed under the ppth power of every linear combination. It therefore inherits a restricted Lie algebra structure. For any group element g∈Γr(G)g\in\Gamma_r(G), the identity (g−1)p=gp−1(g-1)^p=g^p-1 and the injectivity just proved identify this operation with the class of gpg^p in degree prpr. In particular it is independent of the chosen representative.

Finally, the presentation Equation (4) has the universal property of the restricted enveloping algebra: a restricted Lie homomorphism out of Jen∗(G)\Jen_*(G) satisfies its defining relations, and an algebra homomorphism out of the presentation preserves the inherited bracket and pp-mapping. We have consequently proved a graded isomorphism

ψ ⁣:u(Jen∗(G))→∼gr∗kG.\psi\colon u(\Jen_*(G))\xrightarrow{\sim}\mathrm{gr}_*kG.

We now state our main result.

Theorem 8 (Jennings). Let GG be a finite pp-group and kk a field of characteristic pp.

  1. Fr(G)=Γr(G)F_r(G)=\Gamma_r(G) for all r≥1r\geq1.

  2. The Poincaré series of the radical layers of kGkG has the expression

    ∑i≥0dim⁡k ⁣(Rad⁡i(kG)Rad⁡i+1(kG))ti=∏r≥1(1−tpr1−tr)nr,\sum_{i\geq0}\dim_k\!\left(\frac{\Rad^i(kG)}{\Rad^{i+1}(kG)}\right)t^i =\prod_{r\geq1}\left(\frac{1-t^{pr}}{1-t^r}\right)^{n_r},

    where pnr=[Fr(G):Fr+1(G)]p^{n_r}=[F_r(G):F_{r+1}(G)].

  3. The Loewy length of kGkG is

    ℓ(kG)=1+∑r≥1(p−1)rnr.\ell(kG)=1+\sum_{r\geq1}(p-1)r n_r.

Proof. The main part of the result is (1), which we can now prove from our work above. We use induction, and have already established the cases r=1,2r=1,2. Suppose Fr(G)=Γr(G)F_r(G)=\Gamma_r(G), and take g∈Fr+1(G)g\in F_{r+1}(G). Then g∈Γr(G)g\in\Gamma_r(G), and its class in Γr(G)/Γr+1(G)\Gamma_r(G)/\Gamma_{r+1}(G) maps to zero in Rad⁡r(kG)/Rad⁡r+1(kG)\Rad^r(kG)/\Rad^{r+1}(kG), because g−1∈Rad⁡r+1(kG)g-1\in\Rad^{r+1}(kG).

But the map from this group quotient into Jen∗(G)\Jen_*(G) is injective after extension from Fp\F_p to kk, and the map Jen∗(G)→gr∗kG\Jen_*(G)\to\mathrm{gr}_*kG is injective by the dimension comparison. Thus the class of gg is zero, so g∈Γr+1(G)g\in\Gamma_{r+1}(G). The reverse inclusion was already proved, completing the induction.

For (2), we have shown that the elements

x1j1⋯xmjm,0≤ji<p,x_1^{j_1}\cdots x_m^{j_m},\qquad0\leq j_i<p,

are a basis for u(Jen∗(G))u(\Jen_*(G)). The graded isomorphism ψ\psi identifies them with a homogeneous kk-basis for gr∗kG\mathrm{gr}_*kG. There are nrn_r generators of weight rr, each of which can occur with exponent 0,1,…,p−10,1,\dots,p-1. A monomial has weight equal to the sum of the weights of its factors, so counting these choices gives

∑i≥0dim⁡k(Rad⁡i(kG)/Rad⁡i+1(kG))ti=∏r≥1(1+tr+t2r+⋯+t(p−1)r)nr.\sum_{i\geq0}\dim_k\big(\Rad^i(kG)/\Rad^{i+1}(kG)\big)t^i =\prod_{r\geq1}(1+t^r+t^{2r}+\cdots+t^{(p-1)r})^{n_r}.

The expression in (2) is just the geometric-series representation of the factors. Notice that the degree-rr component can contain products of lower-weight generators, as well as the nrn_r generators themselves.

For (3), the Loewy length is one more than the largest degree with a nonzero radical layer. Indeed, if two consecutive radical powers are equal, nilpotence forces that power to be zero. Thus the Loewy length is one more than the degree of the Poincaré polynomial. Each factor has top degree (p−1)rnr(p-1)r n_r, so the required degree is ∑r(p−1)rnr\sum_r(p-1)r n_r. □

Examples and applications

We will conclude by giving the primary application of Jennings' theorem, and then a small example.

Theorem 9. Let kk be a field of characteristic p>0p>0 and GG a finite pp-group. If ℓ\ell is the Loewy length of kGkG, then

Soc⁡n(kG)=Rad⁡ℓ−n(kG),0≤n≤ℓ.\Soc^n(kG)=\Rad^{\ell-n}(kG),\qquad0\leq n\leq\ell.

Thus the radical series and the socle series of kGkG are the same filtration, in reverse order.

Proof. The Poincaré polynomial of Theorem 8 is symmetric: each factor 1+tr+⋯+t(p−1)r1+t^r+\cdots+t^{(p-1)r} is symmetric, so

dim⁡k(Rad⁡i(kG)/Rad⁡i+1(kG))=dim⁡k(Rad⁡ℓ−i−1(kG)/Rad⁡ℓ−i(kG)).\dim_k\big(\Rad^i(kG)/\Rad^{i+1}(kG)\big) =\dim_k\big(\Rad^{\ell-i-1}(kG)/\Rad^{\ell-i}(kG)\big).

The duality between radical and socle layers gives

Soc⁡n+1((kG)∗)/Soc⁡n((kG)∗)≅(Rad⁡n(kG)/Rad⁡n+1(kG))∗;\Soc^{n+1}((kG)^*)/\Soc^n((kG)^*) \cong\big(\Rad^n(kG)/\Rad^{n+1}(kG)\big)^*;

see Webb, Chapter 6, Exercise (7). The regular group module is self-dual: the pairing on the group basis given by (g,h)↦δg,h(g,h)\mapsto\delta_{g,h} is nondegenerate and invariant under simultaneous left multiplication. Thus, as in Webb, Proposition 8.5.2,

dim⁡k(Soc⁡n+1(kG)/Soc⁡n(kG))=dim⁡k(Rad⁡n(kG)/Rad⁡n+1(kG)).\dim_k\big(\Soc^{n+1}(kG)/\Soc^n(kG)\big) =\dim_k\big(\Rad^n(kG)/\Rad^{n+1}(kG)\big).

Finally, remember that

Soc⁡n(kG)⊃Rad⁡ℓ−n(kG).\Soc^n(kG)\supset\Rad^{\ell-n}(kG).

This is because Rad⁡n(kG)⋅Rad⁡ℓ−n(kG)⊂Rad⁡ℓ(kG)=0\Rad^n(kG)\cdot\Rad^{\ell-n}(kG)\subset\Rad^\ell(kG)=0, and the nnth socle is the set of elements annihilated by the nnth radical power. Summing the layer dimensions and using the symmetry above gives

dim⁡kSoc⁡n(kG)=∑i=0n−1dim⁡k(Rad⁡i(kG)/Rad⁡i+1(kG))=∑i=ℓ−nℓ−1dim⁡k(Rad⁡i(kG)/Rad⁡i+1(kG))=dim⁡kRad⁡ℓ−n(kG).\begin{aligned} \dim_k\Soc^n(kG) &=\sum_{i=0}^{n-1}\dim_k\big(\Rad^i(kG)/\Rad^{i+1}(kG)\big) \\ &=\sum_{i=\ell-n}^{\ell-1}\dim_k\big(\Rad^i(kG)/\Rad^{i+1}(kG)\big) \\ &=\dim_k\Rad^{\ell-n}(kG). \end{aligned}

The inclusion is therefore an equality. □

Now we come to an example. Abelian pp-groups are less interesting for this purpose because they are direct products of cyclic pp-groups, and we have a very explicit description of the group rings of cyclic pp-groups and of direct products. The smallest nonabelian examples have order 88; the group D8D_8 is considered in Webb, Example 6.4.2. Recall also that all groups of order p2p^2 are abelian. There are nine nonabelian groups of order 1616, and two nonabelian groups of order 2727. Let us consider one of the latter, namely

M27=C9⋊C3=⟨a,b∣a9=b3=1, bab−1=a4⟩.M_{27}=C_9\rtimes C_3 =\braket{a,b\mid a^9=b^3=1,\ bab^{-1}=a^4}.

Let kk be a fixed field of characteristic 33. We will first compute the Jennings/dimension series of M27M_{27}. The first term is M27M_{27} itself, and the second is the Frattini subgroup:

Φ(M27)=M273[M27,M27].\Phi(M_{27})=M_{27}^3[M_{27},M_{27}].

We have M273={1,a3,a6}M_{27}^3=\{1,a^3,a^6\}; indeed, (aibj)3=a3i(a^ib^j)^3=a^{3i}. The derived subgroup is generated by bab−1a−1=a3bab^{-1}a^{-1}=a^3: the quotient by ⟨a3⟩\braket{a^3} is abelian, and this generator is itself a commutator. Thus both subgroups are ⟨a3⟩\braket{a^3}, and the Frattini subgroup is ⟨a3⟩\braket{a^3}.

The first index is 27/3=927/3=9, so n1=2n_1=2. The only two choices for the next subgroup are ⟨a3⟩\braket{a^3} and 11, corresponding to n2=0n_2=0 and n2=1n_2=1. The power condition requires

Γ3(M27)⊃M273=⟨a3⟩,\Gamma_3(M_{27})\supset M_{27}^3=\braket{a^3},

so Γ3=Γ2\Gamma_3=\Gamma_2 and n2=0n_2=0. To see that the next term is trivial, observe that ⟨a3⟩\braket{a^3} is central of order 33. Consequently the decreasing series

M27,⟨a3⟩,⟨a3⟩,1,1,…M_{27},\quad\braket{a^3},\quad\braket{a^3},\quad1,\quad1,\dots

satisfies both the commutator and power conditions. The fastest such series is contained in it, so Γ4=1\Gamma_4=1. Therefore the Jennings series is

M27⊃⟨a3⟩⊃⟨a3⟩⊃⟨1⟩,n1=2,n2=0,n3=1.M_{27}\supset\braket{a^3}\supset\braket{a^3}\supset\braket{1}, \qquad n_1=2,\quad n_2=0,\quad n_3=1.

Hence the Poincaré polynomial is

∏r=13(1+tr+t2r)nr=(1+t+t2)2(1+t3+t6)=1+2t+3t2+3t3+3t4+3t5+3t6+3t7+3t8+2t9+t10.\begin{aligned} \prod_{r=1}^3(1+t^r+t^{2r})^{n_r} &=(1+t+t^2)^2(1+t^3+t^6) \\ &=1+2t+3t^2+3t^3+3t^4+3t^5 \\ &\qquad+3t^6+3t^7+3t^8+2t^9+t^{10}. \end{aligned}

This gives all the dimensions of the radical layers, and of the socle layers by Theorem 9, and tells us that the Loewy length is 1111.

References

The original course-paper version is available as a PDF.

Back to the blog →