Introduction
Throughout, p p p will be an odd prime number.
Let A A A be the mod p p p Steenrod algebra. The cohomology of A A A , Ext A ∗ ( F p , F p ) \operatorname{Ext}^{*}_{A}(\mathbb{F}_{p},\mathbb{F}_{p}) Ext A ∗ ( F p , F p ) , is the E 2 E_{2} E 2 -page of the classical Adams spectral sequence converging to the p p p -component of the stable homotopy groups of spheres, an object of great interest in stable homotopy theory. The goal of this paper is to try to understand the cohomology of A A A along the lines of the so-called Quillen stratification for the mod 2 Steenrod algebra, as presented in [19 ] and [20 Chapter 5] .
As Milnor [12 ] tells us, the Steenrod algebra is the Hopf algebra with graded dual
A ∨ = F p [ ξ 1 , ξ 2 , ξ 3 , … ] ⊗ Λ ( τ 0 , τ 1 , τ 2 , … ) , A^{\vee}=\mathbb{F}_{p}[\xi_{1},\xi_{2},\xi_{3},\ldots]\otimes\Lambda(\tau_{0},\tau_{1},\tau_{2},\ldots), A ∨ = F p [ ξ 1 , ξ 2 , ξ 3 , … ] ⊗ Λ ( τ 0 , τ 1 , τ 2 , … ) ,
graded by deg ξ n = 2 ( p n − 1 ) \deg\xi_{n}=2(p^{n}-1) deg ξ n = 2 ( p n − 1 ) and deg τ n = 2 p n − 1 \deg\tau_{n}=2p^{n}-1 deg τ n = 2 p n − 1 , with coproduct
ξ n ↦ ∑ i = 0 n ξ n − i p i ⊗ ξ i , \xi_{n}\mapsto\sum_{i=0}^{n}\xi_{n-i}^{p^{i}}\otimes\xi_{i}, ξ n ↦ i = 0 ∑ n ξ n − i p i ⊗ ξ i ,
τ n ↦ τ n ⊗ 1 + ∑ i = 0 n ξ n − i p i ⊗ τ i , \tau_{n}\mapsto\tau_{n}\otimes1+\sum_{i=0}^{n}\xi_{n-i}^{p^{i}}\otimes\tau_{i}, τ n ↦ τ n ⊗ 1 + i = 0 ∑ n ξ n − i p i ⊗ τ i ,
where ξ 0 = 1 \xi_{0}=1 ξ 0 = 1 when it appears in these formulas.
To understand a Hopf algebra like A A A , a good first step is to understand its sub-Hopf algebras. Any sub-Hopf algebra C C C of A A A is dual to a quotient Hopf algebra C ∨ C^{\vee} C ∨ of A ∨ A^{\vee} A ∨ , and by a theorem of Adams and Margolis [1 ] , those quotients must have the form
C ∨ = A ∨ / ( ξ 1 p c 1 , ξ 2 p c 2 , ξ 3 p c 3 , … ; τ 0 e 0 , τ 1 e 1 , τ 2 e 2 , … ) . C^{\vee}=A^{\vee}/(\xi_{1}^{p^{c_{1}}},\xi_{2}^{p^{c_{2}}},\xi_{3}^{p^{c_{3}}},\ldots;\tau_{0}^{e_{0}},\tau_{1}^{e_{1}},\tau_{2}^{e_{2}},\ldots). C ∨ = A ∨ / ( ξ 1 p c 1 , ξ 2 p c 2 , ξ 3 p c 3 , … ; τ 0 e 0 , τ 1 e 1 , τ 2 e 2 , … ) .
for some exponents ( ( c 1 , c 2 , c 3 , … ) ; ( e 0 , e 1 , e 2 , … ) ) ((c_{1},c_{2},c_{3},\ldots);(e_{0},e_{1},e_{2},\ldots)) (( c 1 , c 2 , c 3 , … ) ; ( e 0 , e 1 , e 2 , … )) , where c i c_{i} c i is either a non-negative integer or ∞ \infty ∞ and e i ∈ { 1 , 2 } e_{i}\in\{1,2\} e i ∈ { 1 , 2 } . The list of exponents is called the profile function for the Hopf algebra.
Adams and Margolis characterized precisely which such lists correspond to sub-Hopf algebras, thus providing a complete classification. We review the details in the next section and for now just mention one case: we let D D D be the sub-Hopf algebra of A A A with profile function
( ( 1 , 2 , 3 , … , n , … ) ; ( 2 , 2 , 2 , … ) ) . ((1,2,3,\ldots,n,\ldots);(2,2,2,\ldots)). (( 1 , 2 , 3 , … , n , … ) ; ( 2 , 2 , 2 , … )) .
This algebra is normal in A A A , and more generally, for any sub-Hopf algebra C C C of A A A , C ∩ D C\cap D C ∩ D is normal in C C C . The inclusion map i : C ∩ D → C i:C\cap D\to C i : C ∩ D → C induces a restriction map i ∗ : Ext C ∗ ( F p , F p ) → Ext C ∩ D ∗ ( F p , F p ) i^{*}:\operatorname{Ext}^{*}_{C}(\mathbb{F}_{p},\mathbb{F}_{p})\to\operatorname{Ext}^{*}_{C\cap D}(\mathbb{F}_{p},\mathbb{F}_{p}) i ∗ : Ext C ∗ ( F p , F p ) → Ext C ∩ D ∗ ( F p , F p ) . Because C ∩ D C\cap D C ∩ D is normal in C C C , the restriction map factors through the subalgebra of invariants under the action of the quotient Hopf algebra C / / ( C ∩ D ) : = C ⊗ C ∩ D F p C//(C \cap D) := C \otimes_{C \cap D} \mathbb{F}_p C // ( C ∩ D ) := C ⊗ C ∩ D F p :
Ext C ∗ ( F p , F p ) → i ∗ Ext C ∩ D ∗ ( F p , F p ) ↘ ↑ ( Ext C ∩ D ∗ ( F p , F p ) ) C / / C ∩ D \begin{aligned}
\operatorname{Ext}^{*}_{C}(\mathbb{F}_p,\mathbb{F}_p) &\xrightarrow{i^*} \operatorname{Ext}^{*}_{C\cap D}(\mathbb{F}_p,\mathbb{F}_p) \\
&\searrow\qquad\uparrow\\
&\left(\operatorname{Ext}^{*}_{C\cap D}(\mathbb{F}_p,\mathbb{F}_p)\right)_{C//C\cap D}
\end{aligned} Ext C ∗ ( F p , F p ) i ∗ Ext C ∩ D ∗ ( F p , F p ) ↘ ↑ ( Ext C ∩ D ∗ ( F p , F p ) ) C // C ∩ D
The map ı ˉ ∗ \bar{\imath}^{*} ˉ ∗ will make an appearance in Theorem B below.
Quillen stratification involves understanding cohomology algebras up to F F F -isomorphism, so recall:
Definition 1.1. A map φ \varphi φ of graded commutative algebras over a field of characteristic p p p is an F F F -isomorphism if
every element in the kernel of φ \varphi φ is nilpotent, and
for every y y y in the codomain of φ \varphi φ , there is an n n n so that y p n y^{p^n} y p n is in the image of φ \varphi φ .
We can strengthen the first part of the definition of F F F -isomorphism to echo the nilpotence theorem of Devinatz, Hopkins, and Smith. To do that, we work in a category which behaves like the classical stable homotopy category: given a sub-Hopf algebra C C C of A A A , let Stable ( C ∨ ) \operatorname{Stable}(C^\vee) Stable ( C ∨ ) be the category as defined in [5 Section 9.5] and examined in detail in [20 ] . The objects of Stable ( C ∨ ) \operatorname{Stable}(C^\vee) Stable ( C ∨ ) are cochain complexes of injective graded left comodules over C ∨ C^\vee C ∨ , and the morphisms are cochain homotopy classes of maps. This category is a stable homotopy category in the sense of [5 ] , and we may refer to its objects as “spectra.” In this category, the sphere object S S S is an injective resolution of F p \mathbb{F}_p F p , and if X X X is any object, then we may write π ∗ ( X ) \pi_*(X) π ∗ ( X ) for the chain homotopy classes of maps from S S S to X X X . If X X X is the injective resolution of a comodule M M M , then π ∗ ( X ) ≅ Ext C ∗ ( F p , M ) \pi_*(X) \cong\operatorname{Ext}^{*}_{C}(\mathbb{F}_p,M) π ∗ ( X ) ≅ Ext C ∗ ( F p , M ) . (Note that if M M M is a C ∨ C^\vee C ∨ -comodule, then M M M can also be viewed as a C C C -comodule, and so here Ext is computed in the category of C C C -modules.) Also in this category, any sub-Hopf algebra B B B of C C C defines a ring spectrum called H B HB H B : this is an injective resolution of C ∨ □ B F p ≅ ( C ⊗ B F p ) ∨ C^\vee\square_B \mathbb{F}_p \cong(C \otimes_B \mathbb{F}_p)^\vee C ∨ □ B F p ≅ ( C ⊗ B F p ) ∨ , and its coefficient ring is π ∗ H B = H B ∗ ≅ Ext B ∗ ( F p , F p ) \pi_*HB = HB_* \cong\operatorname{Ext}^{*}_{B}(\mathbb{F}_p,\mathbb{F}_p) π ∗ H B = H B ∗ ≅ Ext B ∗ ( F p , F p ) . If X X X is the injective resolution of a C ∨ C^\vee C ∨ -comodule M M M , then the quotient map C ∨ → B ∨ C^\vee\to B^\vee C ∨ → B ∨ gives M M M the structure of a B ∨ B^\vee B ∨ -comodule, and we have H B ∗ ( X ) ≅ Ext B ∗ ( F p , M ) HB_*(X) \cong\operatorname{Ext}^{*}_{B}(\mathbb{F}_p,M) H B ∗ ( X ) ≅ Ext B ∗ ( F p , M ) .
Definition 1.2. Given a Hopf algebra C C C , we say that a collection S S S of sub-Hopf algebras of C C C detects nilpotence in Stable ( C ∨ ) \operatorname{Stable}(C^\vee) Stable ( C ∨ ) if the following conditions hold:
(a) Fix a ring spectrum R R R and α ∈ π ∗ ( R ) \alpha\in\pi_*(R) α ∈ π ∗ ( R ) . Then α \alpha α is nilpotent if H B ∗ α = 0 HB_*\alpha= 0 H B ∗ α = 0 for all B ∈ S B \in S B ∈ S .
(b) Fix a finite spectrum Y Y Y and a self-map f : Y → Y f:Y \to Y f : Y → Y . Then f f f is nilpotent if and only if H B ∗ f HB_*f H B ∗ f is nilpotent for all B ∈ S B \in S B ∈ S .
(c) Fix a finite spectrum F F F , an arbitrary spectrum X X X , and a map f : F → X f:F \to X f : F → X . Then f f f is smash-nilpotent if $1_{HB} \wedge f:HB \wedge F \to HB \wedge X$ is zero for all B ∈ S B \in S B ∈ S .
We have three main results in this paper; here are the first two.
Theorem A. Fix an odd prime p p p and suppose that C C C is a sub-Hopf algebra of A A A . The sub-Hopf algebra C ∩ D C \cap D C ∩ D detects nilpotence in Stable ( C ∨ ) \operatorname{Stable}(C^\vee) Stable ( C ∨ ) .
Theorem B. Fix an odd prime p p p and suppose that C C C is a sub-Hopf algebra of A A A . The map
ı ˉ ∗ : Ext C ∗ ( F p , F p ) ⟶ ( Ext C ∩ D ∗ ( F p , F p ) ) C / / C ∩ D \bar{\imath}^{*}:\operatorname{Ext}^{*}_{C}(\mathbb{F}_p,\mathbb{F}_p)\longrightarrow\left(\operatorname{Ext}^{*}_{C\cap D}(\mathbb{F}_p,\mathbb{F}_p)\right)_{C//C\cap D} ˉ ∗ : Ext C ∗ ( F p , F p ) ⟶ ( Ext C ∩ D ∗ ( F p , F p ) ) C // C ∩ D is an F F F -isomorphism.
For example if C = A C=A C = A , the first theorem tells us that every element in the kernel of the restriction map i ∗ : Ext A ∗ ( F p , F p ) → Ext D ∗ ( F p , F p ) i^{*}:\operatorname{Ext}^{*}_{A}(\mathbb{F}_p,\mathbb{F}_p)\to\operatorname{Ext}^{*}_{D}(\mathbb{F}_p,\mathbb{F}_p) i ∗ : Ext A ∗ ( F p , F p ) → Ext D ∗ ( F p , F p ) is nilpotent, and the second theorem tells us that the map
ı ˉ ∗ : Ext A ∗ ( F p , F p ) ⟶ ( Ext D ∗ ( F p , F p ) ) A / / D \bar{\imath}^{*}:\operatorname{Ext}^{*}_{A}(\mathbb{F}_p,\mathbb{F}_p)\longrightarrow\left(\operatorname{Ext}^{*}_{D}(\mathbb{F}_p,\mathbb{F}_p)\right)^{A//D} ˉ ∗ : Ext A ∗ ( F p , F p ) ⟶ ( Ext D ∗ ( F p , F p ) ) A // D
is an F F F -isomorphism. is an F F F -isomorphism.
These theorems also hold at the prime 2 [19 ] , and in that setting one can use them to go further, in particular to obtain results involving “elementary” sub-Hopf algebras. A Hopf algebra E E E over a field k k k of characteristic p > 0 p > 0 p > 0 is called elementary if ( I E ) p = 0 (IE)^p = 0 ( I E ) p = 0 , where I E IE I E is the augmentation ideal of E E E . At the prime 2, this means that E E E is isomorphic as an algebra to k [ x 1 , x 2 , … ] / ( x i 2 ) k[x_1,x_2,\ldots]/(x_i^2) k [ x 1 , x 2 , … ] / ( x i 2 ) , whereas at odd primes E E E is isomorphic to a tensor product of a polynomial algebra truncated at height p p p and an exterior algebra:
E ≅ a l g k [ x 1 , x 2 , … ] / ( x i p ) ⊗ Λ ( y 1 , y 2 , … ) . E \cong_{\mathrm{alg}} k[x_1,x_2,\ldots]/(x_i^p) \otimes\Lambda(y_1,y_2,\ldots). E ≅ alg k [ x 1 , x 2 , … ] / ( x i p ) ⊗ Λ ( y 1 , y 2 , … ) .
Up to F F F -isomorphism the cohomology of an elementary Hopf algebra is a polynomial algebra. For the mod 2 Steenrod algebra A A A , we have this:
Theorem 1.3 ([19 ] ). We work at the prime 2. Fix a sub-Hopf algebra C C C of the mod 2 Steenrod algebra.
(a) The elementary sub-Hopf algebras of C C C detect nilpotence.
(b) The map
Ext C ∗ ( F 2 , F 2 ) ⟶ ( lim E Ext E ∗ ( F 2 , F 2 ) ) C \operatorname{Ext}^{*}_{C}(\mathbb{F}_2,\mathbb{F}_2) \longrightarrow\left(\lim_{E} \operatorname{Ext}^{*}_{E}(\mathbb{F}_2,\mathbb{F}_2)\right)^C Ext C ∗ ( F 2 , F 2 ) ⟶ ( E lim Ext E ∗ ( F 2 , F 2 ) ) C is an F F F -isomorphism, where the limit is taken over the category whose objects are the elementary sub-Hopf algebras of C C C and whose morphisms are inclusions.
This theorem deserves to be called Quillen stratification for the mod 2 Steenrod algebra. If k k k is a field of characteristic p p p , Quillen stratification for a finite group G G G provides a description of H ∗ ( G ; k ) H^*(G;k) H ∗ ( G ; k ) : there is an F F F -isomorphism
H ∗ ( G ; k ) ⟶ lim Elem ( G ) H ∗ ( E ; k ) , H^*(G;k) \longrightarrow\lim_{\operatorname{Elem}(G)} H^*(E;k), H ∗ ( G ; k ) ⟶ Elem ( G ) lim H ∗ ( E ; k ) ,
where Elem ( G ) \operatorname{Elem}(G) Elem ( G ) is the category whose objects E E E are elementary abelian p p p -subgroups of G G G and whose morphisms are generated by inclusions and conjugation by elements of G G G . For elementary sub-Hopf algebras of the Steenrod algebra, it turns out that one can separate the inclusions and the conjugations, so the inclusions combine to give lim E Ext E ∗ ( F 2 , F 2 ) \lim_E \operatorname{Ext}^{*}_E(\mathbb{F}_2,\mathbb{F}_2) lim E Ext E ∗ ( F 2 , F 2 ) , while the analog of taking the limit under conjugations is to take the invariants.
Wilkerson showed [23 ] that the elementary sub-Hopf algebras of the odd primary Steenrod algebra do not detect nilpotence, so the precise analog of this at odd primes cannot hold. We can, though, provide a partial replacement. We describe the elementary sub-Hopf algebras of the odd primary Steenrod algebra in the next section; the relevant parts here are that (a) each elementary B ⊆ A B \subseteq A B ⊆ A has a bounded profile function : the profile function has the form ( ( b 1 , b 2 , … ) ; ( e 0 , e 1 , … ) ) ((b_1,b_2,\ldots);(e_0,e_1,\ldots)) (( b 1 , b 2 , … ) ; ( e 0 , e 1 , … )) such that there is a number M M M with each b i ≤ M b_i \leq M b i ≤ M ; and (b) each elementary B ⊆ A B \subseteq A B ⊆ A is contained in D D D ; indeed, D D D is the smallest Hopf algebra containing every elementary sub-Hopf algebra. There are, though, sub-Hopf algebras of D D D with bounded profile functions which are not elementary, say for example the one given by the profile function ( ( 1 , 2 , 3 , 3 , 3 , … ) ; ( 2 , 2 , 2 , … ) ) ((1,2,3,3,3,\ldots);(2,2,2,\ldots)) (( 1 , 2 , 3 , 3 , 3 , … ) ; ( 2 , 2 , 2 , … )) . The following partial replacement for Theorem 1.3 is the third main result of this paper.
Theorem C. Fix a sub-Hopf algebra C ⊆ A C \subseteq A C ⊆ A . Let Bdd ( C ∩ D ) \operatorname{Bdd}(C \cap D) Bdd ( C ∩ D ) be the collection of sub-Hopf algebras of C ∩ D C \cap D C ∩ D with bounded profile function. Then Bdd ( C ∩ D ) \operatorname{Bdd}(C \cap D) Bdd ( C ∩ D ) detects nilpotence in Stable ( C ∨ ) \operatorname{Stable}(C^\vee) Stable ( C ∨ ) .
One might also conjecture that there is an F F F -isomorphism
H ∗ ( C ) ⟶ ( lim Bdd ( B ∩ D ) H ∗ ( B ) ) C , H^*(C) \longrightarrow\left(\lim_{\operatorname{Bdd}(B\cap D)} H^*(B)\right)^C, H ∗ ( C ) ⟶ ( Bdd ( B ∩ D ) lim H ∗ ( B ) ) C ,
where Bdd ( B ∩ D ) \operatorname{Bdd}(B\cap D) Bdd ( B ∩ D ) is made into a category with the morphisms being inclusions. This conjecture is discussed in Section 6. Also see Conjecture 6.1 for a stronger statement, a more direct analog of the p = 2 p=2 p = 2 theorem.
The structure of the paper: in § 2 we provide background, notation, and a few key lemmas. In § 3 we prove a theorem that provides a framework within the Steenrod algebra for proving results like Theorems A and B. We then use that theorem in § 4 to prove an intermediate result: a certain sub-Hopf algebra B i B_i B i (where D ⊆ B i ⊆ A D \subseteq B_i \subseteq A D ⊆ B i ⊆ A ) detects nilpotence and gives an F F F -isomorphism, and we use that in § 5 to show that D D D detects nilpotence and gives an F F F -isomorphism, yielding our first two main theorems. In § 6 we prove Theorem C , and in § 7 we discuss some further directions. A key computation used in § 4 comes from unpublished work of Burklund, Levy, and Pstrągowski; it is valid at all primes, but in an appendix we give an independent proof for the p = 3 p=3 p = 3 case.
Preliminaries
As we have said, A A A denotes the mod p p p Steenrod algebra with p > 2 p>2 p > 2 . Throughout, modules will be understood to be left modules, and the same for comodules.
Hopf algebras related to A A A
First we recall the Adams-Margolis theorem classifying sub-Hopf algebras of A A A [1 ] . We consider a sub-coalgebra B B B of A A A whose graded dual is the quotient algebra of A ∨ A^\vee A ∨ of the form
B ∨ = A ∨ / ( ξ 1 p b 1 , ξ 2 p b 2 , ξ 3 p b 3 , … ; τ 0 e 0 , τ 1 e 1 , τ 2 e 2 , … ) B^\vee= A^\vee/(\xi_1^{p^{b_1}}, \xi_2^{p^{b_2}}, \xi_3^{p^{b_3}}, \ldots; \tau_0^{e_0}, \tau_1^{e_1}, \tau_2^{e_2}, \ldots) B ∨ = A ∨ / ( ξ 1 p b 1 , ξ 2 p b 2 , ξ 3 p b 3 , … ; τ 0 e 0 , τ 1 e 1 , τ 2 e 2 , … )
for some exponents ( ( b 1 , b 2 , b 3 , … ) ; ( e 0 , e 1 , e 2 , … ) ) ((b_1,b_2,b_3,\ldots);(e_0,e_1,e_2,\ldots)) (( b 1 , b 2 , b 3 , … ) ; ( e 0 , e 1 , e 2 , … )) , where b i ∈ { 0 , 1 , 2 , … , ∞ } b_i \in\{0,1,2,\ldots,\infty\} b i ∈ { 0 , 1 , 2 , … , ∞ } and e i ∈ { 1 , 2 } e_i \in\{1,2\} e i ∈ { 1 , 2 } . (If some b i = ∞ b_i=\infty b i = ∞ , then do not mod out by any power of ξ i \xi_i ξ i .) We have the following conditions:
for all i , j ≥ 1 i,j \geq1 i , j ≥ 1 , either b i ≤ j + b i + j b_i \leq j+b_{i+j} b i ≤ j + b i + j or b j ≤ b i + j b_j \leq b_{i+j} b j ≤ b i + j , and
for all i ≥ 1 i \geq1 i ≥ 1 and j ≥ 0 j \geq0 j ≥ 0 , if e i + j = 1 e_{i+j}=1 e i + j = 1 , then either b i ≤ j b_i \leq j b i ≤ j or e j = 1 e_j=1 e j = 1 .
Theorem 2.1 (Theorem 2.4 in [1 ] ). If the sub-coalgebra B ⊆ A B \subseteq A B ⊆ A satisfies these conditions, then it is a sub-Hopf algebra, and conversely, every sub-Hopf algebra of A A A is a sub-coalgebra of this form satisfying these conditions.
We also need to understand when a sub-Hopf algebra of A A A is normal. If H H H is a cocommutative Hopf algebra over F p \mathbb{F}_p F p , then a sub-Hopf algebra B ⊆ H B \subseteq H B ⊆ H is normal if the left and right ideals generated by B B B are equal: H ⋅ I B = I B ⋅ H H \cdot I_B=I_B \cdot H H ⋅ I B = I B ⋅ H , where I B = ker ( B → F p ) I_B=\ker(B \to\mathbb{F}_p) I B = ker ( B → F p ) is the augmentation ideal of $B. It is well documented, for example in [9 ] , which sub-Hopf algebras of the mod 2 Steenrod algebra are normal: those with non-decreasing profile function. The author has not been able to find a good reference for the odd primary case, though.
Proposition 2.2. Suppose that B ⊆ A B \subseteq A B ⊆ A is a sub-Hopf algebra with profile function
( ( b 1 , b 2 , b 3 , … ) ; ( e 0 , e 1 , e 2 , … ) ) . ((b_1,b_2,b_3,\ldots);(e_0,e_1,e_2,\ldots)). (( b 1 , b 2 , b 3 , … ) ; ( e 0 , e 1 , e 2 , … )) . Then B B B is normal in A A A if and only if
(a) the sequence ( b i ) (b_i) ( b i ) is non-decreasing,
(b) the sequence ( e i ) (e_i) ( e i ) is non-decreasing, and
(c) if e n = 1 e_n=1 e n = 1 for some n n n , then b i = 0 b_i=0 b i = 0 for i ≤ n i \leq n i ≤ n .
Proof. The sub-Hopf algebra B ⊆ A B \subseteq A B ⊆ A is normal if and only if there is an extension of Hopf algebras
B ⟶ A ⟶ A / / B , B \longrightarrow A \longrightarrow A//B, B ⟶ A ⟶ A // B ,
and this happens if and only if there is an extension of Hopf algebras
A ∨ □ B ∨ F p → A ∨ → B ∨ . A^{\vee} \square_{B^{\vee}} F_p \to A^{\vee} \to B^{\vee}. A ∨ □ B ∨ F p → A ∨ → B ∨ .
We have
B ∨ = A ∨ / ( ξ 1 p b 1 , ξ 2 p b 2 , ξ 3 p b 3 , … ; τ 0 e 0 , τ 1 e 1 , τ 2 e 2 , … ) , B^{\vee} = A^{\vee}/(\xi_1^{p^{b_1}},\xi_2^{p^{b_2}},\xi_3^{p^{b_3}},\ldots;\tau_0^{e_0},\tau_1^{e_1},\tau_2^{e_2},\ldots), B ∨ = A ∨ / ( ξ 1 p b 1 , ξ 2 p b 2 , ξ 3 p b 3 , … ; τ 0 e 0 , τ 1 e 1 , τ 2 e 2 , … ) ,
and so
A ∨ □ B ∨ F p = F p [ ξ 1 p b 1 , ξ 2 p b 2 , ξ 3 p b 3 , … ] ⊗ Λ ( τ n : e n = 1 ) . A^{\vee} \square_{B^{\vee}} F_p = F_p[\xi_1^{p^{b_1}},\xi_2^{p^{b_2}},\xi_3^{p^{b_3}},\ldots]\otimes\Lambda(\tau_n:e_n=1). A ∨ □ B ∨ F p = F p [ ξ 1 p b 1 , ξ 2 p b 2 , ξ 3 p b 3 , … ] ⊗ Λ ( τ n : e n = 1 ) .
Let C ∨ = A ∨ □ B ∨ F p C^{\vee}=A^{\vee}\square_{B^{\vee}}F_p C ∨ = A ∨ □ B ∨ F p . It is clear that if the conditions hold, then C ∨ C^{\vee} C ∨ is a sub-Hopf algebra: if the conditions hold, then B B B is normal. We need to check the converse, so assume that C ∨ C^{\vee} C ∨ is a sub-Hopf algebra. Since the coproduct on ξ n p b n \xi_n^{p^{b_n}} ξ n p b n contains terms ξ n − i p b n + i ⊗ ξ i p b n \xi_{n-i}^{p^{b_n+i}}\otimes\xi_i^{p^{b_n}} ξ n − i p b n + i ⊗ ξ i p b n , we conclude that b i ≤ b n b_i\leq b_n b i ≤ b n if i < n i<n i < n . Similarly, the coproduct on τ n \tau_n τ n contains terms ξ n − i p i ⊗ τ i \xi_{n-i}^{p^i}\otimes\tau_i ξ n − i p i ⊗ τ i , so if e n = 1 e_n=1 e n = 1 (meaning that τ n \tau_n τ n is a generator of C ∨ C^{\vee} C ∨ ), then e i = 1 e_i=1 e i = 1 when i < n i<n i < n , and ξ n \xi_n ξ n must lie in C ∨ C^{\vee} C ∨ : the exponent b n b_n b n must be 0. Since b i ≤ b n b_i\leq b_n b i ≤ b n for i ≤ n i\leq n i ≤ n , we conclude that b i = 0 b_i=0 b i = 0 for all i ≤ n i\leq n i ≤ n .
In practice we will typically cite the conditions to conclude that a sub-Hopf algebra is normal, and we will also typically be in the situation where each e i = 2 e_i=2 e i = 2 ; or somewhat more generally we will have Hopf algebras B ′ ⊆ B B'\subseteq B B ′ ⊆ B which agree on the exterior parts of their profile functions, and we will want to know when B ′ B' B ′ is normal in B B B .
We define some sub- and quotient Hopf algebras of A A A .
(This one is only used to help define some of the others in this list.) Write Λ ∨ \Lambda^{\vee} Λ ∨ for the exterior quotient Hopf algebra of A ∨ A^{\vee} A ∨ : Λ ∨ : = Λ ( τ 0 , τ 1 , τ 2 , … ) \Lambda^{\vee}:=\Lambda(\tau_0,\tau_1,\tau_2,\ldots) Λ ∨ := Λ ( τ 0 , τ 1 , τ 2 , … ) . Its graded dual is Λ : = Λ ( Q 0 , Q 1 , Q 2 , … ) ⊆ A \Lambda:=\Lambda(Q_0,Q_1,Q_2,\ldots)\subseteq A Λ := Λ ( Q 0 , Q 1 , Q 2 , … ) ⊆ A , which is normal in A A A .
We have already introduced D ⊆ A D\subseteq A D ⊆ A , defined by D ∨ ≅ A ∨ / ( ξ 1 p , ξ 2 p 2 , … , ξ n p n , … ) D^{\vee}\cong A^{\vee}/(\xi_1^p,\xi_2^{p^2},\ldots,\xi_n^{p^n},\ldots) D ∨ ≅ A ∨ / ( ξ 1 p , ξ 2 p 2 , … , ξ n p n , … ) . Its profile function is therefore
( ( 1 , 2 , 3 , … , n , … ) ; ( 2 , 2 , 2 , … ) ) . ((1,2,3,\ldots,n,\ldots);(2,2,2,\ldots)). (( 1 , 2 , 3 , … , n , … ) ; ( 2 , 2 , 2 , … )) .
This is called “D D D ” because the polynomial part of the profile function traces the “diagonal” line n ↦ n n\mapsto n n ↦ n . D D D is normal in A A A .
Define B i ⊆ A Bi\subseteq A B i ⊆ A by B i ∨ ≅ A ∨ / ( ξ 1 p , ξ 2 p 2 , ξ 3 p 4 , … , ξ n p ( n 2 ) + 1 , … ) Bi^{\vee}\cong A^{\vee}/(\xi_1^p,\xi_2^{p^2},\xi_3^{p^4},\ldots,\xi_n^{p^{\binom{n}{2}+1}},\ldots) B i ∨ ≅ A ∨ / ( ξ 1 p , ξ 2 p 2 , ξ 3 p 4 , … , ξ n p ( 2 n ) + 1 , … ) , so its profile function is
( ( 1 , 2 , 4 , 7 , … , ( n 2 ) + 1 , … ) ; ( 2 , 2 , 2 , … ) ) . ((1,2,4,7,\ldots,\binom{n}{2}+1,\ldots);(2,2,2,\ldots)). (( 1 , 2 , 4 , 7 , … , ( 2 n ) + 1 , … ) ; ( 2 , 2 , 2 , … )) .
In case you’re curious, “B i Bi B i ” stands for “binomial.” B i Bi B i is normal in A A A . This algebra will be used as a sort of catalyst: it will appear in some auxiliary results, in particular Corollaries 4.4 and 5.2, which we use to prove our main results.
Define P P P to be dual to the polynomial part of A ∨ A^{\vee} A ∨ , P ∨ = F p [ ξ 1 , ξ 2 , ξ 3 , … ] P^{\vee}=F_p[\xi_1,\xi_2,\xi_3,\ldots] P ∨ = F p [ ξ 1 , ξ 2 , ξ 3 , … ] . Since P ∨ P^{\vee} P ∨ is a sub-Hopf algebra of A ∨ A^{\vee} A ∨ , we see that P P P is naturally a quotient Hopf algebra of A A A : P ≅ A / / Λ P\cong A//\Lambda P ≅ A //Λ . One can also view P P P as a sub-Hopf algebra of A A A : use the profile function
( ( ∞ , ∞ , ∞ , … ) ; ( 1 , 1 , 1 , … ) ) . ((\infty,\infty,\infty,\ldots);(1,1,1,\ldots)). (( ∞ , ∞ , ∞ , … ) ; ( 1 , 1 , 1 , … )) .
As a consequence, A A A can be viewed a semi-direct product of P P P and Λ \Lambda Λ . Because the composite P → A → P P\to A\to P P → A → P of inclusion followed by projection is the identity, we see that the cohomology of P P P splits off of the cohomology of A A A . Somewhat like B i Bi B i , P P P is used as a catalyst, but it appears pretty naturally because of the properties just described.
Define A ≥ n ⊆ A A_{\ge n} \subseteq A A ≥ n ⊆ A by A ≥ n ∨ ≅ A ∨ / ( ξ 1 , … , ξ n − 1 , τ 0 , … , τ n − 1 ) A^{\vee}_{\ge n} \cong A^{\vee}/(\xi_1,\ldots,\xi_{n-1},\tau_0,\ldots,\tau_{n-1}) A ≥ n ∨ ≅ A ∨ / ( ξ 1 , … , ξ n − 1 , τ 0 , … , τ n − 1 ) . Its profile function is therefore
( ( 0 , … , 0 , ∞ , ∞ , ∞ , … ) ; ( 1 , … , 1 , 2 , 2 , 2 , … ) ) . ((0,\ldots,0,\infty,\infty,\infty,\ldots);(1,\ldots,1,2,2,2,\ldots)). (( 0 , … , 0 , ∞ , ∞ , ∞ , … ) ; ( 1 , … , 1 , 2 , 2 , 2 , … )) .
A ≥ n A_{\ge n} A ≥ n is normal in A A A .
A ≤ n − 1 ∨ ≅ F p [ ξ 1 , … , ξ n − 1 ] ⊗ Λ ( τ 0 , … , τ n − 1 ) A^{\vee}_{\le n-1}\cong\mathbb{F}_p[\xi_1,\ldots,\xi_{n-1}]\otimes\Lambda(\tau_0,\ldots,\tau_{n-1}) A ≤ n − 1 ∨ ≅ F p [ ξ 1 , … , ξ n − 1 ] ⊗ Λ ( τ 0 , … , τ n − 1 )
with the same coproduct formulas as in A ∨ A^{\vee} A ∨ . We may write A < n A_{<n} A < n for the same thing.
We will not need to use elementary sub-Hopf algebras of A A A in any crucial way in this paper, but we include the following description for reference and to back up the claim in the introduction that elementary sub-Hopf algebras have bounded profile functions and lie in D D D . The following facts are proved in [20 2.1.7–8] . For each r ≥ − 1 r\ge-1 r ≥ − 1 there is a maximal elementary sub-Hopf algebra E ( r ) E(r) E ( r ) given by the profile function
E ( − 1 ) : ( ( 0 , 0 , 0 , … ) ; ( 2 , 2 , 2 , … ) ) , E(-1):((0,0,0,\ldots);(2,2,2,\ldots)), E ( − 1 ) : (( 0 , 0 , 0 , … ) ; ( 2 , 2 , 2 , … )) ,
and for $r\ge 0
E ( r ) : ( ( 0 , … , 0 , r + 1 , r + 1 , r + 1 , … ) ; ( 1 , … , 1 , 2 , 2 , 2 , … ) ) . E(r):((0,\ldots,0,r+1,r+1,r+1,\ldots);(1,\ldots,1,2,2,2,\ldots)). E ( r ) : (( 0 , … , 0 , r + 1 , r + 1 , r + 1 , … ) ; ( 1 , … , 1 , 2 , 2 , 2 , … )) .
We denote the profile function of E ( r ) E(r) E ( r ) by
( ( f 1 r , f 2 r , … ) ; ( e 0 r , e 1 r , … ) ) . ((f_1^r,f_2^r,\ldots);(e_0^r,e_1^r,\ldots)). (( f 1 r , f 2 r , … ) ; ( e 0 r , e 1 r , … )) .
If for some r r r , the list ( ( b 1 , b 2 , … ) ; ( e 0 , e 1 , … ) ) ((b_1,b_2,\ldots);(e_0,e_1,\ldots)) (( b 1 , b 2 , … ) ; ( e 0 , e 1 , … )) satisfies b i ≤ f i r b_i\le f_i^r b i ≤ f i r and e j ≤ e j r e_j\le e_j^r e j ≤ e j r for all i , j i,j i , j , then this is a valid profile function and corresponds to an elementary sub-Hopf algebra of E ( r ) E(r) E ( r ) . The maximal elementary sub-Hopf algebras are normal; the non-maximal ones need not be.
Remark 2.3. We note that Adams and Margolis have classified the sub-Hopf algebras of A < n A_{<n} A < n : see [1 Theorem 3.3] . These have the same form as for A A A : their duals look like A < n ∨ / ( ξ 1 p b 1 , … , ξ n − 1 p b n − 1 ; τ 0 e 0 , … , τ n − 1 e n − 1 ) A^{\vee}_{<n}/(\xi_1^{p^{b_1}},\ldots,\xi_{n-1}^{p^{b_{n-1}}};\tau_0^{e_0},\ldots,\tau_{n-1}^{e_{n-1}}) A < n ∨ / ( ξ 1 p b 1 , … , ξ n − 1 p b n − 1 ; τ 0 e 0 , … , τ n − 1 e n − 1 ) for exponents ( ( b 1 , … , b n − 1 ) ; ( e 0 , … , e n − 1 ) ) ((b_1,\ldots,b_{n-1});(e_0,\ldots,e_{n-1})) (( b 1 , … , b n − 1 ) ; ( e 0 , … , e n − 1 )) satisfying the same conditions as in Theorem 2.1 above.
Cohomology and Steenrod operations
We will usually write H ∗ ( B ; − ) H^*(B;-) H ∗ ( B ; − ) for Ext B ∗ ( F p , − ) \operatorname{Ext}^*_B(\mathbb{F}_p,-) Ext B ∗ ( F p , − ) , and we may omit the coefficients if they are trivial: H ∗ H^* H ∗ means H ∗ ( B ; F p ) H^*(B;\mathbb{F}_p) H ∗ ( B ; F p ) .
As codified by May in [11 Theorem 11.8] , there are Steenrod operations acting on the mod p p p cohomology of a cocommutative Hopf algebra B B B , and in particular for sub-Hopf algebras of A A A . We recall May’s notation:
P i : H s , t ( B ) → H s + ( 2 i − t ) ( p − 1 ) , p t ( B ) , β P i : H s , t ( B ) → H s + 1 + ( 2 i − t ) ( p − 1 ) , p t ( B ) . \mathcal{P}^i:H^{s,t}(B)\to H^{s+(2i-t)(p-1),pt}(B),
\\
\beta\mathcal{P}^i:H^{s,t}(B)\to H^{s+1+(2i-t)(p-1),pt}(B). P i : H s , t ( B ) → H s + ( 2 i − t ) ( p − 1 ) , pt ( B ) , β P i : H s , t ( B ) → H s + 1 + ( 2 i − t ) ( p − 1 ) , pt ( B ) .
These satisfy the Cartan formulas as well as some instability conditions:
P n ( x y ) = ∑ i + j = n P i ( x ) ⊗ P j ( y ) , β P n ( x y ) = ∑ i + j = n ( β P i ( x ) P j ( y ) + P i ( x ) β P j ( y ) ) , P i = 0 if 2 i < t or 2 i > s + t , β P i = 0 if 2 i < t or 2 i ≥ s + t , P i ( x ) = x p if 2 i = s + t . \begin{aligned}
\mathcal{P}^{n}(xy) &= \sum_{i+j=n} \mathcal{P}^{i}(x) \otimes\mathcal{P}^{j}(y), \\
\beta\mathcal{P}^{n}(xy) &= \sum_{i+j=n} \left(\beta\mathcal{P}^{i}(x)\mathcal{P}^{j}(y) + \mathcal{P}^{i}(x)\beta\mathcal{P}^{j}(y)\right), \\
\mathcal{P}^{i} &= 0 \quad\text{if } 2i < t \text{ or } 2i > s+t, \\
\beta\mathcal{P}^{i} &= 0 \quad\text{if } 2i < t \text{ or } 2i \ge s+t, \\
\mathcal{P}^{i}(x) &= x^{p} \quad\text{if } 2i=s+t.
\end{aligned} P n ( x y ) β P n ( x y ) P i β P i P i ( x ) = i + j = n ∑ P i ( x ) ⊗ P j ( y ) , = i + j = n ∑ ( β P i ( x ) P j ( y ) + P i ( x ) β P j ( y ) ) , = 0 if 2 i < t or 2 i > s + t , = 0 if 2 i < t or 2 i ≥ s + t , = x p if 2 i = s + t .
We also recall May’s regrading [11 p. 227] of these operations:
P ~ i = P i + t : H s , 2 t ⟶ H s + 2 i ( p − 1 ) , 2 p t , \widetilde{\mathcal{P}}^{i}=\mathcal{P}^{i+t}:H^{s,2t}\longrightarrow H^{s+2i(p-1),2pt}, P i = P i + t : H s , 2 t ⟶ H s + 2 i ( p − 1 ) , 2 pt ,
β P ~ i = β P i + t : H s , 2 t ⟶ H s + 2 i ( p − 1 ) + 1 , 2 p t . \beta\widetilde{\mathcal{P}}^{i}=\beta\mathcal{P}^{i+t}:H^{s,2t}\longrightarrow H^{s+2i(p-1)+1,2pt}. β P i = β P i + t : H s , 2 t ⟶ H s + 2 i ( p − 1 ) + 1 , 2 pt .
As May points out, these regraded operations only act on H s , t H^{s,t} H s , t when t t t is even, so they are not useful when dealing with the exterior part of the dual Steenrod algebra and any cohomology classes arising therefrom. When the internal degree t t t is even, the Steenrod operation P t / 2 : H s , t → H s , p t \mathcal{P}^{t/2}:H^{s,t}\to H^{s,pt} P t /2 : H s , t → H s , pt is analogous to the operation P 0 \mathcal{P}^{0} P 0 on the cohomology of a space, and indeed with the regrading, this operation is P ~ 0 \widetilde{\mathcal{P}}^{0} P 0 . It is not, however, the identity in general: it is instead induced by the p p p th power map on the dual — see [11 Proposition 11.10] .
The algebra D ( x ) D(x) D ( x ) and the elements h t s h_{ts} h t s , b t s b_{ts} b t s , and a n a_n a n
Despite the unfortunate notation clash with D D D , we follow Miller-Wilkerson [14 ] and write D ( x ) D(x) D ( x ) for the truncated polynomial algebra D ( x ) = F p [ x ] / ( x p ) D(x)=\mathbb{F}_p[x]/(x^p) D ( x ) = F p [ x ] / ( x p ) , and we write D ( x 1 , x 2 , … ) D(x_1,x_2,\ldots) D ( x 1 , x 2 , … ) for D ( x 1 ) ⊗ D ( x 2 ) ⊗ ⋯ D(x_1)\otimes D(x_2)\otimes\cdots D ( x 1 ) ⊗ D ( x 2 ) ⊗ ⋯ . It is standard that when p p p is odd, H ∗ ( D ( x ) ) ≅ Λ F p ( h ) ⊗ F p [ b ] H^*(D(x))\cong\Lambda_{\mathbb{F}_p}(h)\otimes\mathbb{F}_p[b] H ∗ ( D ( x )) ≅ Λ F p ( h ) ⊗ F p [ b ] , where h ∈ H 1 , ∣ x ∣ h\in H^{1,\lvert x\rvert} h ∈ H 1 , ∣ x ∣ and b ∈ H 2 , p ∣ x ∣ b\in H^{2,p\lvert x\rvert} b ∈ H 2 , p ∣ x ∣ , and b b b can be obtained from h h h as a p p p -fold Massey product, b = ⟨ h , … , h ⟩ b=\langle h,\ldots,h\rangle b = ⟨ h , … , h ⟩ . If D ( x ) D(x) D ( x ) is given the structure of a cocommutative Hopf algebra, e.g. with x x x primitive, then one has Steenrod operations acting on its cohomology, and in that case b b b can also be obtained via b = β P ∣ h ∣ / 2 ( h ) b=\beta\mathcal{P}^{\lvert h\rvert/2}(h) b = β P ∣ h ∣ /2 ( h ) , using the above notation.
The Milnor basis for A A A is obtained by dualizing the monomial basis for A ∨ A^\vee A ∨ . Let P t s \mathcal{P}^{s}_{t} P t s be the Milnor basis element dual to ξ t p s \xi_t^{p^s} ξ t p s . If B B B is a sub-Hopf algebra of A A A and if ξ t p s \xi_t^{p^s} ξ t p s is primitive in its dual B ∨ B^\vee B ∨ , then the algebra D ( P t s ) D(\mathcal{P}^{s}_{t}) D ( P t s ) is a quotient algebra of B B B , and so there is an induced map
H ∗ ( D ( P t s ) ) ⟶ H ∗ ( B ) . H^*(D(\mathcal{P}^{s}_{t}))\longrightarrow H^*(B). H ∗ ( D ( P t s )) ⟶ H ∗ ( B ) .
We denote the algebra generators in the domain by h t , s ∈ H 1 , ∣ ξ t p s ∣ ( D ( P t s ) ) h_{t,s}\in H^{1,\lvert\xi_t^{p^s}\rvert}(D(\mathcal{P}^{s}_{t})) h t , s ∈ H 1 , ∣ ξ t p s ∣ ( D ( P t s )) and b t , s ∈ H 2 , p ∣ ξ t p s ∣ ( D ( P t s ) ) b_{t,s}\in H^{2,p\lvert\xi_t^{p^s}\rvert}(D(\mathcal{P}^{s}_{t})) b t , s ∈ H 2 , p ∣ ξ t p s ∣ ( D ( P t s )) , and we use the same names for their images in H ∗ ( B ) H^*(B) H ∗ ( B ) . Of course algebra generators are only defined up to a scalar multiple, and it might be useful to be more precise, so we define h t , s h_{t,s} h t , s to be cohomology class represented by the cobar element [ ξ t p s ] [\xi_t^{p^s}] [ ξ t p s ] , and we define b t , s b_{t,s} b t , s by b t , s = β P ∣ ξ t p s ∣ / 2 ( h t , s ) b_{t,s}=\beta\mathcal{P}^{\lvert\xi_t^{p^s}\rvert/2}(h_{t,s}) b t , s = β P ∣ ξ t p s ∣ /2 ( h t , s ) . We often omit the commas, writing h t s h_{ts} h t s and b t s b_{ts} b t s instead.
If ξ t p s \xi_t^{p^s} ξ t p s and ξ t p s + 1 \xi_t^{p^{s+1}} ξ t p s + 1 are both primitive in B ∨ B^\vee B ∨ , then because P ~ 0 \widetilde{\mathcal{P}}^0 P 0 is induced by the p p p th power map on the dual, we have P ~ 0 ( h t s ) = h t , s + 1 \widetilde{\mathcal{P}}^0(h_{ts})=h_{t,s+1} P 0 ( h t s ) = h t , s + 1 and P ~ 0 ( b t s ) = b t , s + 1 \widetilde{\mathcal{P}}^0(b_{ts})=b_{t,s+1} P 0 ( b t s ) = b t , s + 1 in the cohomology of B B B .
We also recall that H ∗ ( Λ ( x ) ) ≅ F p [ a ] H^*(\Lambda(x))\cong\mathbb{F}_p[a] H ∗ ( Λ ( x )) ≅ F p [ a ] where a ∈ H 1 , ∣ x ∣ a\in H^{1,\lvert x\rvert} a ∈ H 1 , ∣ x ∣ is represented by [ x ] [x] [ x ] in the cobar complex. For each n n n , we have a sub-Hopf algebra Λ ( Q n ) ⊆ A \Lambda(Q_n)\subseteq A Λ ( Q n ) ⊆ A , dual to the quotient Λ ( τ n ) \Lambda(\tau_n) Λ ( τ n ) of A ∨ A^\vee A ∨ . We let a n a_n a n denote the algebra generator of H ∗ ( Λ ( Q n ) ) H^*(\Lambda(Q_n)) H ∗ ( Λ ( Q n )) . If B B B is a sub-Hopf algebra of A A A with τ n \tau_n τ n primitive in B ∨ B^\vee B ∨ , then Λ ( Q n ) \Lambda(Q_n) Λ ( Q n ) is a quotient Hopf algebra of B B B , and we also use a n ∈ H ∗ ( B ) a_n\in H^*(B) a n ∈ H ∗ ( B ) to denote the image of a n ∈ H ∗ ( Λ ( Q n ) ) a_n\in H^*(\Lambda(Q_n)) a n ∈ H ∗ ( Λ ( Q n )) .
Two lemmas
We present two lemmas which will be used throughout the paper.
As above, given D ( x ) = F p [ x ] / ( x p ) D(x)=\mathbb{F}_p[x]/(x^p) D ( x ) = F p [ x ] / ( x p ) , we write h , b h,b h , b for the algebra generators of H ∗ ( D ( x ) ) ≅ Λ ( h ) ⊗ F p [ b ] H^*(D(x))\cong\Lambda(h)\otimes\mathbb{F}_p[b] H ∗ ( D ( x )) ≅ Λ ( h ) ⊗ F p [ b ] . (This is assuming that p p p is odd; if one wants to consider this when p = 2 p=2 p = 2 , use the Λ ( x ) \Lambda(x) Λ ( x ) case below.) Given an extension of Hopf algebras over F p \mathbb{F}_p F p
C ⟶ B ⟶ D ( x ) , C\longrightarrow B\longrightarrow D(x), C ⟶ B ⟶ D ( x ) ,
there is an induced map H ∗ ( D ( x ) ) ⟶ H ∗ ( B ) H^*(D(x))\longrightarrow H^*(B) H ∗ ( D ( x )) ⟶ H ∗ ( B ) , and we also write h , b h,b h , b for the images of those generators under this map. Similarly, if we have a Hopf algebra extension
C ⟶ B ⟶ Λ ( x ) , C\longrightarrow B\longrightarrow\Lambda(x), C ⟶ B ⟶ Λ ( x ) ,
there is an induced map H ∗ ( Λ ( x ) ) ⟶ H ∗ ( B ) H^*(\Lambda(x))\longrightarrow H^*(B) H ∗ ( Λ ( x )) ⟶ H ∗ ( B ) , and we use a a a to denote both a generator of H ∗ ( Λ ( x ) ) H^*(\Lambda(x)) H ∗ ( Λ ( x )) and its image in H ∗ ( B ) H^*(B) H ∗ ( B ) . The following is due to Quillen and Venkov [21 ] .
Lemma 2.4 (Quillen-Venkov). (a) Given a Hopf algebra extension C ⟶ B ⟶ D ( x ) C\longrightarrow B\longrightarrow D(x) C ⟶ B ⟶ D ( x ) as above, if z ∈ H ∗ ( B ) z\in H^*(B) z ∈ H ∗ ( B ) is in the kernel of the restriction map to H ∗ ( C ) H^*(C) H ∗ ( C ) , then z 2 z^2 z 2 is in the ideal generated by b b b .
(b) Given a Hopf algebra extension C ⟶ B ⟶ Λ ( x ) C\longrightarrow B\longrightarrow\Lambda(x) C ⟶ B ⟶ Λ ( x ) as above, if z ∈ H ∗ ( B ) z\in H^*(B) z ∈ H ∗ ( B ) is in the kernel of the restriction map to H ∗ ( C ) H^*(C) H ∗ ( C ) , then z z z is in the ideal generated by a a a .
Proof. Quillen and Venkov’s result is about group cohomology, but their proof carries over essentially unchanged to this setting. In a bit more detail, for (a) the stated setup for their lemma is a little different: they start with a nonzero element h ∈ H 1 ( G ; Z / p Z ) = Hom ( G , Z / p Z ) h\in H^1(G;\mathbb{Z}/p\mathbb{Z})=\operatorname{Hom}(G,\mathbb{Z}/p\mathbb{Z}) h ∈ H 1 ( G ; Z / p Z ) = Hom ( G , Z / p Z ) , and then view h h h as corresponding to a group extension 1 ⟶ G ′ ⟶ G ⟶ Z / p Z ⟶ 1 1\longrightarrow G'\longrightarrow G\longrightarrow\mathbb{Z}/p\mathbb{Z}\longrightarrow1 1 ⟶ G ′ ⟶ G ⟶ Z / p Z ⟶ 1 , in which case h h h is the image of the fundamental class in H 1 ( Z / p Z ; Z / p Z ) H^1(\mathbb{Z}/p\mathbb{Z};\mathbb{Z}/p\mathbb{Z}) H 1 ( Z / p Z ; Z / p Z ) — the element h h h is doing double duty here. Then they show that if z ∈ H ∗ ( G ; Z / p Z ) z\in H^*(G;\mathbb{Z}/p\mathbb{Z}) z ∈ H ∗ ( G ; Z / p Z ) restricts to zero in H ∗ ( G ′ ; Z / p Z ) H^*(G';\mathbb{Z}/p\mathbb{Z}) H ∗ ( G ′ ; Z / p Z ) , then z 2 z^2 z 2 is in the ideal generated by β ( h ) \beta(h) β ( h ) , where β \beta β is the Bockstein homomorphism.
In our setup we start with a Hopf algebra extension C ⟶ B ⟶ D ( x ) C\longrightarrow B\longrightarrow D(x) C ⟶ B ⟶ D ( x ) and we choose h h h to be the image in H 1 ( B ) H^1(B) H 1 ( B ) of any nonzero element in H 1 ( D ( x ) ) H^1(D(x)) H 1 ( D ( x )) . Given this setup, their proof works.
Their proof works for (b) also, or alternatively, their Remark on p. 317 applies. Indeed, according to [20 Lemma 1.2.15(a)] , when we have an extension C ⟶ B ⟶ Λ ( x ) C\longrightarrow B\longrightarrow\Lambda(x) C ⟶ B ⟶ Λ ( x ) , then we get a long exact sequence
⋯ ⟶ H ∗ ( B ) → ⋅ a H ∗ ( B ) → i ∗ H ∗ ( C ) ⟶ ⋯ , \cdots\longrightarrow H^*(B)\xrightarrow{\cdot a}H^*(B)\xrightarrow{i^*}H^*(C)\longrightarrow\cdots, ⋯ ⟶ H ∗ ( B ) ⋅ a H ∗ ( B ) i ∗ H ∗ ( C ) ⟶ ⋯ ,
where i ∗ i^* i ∗ is the restriction map. So z z z is in the kernel of i ∗ i^* i ∗ if and only if it is in the image of multiplication by a a a .
See also [2 Theorem 5.2.1] for a different proof of the group theory result which also translates well to this situation; see [8 Proposition 2.4] for a version for Hopf algebras, but only in characteristic two; and see also [18 Lemma 3.2] for a closely related result.
The following is an odd primary version of a result of W. H. Lin [8 Theorem 3.1] and our proof is modeled on Lin’s. His result is stronger: He shows that when p = 2 p=2 p = 2 , the analog of b t b n k b_tb_{nk} b t b nk is nilpotent if k ≥ t k\geq t k ≥ t , whereas we need to assume that k ≥ s + t + 1 k\geq s+t+1 k ≥ s + t + 1 . This discrepancy is at least partly unavoidable: Wilkerson showed that at odd primes, b 10 b 21 b_{10}b_{21} b 10 b 21 is non-nilpotent (in the setting of the lemma), but there are cases which are not covered by the lemma and where the nilpotence of b t b n k b_tb_{nk} b t b nk is not known. We discuss this in § 7 .
Lemma 2.5 (Nilpotence lemma). Fix m ≤ ∞ m\leq\infty m ≤ ∞ and let B B B be a sub-Hopf algebra of A < m A_{<m} A < m with graded dual B ∨ B^\vee B ∨ . Fix positive integers t t t and n n n with t < n t<n t < n and t + n < m t+n<m t + n < m .
(a) Fix an integer s > 0 s>0 s > 0 and assume that ξ t p s = 0 \xi_{t}^{p^s}=0 ξ t p s = 0 in B ∨ B^\vee B ∨ for i < t i<t i < t , and assume that ξ t p s ≠ 0 \xi_{t}^{p^s}\neq0 ξ t p s = 0 while ξ t p s + 1 = 0 \xi_{t}^{p^{s+1}}=0 ξ t p s + 1 = 0 . Fix k ≥ s + t + 1 k\geq s+t+1 k ≥ s + t + 1 , and assume that ξ i p k = 0 \xi_i^{p^k}=0 ξ i p k = 0 for i < n i<n i < n . Finally, assume that ξ n p k ≠ 0 \xi_n^{p^k}\neq0 ξ n p k = 0 . Then b t b n k b_tb_{nk} b t b nk is nilpotent in H ∗ ( B ) H^*(B) H ∗ ( B ) .
(b) Assume that τ i = 0 \tau_i=0 τ i = 0 in B ∨ B^\vee B ∨ for i < t i<t i < t , and assume that τ t ≠ 0 \tau_t\neq0 τ t = 0 . Fix k ≥ t k\geq t k ≥ t , and assume that ξ i p k = 0 \xi_i^{p^k}=0 ξ i p k = 0 for i < n i<n i < n . Finally, assume that ξ n p k ≠ 0 \xi_n^{p^k}\neq0 ξ n p k = 0 . Then a t b n k a_tb_{nk} a t b nk is nilpotent in H ∗ ( B ) H^*(B) H ∗ ( B ) .
Here is a sketch of the polynomial part of the profile function of B B B in part (a), where terms in gray are zero in B ∨ B^\vee B ∨ :
Figure 1.
Proof. (a) Note that the assumptions guarantee that ξ t p s \xi_t^{p^s} ξ t p s and ξ n p k \xi_n^{p^k} ξ n p k are primitive, so that b t s b_{ts} b t s and b n k b_{nk} b nk are elements of H 2 ( B ) H^2(B) H 2 ( B ) . Also, the hypothesis that t + n < m t+n<m t + n < m , together with the fact that B B B is a sub-Hopf algebra of A < m A^{<m} A < m , guarantees that ξ t + n ≠ 0 ∈ B ∨ \xi_{t+n}\ne0\in B^\vee ξ t + n = 0 ∈ B ∨ . This is easiest to see from the dual point of view: if ξ t + n = 0 \xi_{t+n}=0 ξ t + n = 0 , then Δ ( ξ t + n ) = 0 \Delta(\xi_{t+n})=0 Δ ( ξ t + n ) = 0 , but Δ ( ξ t + n ) \Delta(\xi_{t+n}) Δ ( ξ t + n ) contains a term ξ n p k ⊗ ξ t \xi_n^{p^k}\otimes\xi_t ξ n p k ⊗ ξ t , and this term is nonzero in B ∨ B^\vee B ∨ . Indeed, after raising everything to the p s p^s p s power, this is the only term in the reduced coproduct on ξ t + n p s \xi_{t+n}^{p^s} ξ t + n p s : we have
ξ t + n p s ⟼ ξ n p s + t ⊗ ξ t p s . \xi_{t+n}^{p^s}\longmapsto\xi_n^{p^{s+t}}\otimes\xi_t^{p^s}. ξ t + n p s ⟼ ξ n p s + t ⊗ ξ t p s .
We proceed by induction, first considering the case when k = s + t + 1 k=s+t+1 k = s + t + 1 . We note that ξ n p s + t \xi_n^{p^{s+t}} ξ n p s + t and ξ t p s \xi_t^{p^s} ξ t p s are both primitive in B ∨ B^\vee B ∨ , so both h n , s + t = [ ξ n p s + t ] h_{n,s+t}=[\xi_n^{p^{s+t}}] h n , s + t = [ ξ n p s + t ] and h t s = [ ξ t p s ] h_{ts}=[\xi_t^{p^s}] h t s = [ ξ t p s ] are cohomology classes. This is clear for ξ t p s \xi_t^{p^s} ξ t p s . For ξ n p s + t \xi_n^{p^{s+t}} ξ n p s + t , its reduced coproduct is
∑ i = t n − t ξ n − i p s + t + i ⊗ ξ i p s + t . \sum_{i=t}^{n-t}\xi_{n-i}^{p^{s+t+i}}\otimes\xi_i^{p^{s+t}}. i = t ∑ n − t ξ n − i p s + t + i ⊗ ξ i p s + t .
We are assuming that ξ j p s = ξ j p s + t + 1 = 0 \xi_j^{p^s}=\xi_j^{p^{s+t+1}}=0 ξ j p s = ξ j p s + t + 1 = 0 for all j < n j<n j < n , so the first tensor factor in each summand is zero. Thus the coproduct on ξ t + n p s \xi_{t+n}^{p^s} ξ t + n p s produces a relation in cohomology:
h n , s + t h t s = 0. h_{n,s+t}h_{ts}=0. h n , s + t h t s = 0.
Applying Steenrod operations to this relation gives the following — we label each line with the operation being applied:
β P ~ 0 : b n , s + t h t , s + 1 − h n , s + t + 1 b t s = 0 , \beta\widetilde{P}^{0}: \quad b_{n,s+t}h_{t,s+1}-h_{n,s+t+1}b_{ts}=0, β P 0 : b n , s + t h t , s + 1 − h n , s + t + 1 b t s = 0 ,
Note that the first term is zero since ξ t p s + 1 = 0 \xi_t^{p^{s+1}}=0 ξ t p s + 1 = 0 .
β P ~ 1 : − b n , s + t + 1 b t s p = 0. \beta\widetilde{P}^{1}: \quad-b_{n,s+t+1}b_{ts}^{p}=0. β P 1 : − b n , s + t + 1 b t s p = 0.
This finishes the case when k = s + t + 1 k=s+t+1 k = s + t + 1 .
The same argument does not immediately apply if k > s + t + 1 k>s+t+1 k > s + t + 1 : we would again want to consider the coproduct on ξ t + n p s \xi_{t+n}^{p^s} ξ t + n p s , but the resulting tensor factors need not be primitive and so would not correspond to cohomology classes, so we would not be able to apply Steenrod operations to them. So we first work in a smaller Hopf algebra in which the terms are primitive, and then we enlarge the Hopf algebra.
The inductive hypothesis is the following: for each k ′ k' k ′ with s + t + 1 ≤ k ′ ≤ k s+t+1\le k'\le k s + t + 1 ≤ k ′ ≤ k , in the cohomology of any Hopf algebra whose dual satisfies (a) ξ i p s = 0 \xi_i^{p^s}=0 ξ i p s = 0 for i < t i<t i < t ,
(b) ξ t p s ≠ 0 \xi_t^{p^s}\ne0 ξ t p s = 0 , ξ t p s + 1 = 0 \xi_t^{p^{s+1}}=0 ξ t p s + 1 = 0 ,
(c) ξ n p k ≠ 0 \xi_n^{p^k}\ne0 ξ n p k = 0 ,
(d) ξ i p k ′ = 0 \xi_i^{p^{k'}}=0 ξ i p k ′ = 0 for t < i < n t<i<n t < i < n ,
the element b t 0 b n , k ′ b_{t0}b_{n,k'} b t 0 b n , k ′ is nilpotent. We have verified this when k = s + t + 1 k=s+t+1 k = s + t + 1 , starting the induction. Note also that the argument that we give is independent of n n n , so when we apply the inductive hypothesis, we can (and will) apply it with different values of n n n .
For the inductive step we first work in a Hopf algebra C C C whose dual satisfies (a), (b), and (c), together with
(d′) ξ i p s + t ≠ 0 \xi_i^{p^{s+t}}\ne0 ξ i p s + t = 0 , ξ i p s + t + 1 = 0 \xi_i^{p^{s+t+1}}=0 ξ i p s + t + 1 = 0 for t < i < n t<i<n t < i < n .
We will show that b t s b n k b_{ts}b_{nk} b t s b nk is nilpotent in H ∗ ( C ) H^*(C) H ∗ ( C ) ; after that, we will have to modify C C C to get a Hopf algebra that sets us up for the next step in the induction. Regarding hypothesis (d′) and in particular the requirement that ξ i p s + t ≠ 0 \xi_i^{p^{s+t}}\ne0 ξ i p s + t = 0 : any Hopf algebra C ′ C' C ′ whose dual satisfies (a), (b), (c), and ξ i p s + t + 1 = 0 \xi_i^{p^{s+t+1}}=0 ξ i p s + t + 1 = 0 for t < i < n t<i<n t < i < n will be a sub-Hopf algebra of C C C ; thus as long as b t s b n k b_{ts}b_{nk} b t s b nk is nilpotent in H ∗ ( C ) H^*(C) H ∗ ( C ) , it will also be nilpotent in H ∗ ( C ′ ) H^*(C') H ∗ ( C ′ ) via the restriction map.
The reduced coproduct in C C C on ξ s + t + n p s \xi_{s+t+n}^{p^s} ξ s + t + n p s is
∑ i = 1 n + t − i ξ s + t + n − i p s ⊗ ξ i p s . \sum_{i=1}^{n+t-i}\xi_{s+t+n-i}^{p^s}\otimes\xi_i^{p^s}. i = 1 ∑ n + t − i ξ s + t + n − i p s ⊗ ξ i p s .
Since ξ i p s = 0 \xi_i^{p^s}=0 ξ i p s = 0 for i < t i<t i < t , the lower limit on the sum can be changed to t t t . Similarly, since ξ n + t − i p s = 0 \xi_{n+t-i}^{p^s}=0 ξ n + t − i p s = 0 when n + t − i < t n+t-i<t n + t − i < t , then the upper limit can be changed to n n n : the reduced coproduct is
∑ i = t n ξ s + t + n − i p s ⊗ ξ i p s . \sum_{i=t}^{n}\xi_{s+t+n-i}^{p^s}\otimes\xi_i^{p^s}. i = t ∑ n ξ s + t + n − i p s ⊗ ξ i p s .
Since ξ j p s + t + 1 = 0 \xi_j^{p^{s+t+1}}=0 ξ j p s + t + 1 = 0 when t < j < n t<j<n t < j < n , the i = t i=t i = t term will be the only one present: the coproduct is ξ n p s + t ⊗ ξ t p s \xi_n^{p^{s+t}}\otimes\xi_t^{p^s} ξ n p s + t ⊗ ξ t p s . Both tensor factors here are primitive, so we get a relation in cohomology: h n , s + t h t s = 0 h_{n,s+t}h_{ts}=0 h n , s + t h t s = 0 . We can then apply Steenrod operations as above to get the desired result:
β ~ P ~ 0 : b n , s + t h t , s + 1 − h n , s + t + 1 b t s = 0 , \widetilde{\beta}\widetilde{P}^{0}: \quad b_{n,s+t}h_{t,s+1}-h_{n,s+t+1}b_{ts}=0, β P 0 : b n , s + t h t , s + 1 − h n , s + t + 1 b t s = 0 ,
Note that the first term is zero since ξ t p s + 1 = 0 \xi_t^{p^{s+1}}=0 ξ t p s + 1 = 0 .
β ~ P ~ 1 : − b n , s + t + 1 b t s p = 0 , \widetilde{\beta}\widetilde{P}^{1}: \quad-b_{n,s+t+1}b_{ts}^{p}=0, β P 1 : − b n , s + t + 1 b t s p = 0 ,
P ~ p : − b n , s + t + 2 b t s p 2 = 0 , \widetilde{P}^{p}: \quad-b_{n,s+t+2}b_{ts}^{p^2}=0, P p : − b n , s + t + 2 b t s p 2 = 0 ,
P ~ p j − 1 : − b n , s + t + j b t s p j = 0. \widetilde{P}^{p^{j-1}}: \quad-b_{n,s+t+j}b_{ts}^{p^j}=0. P p j − 1 : − b n , s + t + j b t s p j = 0.
So by choosing an appropriate j j j , we see that b t s b n k b_{ts}b_{nk} b t s b nk is nilpotent in H ∗ ( C ) H^*(C) H ∗ ( C ) .
Now we need to get a nilpotence result over a different Hopf algebra, one whose dual satisfies (a), (b), and (c) above, along with
(d′′) ξ i p k = 0 \xi_i^{p^k}=0 ξ i p k = 0 for t < i < n t<i<n t < i < n . As noted above, we may assume that ξ i p s + t ≠ 0 \xi_i^{p^{s+t}} \ne0 ξ i p s + t = 0 for t < i < n t < i < n t < i < n , and we can build such a Hopf algebra from C C C by adding in one “box” ξ i p j \xi_i^{p^j} ξ i p j at a time: we assume that the result holds over a Hopf algebra C ′ C' C ′ and we want to show that it continues to hold over C ′ ′ C'' C ′′ , where there is a Hopf algebra extension
C ′ ⟶ C ′ ′ ⟶ D ( P i j ) , C' \longrightarrow C'' \longrightarrow D(P_i^j), C ′ ⟶ C ′′ ⟶ D ( P i j ) ,
where t < i < n t < i < n t < i < n and s + t + 1 ≤ j < k s+t+1 \le j < k s + t + 1 ≤ j < k . (Recall that P i j P_i^j P i j is dual to ξ i p j \xi_i^{p^j} ξ i p j .) We use the Quillen-Venkov lemma and our inductive hypothesis: given z = b t s b n k ∈ H ∗ ( C ′ ′ ) z=b_{ts}b_{nk}\in H^*(C'') z = b t s b nk ∈ H ∗ ( C ′′ ) , since z M ↦ 0 z^M\mapsto0 z M ↦ 0 in H ∗ ( C ′ ) H^*(C') H ∗ ( C ′ ) , then z 2 M z^{2M} z 2 M is in the ideal generated by b i j b_{ij} b ij : we have
b t s 2 M b n k 2 M = b i j y b_{ts}^{2M}b_{nk}^{2M}=b_{ij}y b t s 2 M b nk 2 M = b ij y
for some y y y . Now because s + t + 1 ≤ j < k s+t+1\le j<k s + t + 1 ≤ j < k , by induction (using a lower value for n n n ) we know that b t s b i j b_{ts}b_{ij} b t s b ij is nilpotent, so after multiplying both sides by b t s b_{ts} b t s and raising to a large enough power, we see that the right-hand side is nilpotent, and hence so is the left. This completes the inductive step.
(b) This proof is very similar. First assume that ξ i p t = 0 \xi_i^{p^t}=0 ξ i p t = 0 when i < n i<n i < n . Then we have the reduced coproduct τ t + n ↦ ξ n p t ⊗ τ t \tau_{t+n}\mapsto\xi_n^{p^t}\otimes\tau_t τ t + n ↦ ξ n p t ⊗ τ t , which gives us the relation h n t a t = 0 h_{nt}a_t=0 h n t a t = 0 in cohomology. If we apply the operation β P p t \beta P^{p^t} β P p t , we get b n t a t p = 0 b_{nt}a_t^p=0 b n t a t p = 0 , and then applying further operations, we get b n , t + i a t p i + 1 = 0 b_{n,t+i}a_t^{p^{i+1}}=0 b n , t + i a t p i + 1 = 0 for all i ≥ 0 i\ge0 i ≥ 0 .
Now we use the Quillen-Venkov lemma to get from the hypothesis that ξ i p t = 0 \xi_i^{p^t}=0 ξ i p t = 0 to ξ i p k = 0 \xi_i^{p^k}=0 ξ i p k = 0 for i < n i<n i < n . As in part (a), we will have Hopf algebra extensions of the form
C ′ ⟶ C ′ ′ ⟶ D ( P i j ) C'\longrightarrow C''\longrightarrow D(P_i^j) C ′ ⟶ C ′′ ⟶ D ( P i j )
where i < n i<n i < n and t ≤ j < k t\le j<k t ≤ j < k , and we assume that z = b n k a t ∈ H ∗ ( C ′ ′ ) z=b_{nk}a_t\in H^*(C'') z = b nk a t ∈ H ∗ ( C ′′ ) maps to a nilpotent element in H ∗ ( C ′ ) H^*(C') H ∗ ( C ′ ) , say z M ↦ 0 z^M\mapsto0 z M ↦ 0 . Then in H ∗ ( C ′ ′ ) H^*(C'') H ∗ ( C ′′ ) , the Quillen-Venkov lemma tells us that z 2 M = b n k 2 M a t 2 M = b i j y z^{2M}=b_{nk}^{2M}a_t^{2M}=b_{ij}y z 2 M = b nk 2 M a t 2 M = b ij y for some y y y . By induction we know that b i j a t b_{ij}a_t b ij a t is nilpotent, and that finishes the proof.
Part (a) of this lemma can be improved. We have stated the version that we need, but it is not too hard to get the same conclusion with slightly weaker hypotheses:
Lemma 2.6. Fix m ≤ ∞ m\le\infty m ≤ ∞ and let B B B be a sub-Hopf algebra of A < m A_{<m} A < m with graded dual B ∨ B^\vee B ∨ . Fix positive integers t t t and n n n with t < n t<n t < n and t + n < m t+n<m t + n < m . Fix an integer s ≥ 0 s\ge0 s ≥ 0 . Assume that ξ t p s ≠ 0 \xi_t^{p^s}\ne0 ξ t p s = 0 while ξ t p s + 1 = 0 \xi_t^{p^{s+1}}=0 ξ t p s + 1 = 0 . Fix k ≥ s + t + 1 k\ge s+t+1 k ≥ s + t + 1 , and assume that ξ i p k = 0 \xi_i^{p^k}=0 ξ i p k = 0 for i < n i<n i < n . Assume that for i < t i<t i < t , ξ i p k − i = 0 \xi_i^{p^{k-i}}=0 ξ i p k − i = 0 . Assume that ξ t p s \xi_t^{p^s} ξ t p s is primitive in B B B . Finally, assume that ξ n p t ≠ 0 \xi_n^{p^t}\ne0 ξ n p t = 0 . Then b t s b n k b_{ts}b_{nk} b t s b nk is nilpotent in H ∗ ( B ) H^*(B) H ∗ ( B ) .
The difference is that for i < t i<t i < t , in the Nilpotence Lemma we assume that ξ i p s = 0 \xi_i^{p^s}=0 ξ i p s = 0 , while here we assume that ξ i p k − i = 0 \xi_i^{p^{k-i}}=0 ξ i p k − i = 0 . Since k ≥ s + t + 1 > s + i + 1 k\ge s+t+1>s+i+1 k ≥ s + t + 1 > s + i + 1 , then k − i ≥ s k-i\ge s k − i ≥ s , so this is a weaker assumption. The original assumption that ξ i p s = 0 \xi_i^{p^s}=0 ξ i p s = 0 implies that ξ i p s \xi_i^{p^s} ξ i p s is primitive, while that may not follow from this weaker assumption, so we add the primitivitity assumption explicitly here.
Sketch of proof. We use the Quillen-Venkov Lemma and the Nilpotence Lemma. Suppose that B B B is as in the statement, and let ( b 1 , b 2 , … ) (b_1,b_2,\ldots) ( b 1 , b 2 , … ) be its profile function. We define a sub-Hopf algebra C C C by setting c i = min ( b i , s ) c_i=\min(b_i,s) c i = min ( b i , s ) if i < t i<t i < t , c i = b i c_i=b_i c i = b i if i ≥ t i\ge t i ≥ t . Then C C C satisfies the hypotheses of the Nilpotence Lemma, so b t s b n k b_{ts}b_{nk} b t s b nk is nilpotent in the cohomology of C C C . We can then build B B B from C C C by using induction and adding in one box ξ i p j \xi_i^{p^j} ξ i p j at a time. That is, as in the above proof, in the inductive step we have a Hopf algebra extension
C ′ ⟶ C ′ ′ ⟶ D ( P i j ) , C'\longrightarrow C''\longrightarrow D(P_i^j), C ′ ⟶ C ′′ ⟶ D ( P i j ) ,
where i < t i<t i < t , j < k − i j<k-i j < k − i , so i + j + 1 ≤ k i+j+1\le k i + j + 1 ≤ k . We assume that b t s b n k b_{ts}b_{nk} b t s b nk is nilpotence over C ′ C' C ′ , so by the Quillen-Venkov Lemma, some power of b t s b n k b_{ts}b_{nk} b t s b nk is a multiple of b i j b_{ij} b ij . Since k ≥ i + j + 1 k\ge i+j+1 k ≥ i + j + 1 , b i j b n k b_{ij}b_{nk} b ij b nk is nilpotent, so b t s b n k b_{ts}b_{nk} b t s b nk remains nilpotent over C ′ ′ C'' C ′′ .
Reducing to nilpotence of particular cohomology classes
The goal of this section is to prove Theorem 3.1 , the point of which is to reduce the questions of detection of nilpotence and existence of an F F F -isomorphism to specific cohomology calculations: are certain cohomology classes nilpotent? We need a bit of setup and notation first, so suppose that we have Hopf algebras C ⊆ H ⊆ A C \subseteq H \subseteq A C ⊆ H ⊆ A , and suppose that we have a positive integer n n n . Suppose further that C C C and H H H agree on the exterior part of the Steenrod algebra: we may write their profile functions as
C : ( ( c 1 , c 2 , c 3 , … ) ; ( e 0 , e 1 , e 2 , … ) ) C : ((c_1,c_2,c_3,\ldots);(e_0,e_1,e_2,\ldots)) C : (( c 1 , c 2 , c 3 , … ) ; ( e 0 , e 1 , e 2 , … ))
H : ( ( h 1 , h 2 , h 3 , … ) ; ( e 0 , e 1 , e 2 , … ) ) . H : ((h_1,h_2,h_3,\ldots);(e_0,e_1,e_2,\ldots)). H : (( h 1 , h 2 , h 3 , … ) ; ( e 0 , e 1 , e 2 , … )) .
We write C n C_n C n for the sub-Hopf algebra of H H H with profile function
( ( c 1 , … , c n − 1 , h n , h n + 1 , … ) ; ( e 0 , e 1 , e 2 , … ) ) ((c_1,\ldots,c_{n-1},h_n,h_{n+1},\ldots);(e_0,e_1,e_2,\ldots)) (( c 1 , … , c n − 1 , h n , h n + 1 , … ) ; ( e 0 , e 1 , e 2 , … ))
(Note that C n C_n C n is not perfect notation since the algebra depends on H H H in addition to C C C and n n n , but we think of H H H as being the ambient Hopf algebra and suppress it from the notation.) Here is a picture of the polynomial part of the profile functions:
Figure 1 (PDF p. 12).
For example, we will apply this in the next section with H = A H=A H = A and C = B i C=Bi C = B i , in which case the algebra B i n Bi_n B i n has profile function
( ( 1 , 2 , 4 , … , n − 1 2 + 1 , ∞ , ∞ , ∞ , … ) ; ( 2 , 2 , 2 , … ) ) . ((1,2,4,\ldots,\frac{n-1}{2}+1,\infty,\infty,\infty,\ldots);(2,2,2,\ldots)). (( 1 , 2 , 4 , … , 2 n − 1 + 1 , ∞ , ∞ , ∞ , … ) ; ( 2 , 2 , 2 , … )) .
Theorem 3.1. Fix n ≥ 1 n \ge1 n ≥ 1 . Suppose that we have Hopf algebras C ⊆ H ⊆ A C \subseteq H \subseteq A C ⊆ H ⊆ A , with C C C agreeing with H H H on the exterior part of A A A . Suppose that C C C is normal in H H H , and define C n C_n C n as above. Write the profile function for C C C as ( ( c 1 , c 2 , … ) ; ( e 1 , e 2 , … ) ) ((c_1,c_2,\ldots);(e_1,e_2,\ldots)) (( c 1 , c 2 , … ) ; ( e 1 , e 2 , … )) . Suppose that for each n n n , either c n = ∞ c_n=\infty c n = ∞ , ξ n p c n = 0 \xi_n^{p^{c_n}}=0 ξ n p c n = 0 in C n C_n C n , or b n , c n b_{n,c_n} b n , c n is nilpotent in the cohomology of C n C_n C n . Then
(a) C C C detects nilpotence in Stable ( H ∨ ) \operatorname{Stable}(H^\vee) Stable ( H ∨ ) .
(b) The map H ∗ ( H ) → ( H ∗ ( C ) ) H H^*(H) \to(H^*(C))^H H ∗ ( H ) → ( H ∗ ( C ) ) H is an F F F -isomorphism.
(We could write (b) in terms of ( H ∗ ( C ) ) H / / C (H^*(C))^{H//C} ( H ∗ ( C ) ) H // C , but of course that is equal to ( H ∗ ( C ) ) H (H^*(C))^H ( H ∗ ( C ) ) H . Here and below we may use the shorter version for more compact notation.)
Note that when c n < ∞ c_n<\infty c n < ∞ , then ξ n p c n \xi_n^{p^{c_n}} ξ n p c n will be zero in C n C_n C n if and only if c n = h n c_n=h_n c n = h n . Also note that if ξ n p c n \xi_n^{p^{c_n}} ξ n p c n is nonzero in C n C_n C n , then it will be primitive, so b n , c n b_{n,c_n} b n , c n will be a cohomology class. (Why is it primitive? ξ n p c n \xi_n^{p^{c_n}} ξ n p c n is zero in C ∨ C^\vee C ∨ , so its reduced coproduct is zero in C ∨ ⊗ C ∨ C^\vee\otimes C^\vee C ∨ ⊗ C ∨ . That coproduct is still zero in C n ∨ ⊗ C ∨ C_n^\vee\otimes C^\vee C n ∨ ⊗ C ∨ , which means that ξ n p c n \xi_n^{p^{c_n}} ξ n p c n is primitive there.)
We also point out that if b n , c n b_{n,c_n} b n , c n is nilpotent, then so is b n , j b_{n,j} b n , j for any j ≥ c n j \ge c_n j ≥ c n , because the Steenrod operation P ~ 0 \widetilde{\mathrm{P}}^0 P 0 is an algebra map sending b n , i b_{n,i} b n , i to b n , i + 1 b_{n,i+1} b n , i + 1 for all i i i .
The proof of the theorem is almost identical to the proofs of [20 Theorems 5.1.2 and 5.1.5] when p = 2 p=2 p = 2 . We may assume that ξ n c n ≠ 0 \xi_n^{c_n}\ne0 ξ n c n = 0 in C n ∨ C_n^\vee C n ∨ . For each n n n , C n + 1 C_{n+1} C n + 1 is normal in C n C_n C n , and we have
C n / / C n + 1 ≅ D ( P n c n , P n c n + 1 , … ) . C_n//C_{n+1}\cong D(P_n^{c_n},P_n^{c_n+1},\ldots). C n // C n + 1 ≅ D ( P n c n , P n c n + 1 , … ) .
This has the structure of a C n C_n C n -module, and for any j ≥ c n j\ge c_n j ≥ c n , we have a finite-dimensional quotient C n C_n C n -module
W j = D ( P n c n , P n c n + 1 , … , P n j ) . W_j=D(P_n^{c_n},P_n^{c_n+1},\ldots,P_n^j). W j = D ( P n c n , P n c n + 1 , … , P n j ) .
It also makes sense to let W c n − 1 = F p W_{c_n-1}=\mathbb{F}_p W c n − 1 = F p : in the proof of the following lemma we will use the observation that each W j + 1 W_{j+1} W j + 1 can be built from W j W_j W j , and this process starts with W c n − 1 = F p W_{c_n-1}=\mathbb{F}_p W c n − 1 = F p . We abuse notation and also use the same notation W j W_j W j for the corresponding object in Stable ( C n ∨ ) \operatorname{Stable}(C_n^\vee) Stable ( C n ∨ ) .
Lemma 3.2. We work in the category Stable ( C n ∨ ) \operatorname{Stable}(C_n^\vee) Stable ( C n ∨ ) . Assume that for some j ≥ c n j\ge c_n j ≥ c n , the element b n j ∈ H ∗ ( C n ) b_{nj}\in H^*(C_n) b nj ∈ H ∗ ( C n ) is nilpotent. Then the object W j − 1 W_{j-1} W j − 1 is in the thick subcategory generated by W j W_j W j . So if b n j b_{nj} b nj is nilpotent for all j ≥ c n j\ge c_n j ≥ c n , then F p \mathbb{F}_p F p is in the thick subcategory generated by any W j W_j W j .
This lemma is where the nilpotence of b n j b_{nj} b nj is used.
Proof. By [20 Lemma 1.2.15(b)] , for any j ≥ c n j\ge c_n j ≥ c n there are two cofiber sequences in Stable ( C n ∨ ) \operatorname{Stable}(C_n^\vee) Stable ( C n ∨ ) :
Σ 2 , p ∣ ξ n c n ∣ W j − 1 → b n j W j − 1 ⟶ W ~ j , \Sigma^{2,p|\xi_n^{c_n}|}W_{j-1}\xrightarrow{b_{nj}}W_{j-1}\longrightarrow\widetilde{W}_j, Σ 2 , p ∣ ξ n c n ∣ W j − 1 b nj W j − 1 ⟶ W j ,
Σ 1 , ∣ ξ n c n ∣ W j ⟶ W ~ j ⟶ W j . \Sigma^{1,|\xi_n^{c_n}|}W_j\longrightarrow\widetilde{W}_j\longrightarrow W_j. Σ 1 , ∣ ξ n c n ∣ W j ⟶ W j ⟶ W j .
Therefore W ~ j \widetilde{W}_j W j is in the thick subcategory generated by W j W_j W j , and the nilpotence of b n j b_{nj} b nj tells us that W j − 1 W_{j-1} W j − 1 is in the thick subcategory generated by W ~ j \widetilde{W}_j W j .
Lemma 3.3. Fix j ≥ 0 j\ge0 j ≥ 0 and let B = D ( P n j , P n j + 1 , P n j + 2 , … ) = ( F p [ ξ n p j ] ) ∨ B=D(P_n^j,P_n^{j+1},P_n^{j+2},\ldots)=(\mathbb{F}_p[\xi_n^{p^j}])^\vee B = D ( P n j , P n j + 1 , P n j + 2 , … ) = ( F p [ ξ n p j ] ) ∨ . If M M M is a B B B -module which is zero below degree t t t , then H ∗ , ∗ ( B ; M ) H^{*,*}(B;M) H ∗ , ∗ ( B ; M ) has a vanishing plane: there is a number c c c , independent of M M M , so that H s , u ( B ; M ) = 0 H^{s,u}(B;M)=0 H s , u ( B ; M ) = 0 when u < p 2 ∣ ξ n p j ∣ s + t + c u<\frac{p}{2}|\xi_n^{p^j}|s+t+c u < 2 p ∣ ξ n p j ∣ s + t + c .
Proof. This is an easy computation. We know that H ∗ ( B ; F p ) ≅ Λ ( h n i : i ≥ j ) ⊗ F p [ b n i : i ≥ j ] H^*(B;\mathbb{F}_p)\cong\Lambda(h_{ni}:i\ge j)\otimes\mathbb{F}_p[b_{ni}:i\ge j] H ∗ ( B ; F p ) ≅ Λ ( h ni : i ≥ j ) ⊗ F p [ b ni : i ≥ j ] has a vanishing plane with a slope given by the element b n j b_{nj} b nj , namely m j : = p 2 ∣ ξ n p j ∣ m_j:=\frac{p}{2}|\xi_n^{p^j}| m j := 2 p ∣ ξ n p j ∣ . That is, H s , u ( B ; F p ) = 0 H^{s,u}(B;\mathbb{F}_p)=0 H s , u ( B ; F p ) = 0 when u < m j s + c u<m_js+c u < m j s + c for some intercept c c c . Therefore we can construct an injective resolution of F p \mathbb{F}_p F p , the connectivity of whose terms increases with this slope. Tensor this with M M M to get an injective resolution of M M M ; its terms will have connectivity increasing with the desired slope, and with intercept t + c t+c t + c .
Proof of Theorem 3.1 . We will omit some details here; see [20 Subsections 5.2.1–2] for full details. We think of building H H H from C C C using inclusions of Hopf algebras
C ⊆ ⋯ ⊆ C n + 1 ⊆ C n ⊆ ⋯ ⊆ C 2 ⊆ C 1 = H . C\subseteq\cdots\subseteq C_{n+1}\subseteq C_n\subseteq\cdots\subseteq C_2\subseteq C_1=H. C ⊆ ⋯ ⊆ C n + 1 ⊆ C n ⊆ ⋯ ⊆ C 2 ⊆ C 1 = H .
For the nilpotence part of the proof, we use standard techniques (omitted here) to reduce to the case of a connective ring spectrum R ∈ Stable ( H ∨ ) R\in\operatorname{Stable}(H^\vee) R ∈ Stable ( H ∨ ) and an element α ∈ H ∗ ( H ; R ) \alpha\in H^*(H;R) α ∈ H ∗ ( H ; R ) which restricts to zero in H ∗ ( C ; R ) H^*(C;R) H ∗ ( C ; R ) . Since C C C and C n C_n C n agree through a range of dimensions increasing with n n n , so do their cohomology algebras, so α \alpha α will also restrict to zero in H ∗ ( C n + 1 ; R ) H^*(C_{n+1};R) H ∗ ( C n + 1 ; R ) for n ≫ 0 n\gg0 n ≫ 0 . This starts an inductive process; the inductive step is to show that if α \alpha α restricts to zero over H ∗ ( C n + 1 ; R ) H^*(C_{n+1};R) H ∗ ( C n + 1 ; R ) , then it is nilpotent in H ∗ ( C n ; R ) H^*(C_n;R) H ∗ ( C n ; R ) .
Given the connective ring spectrum R R R and the Hopf algebra extension
C n + 1 ⟶ C n ⟶ C n / / C n + 1 , C_{n+1}\longrightarrow C_n\longrightarrow C_n//C_{n+1}, C n + 1 ⟶ C n ⟶ C n // C n + 1 ,
the Cartan-Eilenberg spectral sequence looks like
E 2 s , t , ∗ ( R ) ≅ H s , ∗ ( C n / / C n + 1 ; H t , ∗ ( C n + 1 ; R ) ) ⇒ H s + t , ∗ ( C n ; R ) . E_2^{s,t,*}(R)\cong H^{s,*}(C_n//C_{n+1};H^{t,*}(C_{n+1};R))\Rightarrow H^{s+t,*}(C_n;R). E 2 s , t , ∗ ( R ) ≅ H s , ∗ ( C n // C n + 1 ; H t , ∗ ( C n + 1 ; R )) ⇒ H s + t , ∗ ( C n ; R ) .
By [20 Proposition 1.4.9] , this may be viewed as an Adams spectral sequence, and [20 Proposition 5.2.2] tells us that we can get as nice a vanishing plane as we want at some page of this spectral sequence. We will get into the details, but first note that there are two cases here depending on whether any power of ξ n \xi_n ξ n is zero in the dual of the ambient Hopf algebra H H H : whether h n < ∞ h_n < \infty h n < ∞ or not. The case when h n < ∞ h_n < \infty h n < ∞ is easier, and we handle it later: for now we assume that h n = ∞ h_n = \infty h n = ∞ .
To get the vanishing plane when h n = ∞ h_n = \infty h n = ∞ : for any j ≥ c n j \ge c_n j ≥ c n , since C n + 1 C_{n+1} C n + 1 acts trivially on W j W_j W j , we have
E 2 s , t , ∗ ( R ∧ W j ) ≅ H s , ∗ ( C n / / C n + 1 ; H t , ∗ ( C n + 1 ; R ∧ W j ) ) ≅ H s , ∗ ( C n / / C n + 1 ; W j ∧ H t , ∗ ( C n + 1 ; R ) ) ≅ H s , ∗ ( ( F p [ ξ n p j + 1 ] ) ∨ ; H t , ∗ ( C n + 1 ; R ) ) . E_2^{s,t,*}(R \wedge W_j) \cong H^{s,*}(C_n//C_{n+1}; H^{t,*}(C_{n+1}; R \wedge W_j))
\cong H^{s,*}(C_n//C_{n+1}; W_j \wedge H^{t,*}(C_{n+1}; R))
\cong H^{s,*}((\mathbb{F}_p[\xi_n^{p^{j+1}}])^\vee; H^{t,*}(C_{n+1}; R)). E 2 s , t , ∗ ( R ∧ W j ) ≅ H s , ∗ ( C n // C n + 1 ; H t , ∗ ( C n + 1 ; R ∧ W j )) ≅ H s , ∗ ( C n // C n + 1 ; W j ∧ H t , ∗ ( C n + 1 ; R )) ≅ H s , ∗ (( F p [ ξ n p j + 1 ] ) ∨ ; H t , ∗ ( C n + 1 ; R )) .
where we have used a change-of-rings isomorphism at the end. By Lemma 3.3 above, with coefficients in a bounded below module, this cohomology has a vanishing plane of the form
E 2 s , t , u ( R ∧ W j ) = 0 when m j + 1 s + t − u > c E_2^{s,t,u}(R \wedge W_j) = 0 \quad\text{when} \quad m_{j+1}s + t - u > c E 2 s , t , u ( R ∧ W j ) = 0 when m j + 1 s + t − u > c
for some c c c , where as above, m j + 1 = p 2 ∣ ξ n p j + 1 ∣ m_{j+1} = \frac{p}{2}\lvert\xi_n^{p^{j+1}}\rvert m j + 1 = 2 p ∣ ξ n p j + 1 ∣ . The lemma is stated for modules, but the existence of such a vanishing plane extends to suitably connective objects in Stable( H ∨ ) (H^\vee) ( H ∨ ) , so since R R R is a connective ring spectrum, we find that E 2 s , t , u ( R ∧ W j ) E_2^{s,t,u}(R \wedge W_j) E 2 s , t , u ( R ∧ W j ) has a good vanishing plane: it will be zero when m j + 1 s + t − u > c m_{j+1}s + t - u > c m j + 1 s + t − u > c , for some c c c . Hence by Lemma 3.2 and genericity of vanishing planes in the Adams spectral sequence — see [6 ] — for some c ′ c' c ′ and some r r r , we see that E r s , t , u ( R ) E_r^{s,t,u}(R) E r s , t , u ( R ) will be zero when m j + 1 s + t − u > c ′ m_{j+1}s + t - u > c' m j + 1 s + t − u > c ′ .
If α ∈ H ∗ ( C n ; R ) \alpha\in H^*(C_n; R) α ∈ H ∗ ( C n ; R ) restricts to zero in H ∗ ( C n + 1 ; R ) H^*(C_{n+1}; R) H ∗ ( C n + 1 ; R ) , it will not lie on the s = 0 s = 0 s = 0 edge of the spectral sequence: it will be represented by a class a ∈ E 2 s , t , u a \in E_2^{s,t,u} a ∈ E 2 s , t , u with s > 0 s > 0 s > 0 . Now if we choose a large enough j j j so as to obtain a low enough vanishing plane, some power of a a a will lie above the vanishing plane and hence will be zero; therefore α \alpha α will be nilpotent in the abutment, as desired.
In the case when h n < ∞ h_n < \infty h n < ∞ , then
C n / / C n + 1 ≅ D ( P n c n , P n c n + 1 , … , P n h n − 1 ) = W h n − 1 . C_n//C_{n+1} \cong D(P_n^{c_n}, P_n^{c_{n+1}}, \ldots, P_n^{h_n-1}) = W_{h_n-1}. C n // C n + 1 ≅ D ( P n c n , P n c n + 1 , … , P n h n − 1 ) = W h n − 1 .
The spectral sequence has
E 2 s , t , ∗ ( R ∧ W h n − 1 ) ≅ H s , ∗ ( C n / / C n + 1 ; W h n − 1 ∧ H t , ∗ ( C n + 1 ; R ) ) . E_2^{s,t,*}(R \wedge W_{h_n-1}) \cong H^{s,*}(C_n//C_{n+1}; W_{h_n-1} \wedge H^{t,*}(C_{n+1}; R)). E 2 s , t , ∗ ( R ∧ W h n − 1 ) ≅ H s , ∗ ( C n // C n + 1 ; W h n − 1 ∧ H t , ∗ ( C n + 1 ; R )) .
The coefficient module is free over C n / / C n + 1 C_n//C_{n+1} C n // C n + 1 , so E 2 s , t , ∗ ( R ∧ W h n − 1 ) = 0 E_2^{s,t,*}(R \wedge W_{h_n-1}) = 0 E 2 s , t , ∗ ( R ∧ W h n − 1 ) = 0 when s > 0 s > 0 s > 0 . This is a very strong vanishing plane! By Lemma 3.2 and the genericity of vanishing planes, for some c ′ c' c ′ and some r r r , we see that E r s , t , u ( R ) E_r^{s,t,u}(R) E r s , t , u ( R ) will be zero when s > c ′ s > c' s > c ′ . The rest of the proof proceeds as above. (Alternatively, one could perhaps state and prove a version of the Quillen-Venkov lemma with coefficients in a ring spectrum in Stable( C ∨ ) (C^\vee) ( C ∨ ) , and the deal with one “box” ξ n p n \xi_n^{p^n} ξ n p n at a time.)
For the F F F -isomorphism part of the proof, it remains to show that if y y y is in the codomain of the restriction map ι ∗ \iota^* ι ∗ , then some p p p th power of y y y is in the image. It is easy to see that if y ∈ H ∗ ( C ) y \in H^*(C) y ∈ H ∗ ( C ) , then y y y is in the image of the restriction map H ∗ ( C n ) → H ∗ ( C ) H^*(C_n) \to H^*(C) H ∗ ( C n ) → H ∗ ( C ) for n ≫ 0 n \gg0 n ≫ 0 , so again, the issue is going from C n + 1 C_{n+1} C n + 1 to C n C_n C n : if y y y is in the invariant subalgebra ( H ∗ ( C n + 1 ) ) C n (H^*(C_{n+1}))^{C_n} ( H ∗ ( C n + 1 ) ) C n , then we need to show that some power of y y y is in the image of the restriction map ι ∗ : H ∗ ( C n ) → ( H ∗ ( C n + 1 ) ) C n \iota^*: H^*(C_n) \to(H^*(C_{n+1}))^{C_n} ι ∗ : H ∗ ( C n ) → ( H ∗ ( C n + 1 ) ) C n . We use the same spectral sequence. The first thing to note is that
E 2 0 , t , u ( F p ) ≅ H 0 , ∗ ( C n / / C n + 1 ; H t , ∗ ( C n + 1 ) ) ≅ ( H t , ∗ ( C n + 1 ) ) C n / / C n + 1 . E_2^{0,t,u}(\mathbb{F}_p) \cong H^{0,*}(C_n//C_{n+1}; H^{t,*}(C_{n+1})) \cong(H^{t,*}(C_{n+1}))^{C_n//C_{n+1}}. E 2 0 , t , u ( F p ) ≅ H 0 , ∗ ( C n // C n + 1 ; H t , ∗ ( C n + 1 )) ≅ ( H t , ∗ ( C n + 1 ) ) C n // C n + 1 .
This is where the invariants come into play. The element y y y is represented by a class in E 2 0 , t , u E_2^{0,t,u} E 2 0 , t , u , and using Steenrod operations acting on the spectral sequence, one can identify the possible locations of the images of y y y and its p p p th powers under the differentials — see [20 Proposition 1.4.11] , which says that y p n y^{p^n} y p n survives to E p n + 1 E_{p^n+1} E p n + 1 . The possible images of the differentials of the classes y p n y^{p^n} y p n lie on a line with a positive slope in the spectral sequence. That line will eventually go above a low enough vanishing plane, so we see that some y p n y^{p^n} y p n will be a permanent cycle. ∙იდ бзиа
Since P P P is a sub-Hopf algebra of A A A , we can of course apply the previous result to sub-Hopf algebras of P P P . In fact it suffices to work with sub-Hopf algebras of P P P , as we now demonstrate. Fix a sub-Hopf algebra C ⊆ A C \subseteq A C ⊆ A . Just as we can view P P P as both a sub- and quotient Hopf algebra of A A A , we can view C ∩ P C \cap P C ∩ P as both a sub- and quotient Hopf algebra of C C C .
Proposition 3.4. Given a sub-Hopf algebra C ⊆ A C \subseteq A C ⊆ A , suppose that ξ n p k \xi_n^{p^k} ξ n p k is primitive in C ∨ C^\vee C ∨ , and hence in ( C ∩ P ) ∨ (C \cap P)^\vee ( C ∩ P ) ∨ . If b n k ∈ H ∗ ( C ∩ P ) b_{nk} \in H^*(C \cap P) b nk ∈ H ∗ ( C ∩ P ) is nilpotent, so is b n k ∈ H ∗ ( C ) b_{nk} \in H^*(C) b nk ∈ H ∗ ( C ) .
Proof. The quotient map C → C ∩ P C \to C \cap P C → C ∩ P induces an algebra map H ∗ ( C ∩ P ) → H ∗ ( C ) H^*(C \cap P) \to H^*(C) H ∗ ( C ∩ P ) → H ∗ ( C ) , sending b n k b_{nk} b nk to b n k b_{nk} b nk .
So given C ⊆ H ⊆ A C \subseteq H \subseteq A C ⊆ H ⊆ A as in Theorem 3.1 , it suffices to check the nilpotence of each b n , c n ∈ H ∗ ( ( C ∩ P ) n ) = H ∗ ( C n ∩ P ) b_{n,c_n} \in H^*((C \cap P)_n) = H^*(C_n \cap P) b n , c n ∈ H ∗ (( C ∩ P ) n ) = H ∗ ( C n ∩ P ) . If we can do that, we may conclude that
C C C detects nilpotence in Stable ( H ∨ ) \operatorname{Stable}(H^\vee) Stable ( H ∨ ) ,
C ∩ P C \cap P C ∩ P detects nilpotence in Stable ( ( H ∩ P ) ∨ ) \operatorname{Stable}((H \cap P)^\vee) Stable (( H ∩ P ) ∨ ) ,
the map H ∗ ( H ) → H ∗ ( C ) H H^*(H) \to H^*(C)^H H ∗ ( H ) → H ∗ ( C ) H is an F F F -isomorphism, and
the map H ∗ ( H ∩ P ) → H ∗ ( C ∩ P ) H ∩ P H^*(H \cap P) \to H^*(C \cap P)^{H \cap P} H ∗ ( H ∩ P ) → H ∗ ( C ∩ P ) H ∩ P is an F F F -isomorphism.
Restricting to the algebra Binom
To prove Theorems A and B, we will apply Theorem 3.1 twice, to each of the inclusions D ⊆ B i D \subseteq Bi D ⊆ B i and B i ⊆ A Bi \subseteq A B i ⊆ A , or more generally given C ⊆ A C \subseteq A C ⊆ A , we apply the theorem to C ∩ D ⊆ C ∩ B i C \cap D \subseteq C \cap Bi C ∩ D ⊆ C ∩ B i and C ∩ B i ⊆ C C \cap Bi \subseteq C C ∩ B i ⊆ C , to get F F F -isomorphisms
H ∗ ( C ) → ( H ∗ ( C ∩ B i ) ) C → ( ( H ∗ ( C ∩ D ) ) C ∩ B i ) C ≅ ( H ∗ ( C ∩ D ) ) C H^*(C) \to\left(H^*(C \cap Bi)\right)^C \to\left(\left(H^*(C \cap D)\right)^{C \cap Bi}\right)^C \cong\left(H^*(C \cap D)\right)^C H ∗ ( C ) → ( H ∗ ( C ∩ B i ) ) C → ( ( H ∗ ( C ∩ D ) ) C ∩ B i ) C ≅ ( H ∗ ( C ∩ D ) ) C
and similarly for detection of nilpotence. This means that we have to study the nilpotence of classes b n j b_{nj} b nj in the cohomology algebras of certain Hopf algebras. We focus on the inclusion C ∩ B i ⊆ C C \cap Bi \subseteq C C ∩ B i ⊆ C in this section; in the next section we will address C ∩ D ⊆ C ∩ B i C \cap D \subseteq C \cap Bi C ∩ D ⊆ C ∩ B i .
Proposition 3.4 lets us focus on sub-Hopf algebras of P P P rather than A A A . The key to making this work is the following, an unpublished result of Burklund, Levy, and Pstrągowski. Recall the notation P ≥ n P_{\geq n} P ≥ n from Subsection 2.1: this is the Hopf algebra with graded dual P ≥ n ∨ = F p [ ξ n , ξ n + 1 , … ] P_{\geq n}^\vee= \mathbb{F}_p[\xi_n,\xi_{n+1},\ldots] P ≥ n ∨ = F p [ ξ n , ξ n + 1 , … ] .
Theorem 4.1 ([3 ] ). The element b n n ∈ H 2 ( P ≥ n ) b_{nn} \in H^2(P_{\geq n}) b nn ∈ H 2 ( P ≥ n ) is nilpotent.
We give an independent proof of this when p = 3 p = 3 p = 3 — see Theorem A.1 in the appendix — and we prove a related result in Theorem 5.1 .
Corollary 4.2. Suppose that C C C is a sub-Hopf algebra of either A A A or P P P . If j ≥ n j \geq n j ≥ n and ξ n φ j ≠ 0 \xi_n^{\varphi^j} \ne0 ξ n φ j = 0 in B ∨ B^\vee B ∨ , then the element b n j ∈ H 2 ( C ≥ n ) b_{nj} \in H^2(C_{\geq n}) b nj ∈ H 2 ( C ≥ n ) is nilpotent.
Proof. If C C C is a sub-Hopf algebra of P P P , then we have the restriction map H ∗ ( P ≥ n ) → H ∗ ( C ≥ n ) H^*(P_{\geq n}) \to H^*(C_{\geq n}) H ∗ ( P ≥ n ) → H ∗ ( C ≥ n ) , and so b n n ∈ H ∗ ( C ≥ n ) b_{nn} \in H^*(C_{\geq n}) b nn ∈ H ∗ ( C ≥ n ) is nilpotent. The Steenrod operation P ~ 0 \widetilde{P}^0 P 0 is an algebra map and sends b i j b_{ij} b ij to b i , j + 1 b_{i,j+1} b i , j + 1 as long as these classes are defined, so we conclude that b n j ∈ H 2 ( C ≥ n ) b_{nj} \in H^2(C_{\geq n}) b nj ∈ H 2 ( C ≥ n ) is nilpotent for all j ≥ n j \geq n j ≥ n .
If C C C is a sub-Hopf algebra of A A A , we note that P ≥ n P_{\geq n} P ≥ n is a quotient of A ≥ n A_{\geq n} A ≥ n , so we get an induced map H ∗ ( P ≥ n ) → H ∗ ( A ≥ n ) H^*(P_{\geq n}) \to H^*(A_{\geq n}) H ∗ ( P ≥ n ) → H ∗ ( A ≥ n ) , whence b n n ∈ H ∗ ( A ≥ n ) b_{nn} \in H^*(A_{\geq n}) b nn ∈ H ∗ ( A ≥ n ) is nilpotent. Now as in the previous paragraph, use Steenrod operations and the restriction map H ∗ ( A ≥ n ) → H ∗ ( C ≥ n ) H^*(A_{\geq n}) \to H^*(C_{\geq n}) H ∗ ( A ≥ n ) → H ∗ ( C ≥ n ) .
For the rest of this section we work within the Hopf algebra P P P , so all profile functions have only a polynomial component. If C C C has profile function ( c 1 , c 2 , … ) (c_1,c_2,\ldots) ( c 1 , c 2 , … ) , then C ∩ B i C \cap Bi C ∩ B i will have profile function with n n n th term min ( c n , ( n 2 ) + 1 ) \min(c_n,\binom{n}{2}+1) min ( c n , ( 2 n ) + 1 ) . Recall from the start of § 3 that, given the inclusion C ∩ B i ⊆ C C \cap Bi \subseteq C C ∩ B i ⊆ C , we write ( C ∩ B i ) n (C \cap B i)_n ( C ∩ B i ) n for the algebra whose profile function is the same as that for C ∩ B i C \cap B i C ∩ B i for the first n − 1 n-1 n − 1 terms, the same as that for C C C afterwards:
( min ( c 1 , 1 ) , min ( c 2 , 2 ) , min ( c 3 , 4 ) , … , min ( c n − 1 , ( n − 1 2 ) + 1 ) , c n , c n + 1 , … ) . \left(\min(c_1,1),\min(c_2,2),\min(c_3,4),\ldots,\min\left(c_{n-1},\binom{n-1}{2}+1\right),c_n,c_{n+1},\ldots\right). ( min ( c 1 , 1 ) , min ( c 2 , 2 ) , min ( c 3 , 4 ) , … , min ( c n − 1 , ( 2 n − 1 ) + 1 ) , c n , c n + 1 , … ) .
According to Proposition 3.4 and Theorem 3.1 , we want to prove the following.
Proposition 4.3. Fix a sub-Hopf algebra C ⊆ P C \subseteq P C ⊆ P . If k ≥ ( n 2 ) + 1 k \ge\binom{n}{2}+1 k ≥ ( 2 n ) + 1 and ξ n k ≠ 0 \xi_n^k \ne0 ξ n k = 0 in B ∨ B^\vee B ∨ , then the element b n k ∈ H 2 ( ( C ∩ B i ) n ) b_{nk} \in H^2((C \cap B i)_n) b nk ∈ H 2 (( C ∩ B i ) n ) is nilpotent.
Corollary 4.4. For any C ⊆ A C \subseteq A C ⊆ A , C ∩ B i C \cap B i C ∩ B i detects nilpotence in Stable ( B ∨ ) \operatorname{Stable}(B^\vee) Stable ( B ∨ ) , and the restriction map
H ∗ ( C ) ⟶ ( H ∗ ( C ∩ B i ) ) C H^*(C) \longrightarrow\left(H^*(C \cap B i)\right)^C H ∗ ( C ) ⟶ ( H ∗ ( C ∩ B i ) ) C is an F F F -isomorphism.
Proof of Proposition 4.3 . Let B = ( C ∩ B i ) n B=(C \cap B i)_n B = ( C ∩ B i ) n . Let ( b 1 , b 2 , … ) (b_1,b_2,\ldots) ( b 1 , b 2 , … ) be the profile function for B B B , which means that b i = min ( c i , ( i 2 ) + 1 ) b_i=\min(c_i,\binom{i}{2}+1) b i = min ( c i , ( 2 i ) + 1 ) when i < n i<n i < n , while b i = c i b_i=c_i b i = c i when i ≥ n i\ge n i ≥ n . Note that B ≥ n = C ≥ n B_{\ge n}=C_{\ge n} B ≥ n = C ≥ n . Fix k ≥ ( n 2 ) + 1 k\ge\binom{n}{2}+1 k ≥ ( 2 n ) + 1 ; we may assume that ξ n k ≠ 0 \xi_n^k\ne0 ξ n k = 0 in B ∨ B^\vee B ∨ and hence also in C ∨ C^\vee C ∨ . We use downward induction via a sequence of sub-Hopf algebras of B n B_n B n , first of the form
C ≥ n = B ≥ n ↪ B ≥ n − 1 ↪ ⋯ ↪ B ≥ 2 ↪ B ≥ 1 = C = ( C ∩ B i ) n , C_{\ge n}=B_{\ge n}\hookrightarrow B_{\ge n-1}\hookrightarrow\cdots\hookrightarrow B_{\ge2}\hookrightarrow B_{\ge1}=C=(C\cap B i)_n, C ≥ n = B ≥ n ↪ B ≥ n − 1 ↪ ⋯ ↪ B ≥ 2 ↪ B ≥ 1 = C = ( C ∩ B i ) n ,
and then we will interpolate further, showing that b n k b_{nk} b nk is nilpotent over each of them. The induction starts with Theorem 4.1 and its corollary: b n k b_{nk} b nk is nilpotent in H ∗ ( B ≥ n ) H^*(B_{\ge n}) H ∗ ( B ≥ n ) . We want to show that for each m ≤ n − 1 m\le n-1 m ≤ n − 1 , if b n k b_{nk} b nk is nilpotent in H ∗ ( B ≥ m + 1 ) H^*(B_{\ge m+1}) H ∗ ( B ≥ m + 1 ) , then it remains nilpotent in H ∗ ( B ≥ m ) H^*(B_{\ge m}) H ∗ ( B ≥ m ) .
By definition, B ≥ m + 1 B_{\ge m+1} B ≥ m + 1 has profile function
B ≥ m + 1 : ( 0 , … , 0 ⏟ m − 1 , 0 , b m + 1 , … ) . B_{\ge m+1}:(\underbrace{0,\ldots,0}_{m-1},0,b_{m+1},\ldots). B ≥ m + 1 : ( m − 1 0 , … , 0 , 0 , b m + 1 , … ) .
We define the algebra B ≥ m + 1 , j B_{\ge m+1,j} B ≥ m + 1 , j by the profile function
( 0 , … , 0 ⏟ m − 1 , j , b m + 1 , … ) (\underbrace{0,\ldots,0}_{m-1},j,b_{m+1},\ldots) ( m − 1 0 , … , 0 , j , b m + 1 , … )
for each 0 ≤ j ≤ b m 0\le j\le b_m 0 ≤ j ≤ b m , so B ≥ m + 1 , 0 = B ≥ m + 1 = B ≥ m + 2 , b m + 1 B_{\ge m+1,0}=B_{\ge m+1}=B_{\ge m+2,b_{m+1}} B ≥ m + 1 , 0 = B ≥ m + 1 = B ≥ m + 2 , b m + 1 . For each j j j with 0 ≤ j ≤ b m − 1 0\le j\le b_m-1 0 ≤ j ≤ b m − 1 , we have a Hopf algebra extension
B ≥ m + 1 , j ⟶ B ≥ m + 1 , j + 1 ⟶ D ( P m j ) . B_{\ge m+1,j}\longrightarrow B_{\ge m+1,j+1}\longrightarrow D(P_m^j). B ≥ m + 1 , j ⟶ B ≥ m + 1 , j + 1 ⟶ D ( P m j ) .
Figure 2 (PDF p. 16).
We proceed by induction, so we assume that b n k b_{nk} b nk is nilpotent over B ≥ m + 1 , j B_{\ge m+1,j} B ≥ m + 1 , j , and we want to prove that b n k b_{nk} b nk remains nilpotent over B ≥ m + 1 , j + 1 B_{\ge m+1,j+1} B ≥ m + 1 , j + 1 . If b n k M = 0 b_{nk}^{M}=0 b nk M = 0 in the cohomology of B ≥ m + 1 , j B_{\ge m+1,j} B ≥ m + 1 , j , then b n k M b_{nk}^{M} b nk M is in the kernel of the restriction map H ∗ ( B ≥ m + 1 , j + 1 ) → H ∗ ( B ≥ m + 1 , j ) H^*(B_{\ge m+1,j+1}) \to H^*(B_{\ge m+1,j}) H ∗ ( B ≥ m + 1 , j + 1 ) → H ∗ ( B ≥ m + 1 , j ) , so by the Quillen-Venkov Lemma 2.4, we see that b n k 2 M = b m j x b_{nk}^{2M}=b_{mj}x b nk 2 M = b mj x for some x ∈ H ∗ ( B ≥ m + 1 , j + 1 ) x \in H^*(B_{\ge m+1,j+1}) x ∈ H ∗ ( B ≥ m + 1 , j + 1 ) . Now we appeal to the Nilpotence Lemma 2.5 . We know that m ≤ n − 1 m \le n-1 m ≤ n − 1 and j < ( m 2 ) + 1 j < \binom{m}{2}+1 j < ( 2 m ) + 1 , and given this, the lemma says that if m + j + 1 ≤ k m+j+1 \le k m + j + 1 ≤ k then b m j b n k b_{mj}b_{nk} b mj b nk is nilpotent. We know that ( n 2 ) + 1 ≤ k \binom{n}{2}+1 \le k ( 2 n ) + 1 ≤ k , so we check that m + j + 1 ≤ ( n 2 ) + 1 m+j+1 \le\binom{n}{2}+1 m + j + 1 ≤ ( 2 n ) + 1 , or equivalently that m + j ≤ ( n 2 ) m+j \le\binom{n}{2} m + j ≤ ( 2 n ) . This is easy: since m ≤ n − 1 m \le n-1 m ≤ n − 1 and j ≤ ( m 2 ) j \le\binom{m}{2} j ≤ ( 2 m ) , we have
m + j ≤ n − 1 + ( n − 1 2 ) = ( n 2 ) . m+j \le n-1+\binom{n-1}{2}=\binom{n}{2}. m + j ≤ n − 1 + ( 2 n − 1 ) = ( 2 n ) .
Thus b m j b n k b_{mj}b_{nk} b mj b nk is nilpotent. Since b n k 2 M b_{nk}^{2M} b nk 2 M is a multiple of b m j b_{mj} b mj , we conclude that b n k b_{nk} b nk is nilpotent. This completes the inductive step and hence the proof. □
Restricting to the algebra D D D
Fix a sub-Hopf algebra C ⊆ A C \subseteq A C ⊆ A . The goal in this section is to use Theorem 3.1 to show that C ∩ D C \cap D C ∩ D detects nilpotence in Stable ( ( C ∩ B i ) ∨ ) \operatorname{Stable}((C \cap Bi)^\vee) Stable (( C ∩ B i ) ∨ ) and that H ∗ ( C ∩ B i ) → ( H ∗ ( C ∩ D ) ) C ∩ B i H^*(C \cap Bi) \to(H^*(C \cap D))^{C\cap Bi} H ∗ ( C ∩ B i ) → ( H ∗ ( C ∩ D ) ) C ∩ B i is an F F F -isomorphism. As in the previous section, we work entirely with sub-Hopf algebras of P P P and then appeal to Proposition 3.4 .
Suppose that B B B is a sub-Hopf algebra of P P P with profile function ( b 1 , b 2 , … ) (b_1,b_2,\ldots) ( b 1 , b 2 , … ) . We say that B B B has a finite profile function if each b i < ∞ b_i < \infty b i < ∞ . Here is the main theorem in this section:
Theorem 5.1. Suppose that B ⊆ P B \subseteq P B ⊆ P has a finite profile function . Given k ≥ n ≥ 1 k \ge n \ge1 k ≥ n ≥ 1 , if ξ n k \xi_n^k ξ n k is primitive in B ∨ B^\vee B ∨ , then b n k b_{nk} b nk is nilpotent in H ∗ ( B ) H^*(B) H ∗ ( B ) .
Given this, to apply Theorem 3.1 and Proposition 3.4 to the inclusion D ⊆ B i D \subseteq Bi D ⊆ B i , we intersect everything with P P P and then look at the Hopf algebra with profile function which looks like D ∩ P D \cap P D ∩ P for the first n − 1 n-1 n − 1 terms, and after that it looks like B i ∩ P Bi \cap P B i ∩ P : the profile function has the form ( 1 , 2 , 3 , … , n − 1 , ( n 2 ) + 1 , ( n + 1 2 ) + 1 , … ) (1,2,3,\ldots,n-1,\binom{n}{2}+1,\binom{n+1}{2}+1,\ldots) ( 1 , 2 , 3 , … , n − 1 , ( 2 n ) + 1 , ( 2 n + 1 ) + 1 , … ) . We need to know whether the cohomology element b n n b_{nn} b nn is nilpotent over this Hopf algebra, but this follows immediately from Theorem 5.1 .
Corollary 5.2. For any C ⊆ A C \subseteq A C ⊆ A , C ∩ D C \cap D C ∩ D detects nilpotence in Stable ( ( C ∩ B i ) ∨ ) \operatorname{Stable}((C \cap Bi)^\vee) Stable (( C ∩ B i ) ∨ ) , and the restriction map
H ∗ ( C ∩ B i ) → ( H ∗ ( C ∩ D ) ) C ∩ B i H^*(C \cap Bi) \to(H^*(C \cap D))^{C\cap Bi} H ∗ ( C ∩ B i ) → ( H ∗ ( C ∩ D ) ) C ∩ B i is an F F F -isomorphism.
If we combine Corollaries 4.4 and 5.2, we get Theorems A and B. The F F F -isomorphism aspect is immediate: just compose the F F F -isomorphisms from the corollaries. Detection of nilpotence perhaps require a bit more consideration, so let’s take care of that.
Proof of Theorem A. Given C ⊆ A C \subseteq A C ⊆ A , we know that C ∩ B i C \cap Bi C ∩ B i detects nilpotence in Stable ( B ∨ ) \operatorname{Stable}(B^\vee) Stable ( B ∨ ) , and also that C ∩ D C \cap D C ∩ D detects nilpotence in Stable ( ( C ∩ B i ) ∨ ) \operatorname{Stable}((C \cap Bi)^\vee) Stable (( C ∩ B i ) ∨ ) . We may need to work in categories Stable ( B ∨ ) \operatorname{Stable}(B^\vee) Stable ( B ∨ ) for different sub-Hopf algebras C C C , so we may add a superscript to indicate the algebra, as in π ∗ C ( − ) \pi_*^C(-) π ∗ C ( − ) . Given a sub-Hopf algebra B ⊆ C B \subseteq C B ⊆ C and an object X X X in Stable ( B ∨ ) \operatorname{Stable}(B^\vee) Stable ( B ∨ ) , we use the notation X ↓ B X\downarrow_B X ↓ B to denote the restriction of X X X to Stable ( B ∨ ) \operatorname{Stable}(B^\vee) Stable ( B ∨ ) .
Detecting nilpotence is defined in terms of objects H ( C ∩ B i ) H(C \cap Bi) H ( C ∩ B i ) and H ( C ∩ D ) H(C \cap D) H ( C ∩ D ) , so for example we want to verify this:
Since C ∩ D C \cap D C ∩ D detects nilpotence in Stable ( ( C ∩ B i ) ∨ ) \operatorname{Stable}((C \cap Bi)^\vee) Stable (( C ∩ B i ) ∨ ) , we know that α ∈ π ∗ C ∩ B i ( R ↓ C ∩ B i ) \alpha\in\pi_*^{C\cap Bi}(R\downarrow_{C\cap Bi}) α ∈ π ∗ C ∩ B i ( R ↓ C ∩ B i ) is nilpotent, and by replacing α \alpha α with a power, we may assume that this element is in fact zero. Now we appeal to [20 Corollary 1.2.7] :
π ∗ C ∩ B i ( R ↓ C ∩ B i ) ≅ π ∗ C ( H ( C ∩ B i ) ∧ R ) = H ( C ∩ B i ) ∗ R . \pi_*^{C\cap Bi}(R\downarrow_{C\cap Bi}) \cong\pi_*^C(H(C\cap Bi)\wedge R)=H(C\cap Bi)_*R. π ∗ C ∩ B i ( R ↓ C ∩ B i ) ≅ π ∗ C ( H ( C ∩ B i ) ∧ R ) = H ( C ∩ B i ) ∗ R .
Since C ∩ B i C \cap Bi C ∩ B i detects nilpotence in Stable ( B ∨ ) \operatorname{Stable}(B^\vee) Stable ( B ∨ ) , we conclude that α ∈ π ∗ B R \alpha\in\pi_*^B R α ∈ π ∗ B R is nilpotent.
One can check the other conditions in the definition of detecting nilpotence similarly. □
To prove Theorem 5.1 , we use the observation that any B ⊆ P B \subseteq P B ⊆ P with a finite profile function comes equipped with surjections to finite-dimensional Hopf algebras: for each j ≥ 1 j \ge1 j ≥ 1 we have a quotient map B → B ≤ j B \to B_{\leq j} B → B ≤ j . So if we can show that ξ n p k \xi_n^{p^k} ξ n p k is primitive in some B ≤ j B_{\leq j} B ≤ j and that b n k ∈ H ∗ ( B ≤ j ) b_{nk} \in H^*(B_{\leq j}) b nk ∈ H ∗ ( B ≤ j ) is nilpotent, then its image b n k ∈ H ∗ ( B ) b_{nk} \in H^*(B) b nk ∈ H ∗ ( B ) is also nilpotent. Since B ≤ j B_{\leq j} B ≤ j is finite-dimensional, we can appeal to [18 Theorem 1.2] : we can detect nilpotence in H ∗ ( B ≤ j ) H^*(B_{\leq j}) H ∗ ( B ≤ j ) by restricting to the “quasi-elementary” sub-Hopf algebras of B ≤ j B_{\leq j} B ≤ j . This means that we need to understand, at least partially, the quasi-elementary sub-Hopf algebras of an algebra like B ≤ j B_{\leq j} B ≤ j — that is, a finite-dimensional sub-Hopf algebra of P ≤ j P_{\leq j} P ≤ j . We investigate this in the next subsection.
Quasi-elementary Hopf algebras
Serre showed that a p p p -group is elementary abelian if and only if its mod p p p cohomology satisfies a certain condition, and that cohomological condition is a key part in proving many facts about group cohomology, such as detecting nilpotence and Quillen stratification. Wilkerson coined the term “quasi-elementary” for those Hopf algebras which satisfy Serre’s cohomological criterion (Definition 5.3 ). It is clear that any elementary Hopf algebra is quasi-elementary, and he showed that the converse holds for sub-Hopf algebras of the mod 2 2 2 Steenrod algebra: this is very useful in proving results at the prime 2 2 2 like Theorem 1.3 . Unfortunately these concepts differ for sub-Hopf algebras of the odd primary Steenrod algebra: there are quasi-elementary sub-Hopf algebras of A A A which are not elementary. (The same goes for sub-Hopf algebras of P P P : the issue is not the exterior part of the Steenrod algebra.) The main obstacle to obtaining a result like Theorem 1.3 at odd primes is that we do not have a full classification of the quasi-elementary sub-Hopf algebras of A A A or of P P P . Nakano and the author claimed in [16 ] to have such a classification. The proof is incomplete, though, and we do not know how to fill in the gaps. We provide a partial replacement here: see Theorem 5.4 below. We also discuss the classification problem in § 7 .
Serre’s cohomological condition says that a p p p -group is elementary abelian if and only if every product of Bocksteins applied to nonzero cohomology classes in H 1 H^1 H 1 is non-nilpotent. For Hopf algebras, it turns out that it is not necessary to look at all degree one elements; in the evenly-graded connected case, it is sufficient to look at degree one elements in the kernel of the first nonvanishing Steenrod operation on H s , t H^{s,t} H s , t , which is P t / 2 \mathcal{P}^{t/2} P t /2 or P 0 \mathcal{P}^0 P 0 , depending on the grading convention. Since we are only dealing with evenly graded Hopf algebras in this section, we will use the P \mathcal{P} P notation.
The following definition was first published in [18 ] , although as noted above, the idea is really due to Wilkerson. We focus on the case of evenly graded connected Hopf algebras, and that simplifies the definition a bit.
Definition 5.3. Fix an odd prime p p p , and let Q Q Q be a connected evenly graded cocommutative Hopf algebra over F p \mathbb{F}_p F p . A nonzero element v ∈ H 2 ( Q ) v \in H^2(Q) v ∈ H 2 ( Q ) is a Serre element if v = β P ~ 0 ( w ) v = \beta\widetilde{\mathcal{P}}^0(w) v = β P 0 ( w ) for some w ∈ H 1 ( Q ) ∩ ker P ~ 0 w \in H^1(Q) \cap\ker\widetilde{\mathcal{P}}^0 w ∈ H 1 ( Q ) ∩ ker P 0 . The Hopf algebra Q Q Q is quasi-elementary if no product of Serre elements is nilpotent.
Note first that every elementary Hopf algebra is quasi-elementary. Note also that the definition of quasi-elementary does not work well for sub-Hopf algebras of P P P with infinite profile functions, or indeed for P P P itself: there are no Serre elements in H ∗ ( P ) H^*(P) H ∗ ( P ) , for example, because P ~ 0 \widetilde{\mathcal{P}}^0 P 0 is injective on H 1 ( P ) H^1(P) H 1 ( P ) . So in the case of sub-Hopf algebras of the Steenrod algebra, we should add conditions, for example assuming a finite profile function, in order to get meaningful results.
Here is a partial classification of quasi-elementary sub-Hopf algebras of P P P ; this is the main result of this section.
Theorem 5.4. Fix N N N with 1 < N ≤ ∞ 1 < N \leq\infty 1 < N ≤ ∞ and suppose that Q ⊆ P < N Q \subseteq P_{<N} Q ⊆ P < N has finite profile function ( q 1 , q 2 , q 3 , … ) (q_1,q_2,q_3,\ldots) ( q 1 , q 2 , q 3 , … ) . If Q Q Q is quasi-elementary, then q i ≤ i q_i \leq i q i ≤ i when i < N / 2 i < N/2 i < N /2 . If N = ∞ N = \infty N = ∞ , then Q Q Q has a bounded profile function.
(Recall from Remark 2.3 that the sub-Hopf algebras of P < N P_{<N} P < N are determined by their profile functions, just as for sub-Hopf algebras of P P P .)
Corollary 5.5. Every quasi-elementary sub-Hopf algebra of P P P is contained in D D D .
This corollary verifies [20 Conjecture 2.1.13] . See § 7 , for discussion regarding the quasi-elementary sub-Hopf algebras of P P P .
The following is a useful criterion for identifying non-quasi-elementary sub-Hopf algebras of P < N P_{<N} P < N .
Lemma 5.6. Fix N N N with 1 < N ≤ ∞ 1 < N \le\infty 1 < N ≤ ∞ . Fix integers t , n < N t,n < N t , n < N , and fix a sub-Hopf algebra Q Q Q of P < N P_{<N} P < N with profile function ( q 1 , q 2 , … ) (q_1,q_2,\ldots) ( q 1 , q 2 , … ) . Suppose that
(a) ξ t p s \xi_t^{p^s} ξ t p s and ξ n p k \xi_n^{p^k} ξ n p k are both nonzero and primitive in Q ∨ Q^\vee Q ∨ , and
(b) The profile function entries q t q_t q t and q n q_n q n are finite.
If b t , s b n , k b_{t,s}b_{n,k} b t , s b n , k is nilpotent in H ∗ ( Q ) H^*(Q) H ∗ ( Q ) , then Q Q Q is not quasi-elementary.
Of course condition (b) is automatically satisfied if Q Q Q has a finite profile function. Note that the integers t , n t,n t , n need not be distinct.
Proof. Let L = q t − 1 L=q_t-1 L = q t − 1 and M = q n − 1 M=q_n-1 M = q n − 1 , so that ξ t p L \xi_t^{p^L} ξ t p L and ξ n p M \xi_n^{p^M} ξ n p M are the largest p p p th powers of these generators which are nonzero in Q ∨ Q^\vee Q ∨ . Then b t , L b_{t,L} b t , L and b n , M b_{n,M} b n , M are Serre elements. (Since ξ t p s \xi_t^{p^s} ξ t p s and ξ n p k \xi_n^{p^k} ξ n p k are primitive, so are any larger p p p th powers of these generators, and so b t , L b_{t,L} b t , L and b n , M b_{n,M} b n , M are elements of H 2 H^2 H 2 .) We will show that b t , L b n , M b_{t,L}b_{n,M} b t , L b n , M is nilpotent.
Suppose that ( b t , s b n , k ) i = 0 (b_{t,s}b_{n,k})^i=0 ( b t , s b n , k ) i = 0 . We may apply the algebra map P ~ 0 \widetilde{\mathcal{P}}^0 P 0 repeatedly to get b t , s + ℓ b n , k + ℓ i = 0 b_{t,s+\ell}b_{n,k+\ell}^i=0 b t , s + ℓ b n , k + ℓ i = 0 for any ℓ ≥ 0 \ell\ge0 ℓ ≥ 0 , and so without loss of generality, we may assume that k = M k=M k = M : we may assume that the relation has the form b t , s b n , M i = 0 b_{t,s}b_{n,M}^i=0 b t , s b n , M i = 0 , with s ≤ L s\le L s ≤ L . If s = L s=L s = L , then we have a product b t , L b n , M b_{t,L}b_{n,M} b t , L b n , M of Serre elements which is nilpotent, so we may assume that s < L s<L s < L .
Choose d d d so that p d > i p^d>i p d > i , and multiply this relation by b n , M p d − i b_{n,M}^{p^d-i} b n , M p d − i to get b t , s i b n , M p d = 0 b_{t,s}^i b_{n,M}^{p^d}=0 b t , s i b n , M p d = 0 . Now apply Steenrod operations that increase the power of b n , M b_{n,M} b n , M and increase the second index of b t , s b_{t,s} b t , s . That is, apply P ~ p d + j − 1 ⋯ P ~ p d + 1 P ~ p d \widetilde{\mathcal{P}}^{p^{d+j-1}}\cdots\widetilde{\mathcal{P}}^{p^{d+1}}\widetilde{\mathcal{P}}^{p^d} P p d + j − 1 ⋯ P p d + 1 P p d to get
b t , s + j i b n , M p d + j = 0. b_{t,s+j}^i b_{n,M}^{p^{d+j}}=0. b t , s + j i b n , M p d + j = 0.
When j = L − s j=L-s j = L − s , we get b t , L i b n , M p d + L − s = 0 b_{t,L}^i b_{n,M}^{p^{d+L-s}}=0 b t , L i b n , M p d + L − s = 0 , as desired. □
Now we use the Nilpotence Lemma 2.5 to prove Theorem 5.4 : if Q Q Q does not satisfy the hypotheses, we find relations b t , s b n , k b_{t,s}b_{n,k} b t , s b n , k in H ∗ ( Q ) H^*(Q) H ∗ ( Q ) and deduce from Lemma 5.6 that Q Q Q is not quasi-elementary.
Proof of Theorem 5.4 . Fix a sub-Hopf algebra Q Q Q of P < N P_{<N} P < N with finite profile function.
If N = ∞ N=\infty N = ∞ and if the profile function for Q Q Q is not bounded, then find t t t so that ξ i = 0 \xi_i=0 ξ i = 0 for i < t i<t i < t while ξ t ≠ 0 \xi_t\ne0 ξ t = 0 . Let s s s be the largest number so that ξ t p s ≠ 0 \xi_t^{p^s}\ne0 ξ t p s = 0 . Since the profile function is not bounded, we can find k ≥ s + t + 1 k\ge s+t+1 k ≥ s + t + 1 so that ξ n p k ≠ 0 \xi_n^{p^k}\ne0 ξ n p k = 0 for some n n n . Find the minimal such n n n and apply Lemmas 2.5 and 5.6 to deduce that Q Q Q is not quasi-elementary. This takes care of the last sentence of Theorem 5.4 .
Now assume that N N N is arbitrary, and assume that some entry in the profile function satisfies q n > n q_n>n q n > n with n < N / 2 n<N/2 n < N /2 . Choose the minimal such n n n , meaning that ξ j p j = 0 \xi_j^{p^j}=0 ξ j p j = 0 for all j < n j<n j < n . We want to show that Q Q Q is not quasi-elementary. If ξ n \xi_n ξ n is the first non-vanishing generator—that is, if ξ i = 0 \xi_i=0 ξ i = 0 for all i < n i<n i < n —then essentially by the argument in the proof of [14 Proposition 4.1] , we can see that Q Q Q is not quasi-elementary.
In more detail (to fill in what is meant by “essentially” in the previous sentence): if ξ i = 0 \xi_i=0 ξ i = 0 for all i < n i<n i < n and if ξ n p n ≠ 0 \xi_n^{p^n}\ne0 ξ n p n = 0 , then the reduced coproduct on ξ 2 n \xi_{2n} ξ 2 n is ξ n p n ⊗ ξ n \xi_n^{p^n}\otimes\xi_n ξ n p n ⊗ ξ n . Since this tensor product is nonzero in Q ∨ ⊗ Q ∨ Q^\vee\otimes Q^\vee Q ∨ ⊗ Q ∨ , ξ 2 n \xi_{2n} ξ 2 n must be nonzero in Q ∨ Q^\vee Q ∨ ; equivalently, ξ 2 n ≠ 0 \xi_{2n}\ne0 ξ 2 n = 0 can be deduced from ξ n p n ≠ 0 a n d t h e x r e f ‘ A d a m s − M a r g o l i s T h e o r e m 2.1 ‘ o n p r o f i l e f u n c t i o n s . T h e r e d u c e d c o p r o d u c t \xi_n^{p^n}\ne0 and the {xref}`Adams-Margolis Theorem 2.1` on profile functions. The reduced coproduct ξ n p n = 0 an d t h e x r e f ‘ A d am s − M a r g o l i s T h eor e m 2.1‘ o n p r o f i l e f u n c t i o n s . T h er e d u ce d co p r o d u c t \xi_{2n}\mapsto \xi_n^{p^n}\otimes\xi_n$ yields the relation h n h n 0 = 0 h_nh_{n0}=0 h n h n 0 = 0 in cohomology (both h n 0 h_{n0} h n 0 and h n n h_{nn} h nn are cohomology classes because ξ n \xi_n ξ n is primitive in Q ∨ Q^\vee Q ∨ ). Applying Steenrod operations to this relation gives
β P ~ 0 : b n n b n 1 − b n , n + 1 b n 0 = 0 , \beta\widetilde{\mathcal{P}}^0:\quad b_{nn}b_{n1}-b_{n,n+1}b_{n0}=0, β P 0 : b nn b n 1 − b n , n + 1 b n 0 = 0 ,
β P ~ 1 : b n n p b n 1 − b n , n + 1 b n 0 p = 0 , \beta\widetilde{\mathcal{P}}^1:\quad b_{nn}^{p}b_{n1}-b_{n,n+1}b_{n0}^{p}=0, β P 1 : b nn p b n 1 − b n , n + 1 b n 0 p = 0 ,
β P ~ p : b n n p 2 b n 2 − b n , n + 2 b n 0 p 2 = 0 , \beta\widetilde{\mathcal{P}}^p:\quad b_{nn}^{p^2}b_{n2}-b_{n,n+2}b_{n0}^{p^2}=0, β P p : b nn p 2 b n 2 − b n , n + 2 b n 0 p 2 = 0 ,
and in general
b n n p d b n d − b n , n + d b n 0 p d = 0. b_{nn}^{p^d}b_{nd}-b_{n,n+d}b_{n0}^{p^d}=0. b nn p d b n d − b n , n + d b n 0 p d = 0.
By assumption Q Q Q has a finite profile function, so ξ n p n + d = 0 \xi_n^{p^n+d}=0 ξ n p n + d = 0 for some d d d , and if we choose the smallest d d d making this hold, then we get the monomial relation b n n p n b n d = 0 b_{nn}^{p^n}b_{nd}=0 b nn p n b n d = 0 . By Lemma 5.6 , Q Q Q is not quasi-elementary. This deals with the case when ξ n p n ≠ 0 \xi_n^{p^n}\ne0 ξ n p n = 0 and ξ i = 0 \xi_i=0 ξ i = 0 for i < n i<n i < n .
To deal with the remaining case, we may assume that for some t < n t<n t < n , we have in Q ∨ Q^\vee Q ∨ :
ξ i = 0 \xi_i=0 ξ i = 0 for i < t i<t i < t ,
ξ t ≠ 0 \xi_t\ne0 ξ t = 0 ,
ξ j p j = 0 \xi_j^{p^j}=0 ξ j p j = 0 for all j < n j<n j < n ,
ξ n p n ≠ 0 \xi_n^{p^n}\ne0 ξ n p n = 0 ,
ξ k p j = 0 \xi_k^{p^j}=0 ξ k p j = 0 for all j , k j,k j , k with N > k + t ≥ 2 t + 1 N>k+t\ge2t+1 N > k + t ≥ 2 t + 1 and j ≥ 2 t j\ge2t j ≥ 2 t .
Explanation: (1) and (2) say that ξ t \xi_t ξ t is the first generator present in Q ∨ Q^\vee Q ∨ . (3) and (4) say that ξ n \xi_n ξ n is the first generator with ξ n p n ≠ 0 \xi_n^{p^n}\ne0 ξ n p n = 0 . (5) is because of annihilator considerations: if ξ t p s ≠ 0 \xi_t^{p^s}\ne0 ξ t p s = 0 with s < t s<t s < t and ξ k p j ≠ 0 \xi_k^{p^j}\ne0 ξ k p j = 0 , then by Lemma 2.5 , b t s b k j b_{ts}b_{kj} b t s b k j is nilpotent if N > k + t N>k+t N > k + t , k ≥ t + 1 k\ge t+1 k ≥ t + 1 , and j ≥ 2 t ≥ t + s + 1 j\ge2t\ge t+s+1 j ≥ 2 t ≥ t + s + 1 , so if any such ξ k p j \xi_k^{p^j} ξ k p j were nonzero, we would get a monomial relation in H ∗ H^* H ∗ , and Lemma 5.6 would then tell us that Q Q Q is not quasi-elementary.
Combining (4) and (5), along with the assumption that n > t n>t n > t , gives
n < 2 t , or equivalently n ≤ 2 t − 1. n<2t,\text{ or equivalently }n\le2t-1. n < 2 t , or equivalently n ≤ 2 t − 1.
That is: n n n satisfies n > t n>t n > t so n ≥ t + 1 n\ge t+1 n ≥ t + 1 , so n + t ≥ 2 t + 1 n+t\ge2t+1 n + t ≥ 2 t + 1 . Also 2 n < N 2n<N 2 n < N so n + t < 2 n < N n+t<2n<N n + t < 2 n < N , so n n n satisfies the first inequality in (5): N > n + t ≥ 2 t + 1 N>n+t\ge2t+1 N > n + t ≥ 2 t + 1 . Since ξ n p n ≠ 0 \xi_n^{p^n}\ne0 ξ n p n = 0 , the exponent n n n must fail to satisfy the second inequality in (5): we must have n < 2 t n<2t n < 2 t .
The reduced coproduct on ξ 2 n \xi_{2n} ξ 2 n is
∑ j = 1 2 n − 1 ξ 2 n − j p j ⊗ ξ j . \sum_{j=1}^{2n-1}\xi_{2n-j}^{p^j}\otimes\xi_j. j = 1 ∑ 2 n − 1 ξ 2 n − j p j ⊗ ξ j .
Because of (1), we can change the limits on the sum to go from t t t to 2 n − t 2n-t 2 n − t . Because of (3), we can omit more terms: when j > n j>n j > n , then 2 n − j < n < j 2n-j<n<j 2 n − j < n < j , so ξ 2 n − j p j = 0 \xi_{2n-j}^{p^j}=0 ξ 2 n − j p j = 0 . So the reduced coproduct on ξ 2 n \xi_{2n} ξ 2 n is
∑ j = t n ξ 2 n − j p j ⊗ ξ j . \sum_{j=t}^{n}\xi_{2n-j}^{p^j}\otimes\xi_j. j = t ∑ n ξ 2 n − j p j ⊗ ξ j .
Sublemma. If the Hopf algebra Q ∨ Q^\vee Q ∨ satisfies conditions (1)–(6), then the elements ξ 2 n − j p j \xi_{2n-j}^{p^j} ξ 2 n − j p j and ξ j \xi_j ξ j are primitive when t ≤ j ≤ n t\le j\le n t ≤ j ≤ n .
Proof. Combining (6) with the inequality j ≤ n j\le n j ≤ n gives j < 2 t j<2t j < 2 t . Therefore each element ξ j \xi_j ξ j is primitive: each term in its coproduct involves ξ i \xi_i ξ i and ξ j − i \xi_{j-i} ξ j − i , and either i i i or j − i j-i j − i will be less than t t t , so assumption (1) tells us that each term is zero.
Now we examine ξ 2 n − j p j \xi_{2n-j}^{p^j} ξ 2 n − j p j when t ≤ j ≤ n t\le j\le n t ≤ j ≤ n . The reduced coproduct on this element is
ξ 2 n − j p j ↦ ∑ i = 1 2 n − j − 1 ξ 2 n − j − i p i + j ⊗ ξ i p j . \xi_{2n-j}^{p^j}\mapsto\sum_{i=1}^{2n-j-1}\xi_{2n-j-i}^{p^{i+j}}\otimes\xi_i^{p^j}. ξ 2 n − j p j ↦ i = 1 ∑ 2 n − j − 1 ξ 2 n − j − i p i + j ⊗ ξ i p j .
The terms with i < t i<t i < t are zero, by (1). Combining (6) with i ≥ t i\ge t i ≥ t and j ≥ t j\ge t j ≥ t , we get
n ≥ n + ( n − 2 t + 1 ) = 2 n − 2 t + 1 ≥ 2 n − j − i + 1 > 2 n − j − i . n\ge n+(n-2t+1)=2n-2t+1\ge2n-j-i+1>2n-j-i. n ≥ n + ( n − 2 t + 1 ) = 2 n − 2 t + 1 ≥ 2 n − j − i + 1 > 2 n − j − i .
This means that assumption (3) applies to ξ 2 n − j − i c p i + j \xi_{2n-j-i}^{cp^{i+j}} ξ 2 n − j − i c p i + j . We also have
2 i + 2 j ≥ 2 t + 2 t = 4 t > 2 n , 2i+2j \ge2t+2t=4t>2n, 2 i + 2 j ≥ 2 t + 2 t = 4 t > 2 n ,
so
i + j > 2 n − j − i . i+j>2n-j-i. i + j > 2 n − j − i .
So by (3), we have ξ 2 n − j − i c p i + j = 0 \xi_{2n-j-i}^{cp^{i+j}}=0 ξ 2 n − j − i c p i + j = 0 for all terms in the coproduct. Therefore ξ 2 n − j c p j \xi_{2n-j}^{cp^j} ξ 2 n − j c p j is primitive.
As a result, the coproduct on ξ 2 n \xi_{2n} ξ 2 n produces a relation
∑ j = t n h 2 n − j , j h j 0 = 0 \sum_{j=t}^{n}h_{2n-j,j}h_{j0}=0 j = t ∑ n h 2 n − j , j h j 0 = 0
where each individual factor h k ℓ h_{k\ell} h k ℓ is a cohomology class, so we can apply Steenrod operations and use the Cartan formula. Applying ( β P ~ 1 ) ( β P ~ 0 ) (\beta\widetilde{\mathcal P}^{1})(\beta\widetilde{\mathcal P}^{0}) ( β P 1 ) ( β P 0 ) to this yields
∑ j = t n ( b 2 n − j , j p b j 1 − b 2 n − j , j + 1 b j 0 p ) = 0. \sum_{j=t}^{n}\left(b_{2n-j,j}^{p}b_{j1}-b_{2n-j,j+1}b_{j0}^{p}\right)=0. j = t ∑ n ( b 2 n − j , j p b j 1 − b 2 n − j , j + 1 b j 0 p ) = 0.
Now apply P ~ p n − 2 ⋯ P ~ p 2 P ~ p \widetilde{\mathcal P}^{p^{n-2}}\cdots\widetilde{\mathcal P}^{p^{2}}\widetilde{\mathcal P}^{p} P p n − 2 ⋯ P p 2 P p :
∑ j = t n ( b 2 n − j , j p n − 1 b j , n − 1 − b 2 n − j , n + j − 1 b j 0 p n − 1 ) = 0. \sum_{j=t}^{n}\left(b_{2n-j,j}^{p^{n-1}}b_{j,n-1}-b_{2n-j,n+j-1}b_{j0}^{p^{n-1}}\right)=0. j = t ∑ n ( b 2 n − j , j p n − 1 b j , n − 1 − b 2 n − j , n + j − 1 b j 0 p n − 1 ) = 0.
If j < n j<n j < n , then b j , n − 1 = 0 b_{j,n-1}=0 b j , n − 1 = 0 by (3). Also 2 t ≤ n + t − 1 ≤ n + j − 1 2t\le n+t-1\le n+j-1 2 t ≤ n + t − 1 ≤ n + j − 1 , so b 2 n − j , n + j − 1 = 0 b_{2n-j,n+j-1}=0 b 2 n − j , n + j − 1 = 0 by (5). So all terms with j < n j<n j < n vanish.
That leaves us with the two j = n j=n j = n terms, the second of which involves b n , 2 n − 1 b_{n,2n-1} b n , 2 n − 1 , but 2 n − 1 ≥ 2 t 2n-1\ge2t 2 n − 1 ≥ 2 t , so b n , 2 n − 1 = 0 b_{n,2n-1}=0 b n , 2 n − 1 = 0 by (5). So we have a monomial relation: b n , n p n − 1 b n , n − 1 = 0 b_{n,n}^{p^{n-1}}b_{n,n-1}=0 b n , n p n − 1 b n , n − 1 = 0 , and therefore Q Q Q cannot be quasi-elementary by Lemma 5.6 . Note that applying P ~ p n − 1 \widetilde{\mathcal P}^{p^{n-1}} P p n − 1 to this yields b n , n p n + 1 = 0 b_{n,n}^{p^{n+1}}=0 b n , n p n + 1 = 0 , if you want a “cleaner” relation.
Proof of Theorem 5.1
Now we prove Theorem 5.1 , so suppose that B ⊆ P B\subseteq P B ⊆ P has a finite profile function. Assume that k ≥ n k\ge n k ≥ n and that ξ n p k \xi_n^{p^k} ξ n p k is primitive and nonzero in B ∨ B^\vee B ∨ . We want to show that b n k b_{nk} b nk is nilpotent in H ∗ ( B ) H^*(B) H ∗ ( B ) .
There is a quotient map B → B < j B\to B_{<j} B → B < j for each j > 0 j>0 j > 0 , and since B B B has a finite profile function, B < j B_{<j} B < j is finite-dimensional. Note that if j > n j>n j > n , then b n k b_{nk} b nk is a cohomology class in H ∗ ( B < j ) H^*(B_{<j}) H ∗ ( B < j ) , and b n k ∈ H ∗ ( B < j ) b_{nk}\in H^*(B_{<j}) b nk ∈ H ∗ ( B < j ) maps to b n k ∈ H ∗ ( B ) b_{nk}\in H^*(B) b nk ∈ H ∗ ( B ) . If we can show that b n k b_{nk} b nk is nilpotent in the cohomology of H ∗ ( B < j ) H^*(B_{<j}) H ∗ ( B < j ) for some j > n j>n j > n , the proof will be complete.
Choose finite N N N so that N > 2 n N>2n N > 2 n . By replacing B B B by B < N B_{<N} B < N , we may assume that we have a finite-dimensional sub-Hopf algebra B ⊆ P < N B\subseteq P_{<N} B ⊆ P < N such that ξ n p k \xi_n^{p^k} ξ n p k is primitive in B ∨ B^\vee B ∨ and that k ≥ n k\ge n k ≥ n . Now we are in position to use Theorem 5.4 , the partial classification of quasi-elementary sub-Hopf algebras in P < N P_{<N} P < N : in the dual of every quasi-elementary sub-Hopf algebra Q Q Q of B B B , we have ξ n p n = 0 \xi_n^{p^n}=0 ξ n p n = 0 , so (since k ≥ n k\ge n k ≥ n ) ξ n p k = 0 \xi_n^{p^k}=0 ξ n p k = 0 , so b n k ↦ 0 b_{nk}\mapsto0 b nk ↦ 0 via the restriction map H ∗ ( B ) → H ∗ ( Q ) H^*(B)\to H^*(Q) H ∗ ( B ) → H ∗ ( Q ) . Since quasi-elementary Hopf algebras detect nilpotence in finite-dimensional Hopf algebras by [18 Theorem 1.2] , we conclude that b n k ∈ H ∗ ( B ) b_{nk}\in H^*(B) b nk ∈ H ∗ ( B ) is nilpotent.
Bounded profile functions
At the prime 2, the elementary sub-Hopf algebras of the mod 2 Steenrod algebra A A A detect nilpotence and give an F F F -isomorphism H ∗ ( A ) → ( lim H ∗ ( E ) ) A H^*(A)\to\left(\lim H^*(E)\right)^A H ∗ ( A ) → ( lim H ∗ ( E ) ) A — see Theorem 1.3 . The goal of this section is to prove Theorem C , a partial replacement for this.
Recall the theorem: given a sub-Hopf algebra C ⊆ A C \subseteq A C ⊆ A , let C ′ = C ∩ D C^\prime= C \cap D C ′ = C ∩ D and let Bdd ( C ′ ) \operatorname{Bdd}(C^\prime) Bdd ( C ′ ) be the category whose objects are sub-Hopf algebras B ⊆ C ′ B \subseteq C^\prime B ⊆ C ′ with bounded profile function and whose morphisms are inclusions. The theorem asserts that the objects in Bdd ( C ′ ) \operatorname{Bdd}(C^\prime) Bdd ( C ′ ) detect nilpotence in Stable ( C ∨ ) \operatorname{Stable}(C^\vee) Stable ( C ∨ ) . We will get to the proof of this soon, but first let’s investigate the potential of having an F F F -isomorphism in this setting. We note that we can define an action of C C C on lim Bdd ( C ′ ) H ∗ ( B ) \lim_{\operatorname{Bdd}(C^\prime)} H^*(B) lim Bdd ( C ′ ) H ∗ ( B ) as follows: let nBdd ( C ′ ) \operatorname{nBdd}(C^\prime) nBdd ( C ′ ) be the full subcategory of Bdd ( C ′ ) \operatorname{Bdd}(C^\prime) Bdd ( C ′ ) consisting of the normal sub-Hopf algebras of C ′ C^\prime C ′ with bounded profile function. This is cofinal in Bdd ( C ′ ) \operatorname{Bdd}(C^\prime) Bdd ( C ′ ) , so we have
lim Bdd ( C ′ ) H ∗ ( B ) ≅ lim nBdd ( C ′ ) H ∗ ( B ) . \lim_{\operatorname{Bdd}(C^\prime)} H^*(B) \cong\lim_{\operatorname{nBdd}(C^\prime)} H^*(B). Bdd ( C ′ ) lim H ∗ ( B ) ≅ nBdd ( C ′ ) lim H ∗ ( B ) .
There are compatible actions of C C C on each term on the right, and therefore an induced action on the limit. For B ∈ nBdd ( C ′ ) B \in\operatorname{nBdd}(C^\prime) B ∈ nBdd ( C ′ ) , the map H ∗ ( C ′ ) → H ∗ ( B ) H^*(C^\prime) \to H^*(B) H ∗ ( C ′ ) → H ∗ ( B ) is one of C C C -algebras, and so we have an induced map
( H ∗ ( C ′ ) ) C → ( lim Bdd ( C ′ ) H ∗ ( B ) ) C . \left(H^*(C^\prime)\right)^C \to\left(\lim_{\operatorname{Bdd}(C^\prime)} H^*(B)\right)^C. ( H ∗ ( C ′ ) ) C → ( Bdd ( C ′ ) lim H ∗ ( B ) ) C .
Theorem B says that we have an F F F -isomorphism H ∗ ( C ) → ( H ∗ ( C ′ ) ) C H^*(C) \to\left(H^*(C^\prime)\right)^C H ∗ ( C ) → ( H ∗ ( C ′ ) ) C . Once we have established that Bdd ( C ′ ) \operatorname{Bdd}(C^\prime) Bdd ( C ′ ) detects nilpotence in Stable ( C ∨ ) \operatorname{Stable}(C^\vee) Stable ( C ∨ ) , to show that the map
H ∗ ( C ′ ) C → ( lim Bdd ( C ′ ) H ∗ ( B ) ) C H^*(C^\prime)^C \to\left(\lim_{\operatorname{Bdd}(C^\prime)} H^*(B)\right)^C H ∗ ( C ′ ) C → ( Bdd ( C ′ ) lim H ∗ ( B ) ) C
is an F F F -isomorphism, we would just need to show that this is an F F F -epimorphism. In fact it suffices to show that the map
H ∗ ( C ′ ) → lim Bdd ( C ′ ) H ∗ ( B ) H^*(C^\prime) \to\lim_{\operatorname{Bdd}(C^\prime)} H^*(B) H ∗ ( C ′ ) → Bdd ( C ′ ) lim H ∗ ( B )
is an F F F -epimorphism. At the prime 2 the corresponding result, in which the inverse limit is taken over the elementary sub-Hopf algebras of C C C , can be established by combining several facts: first, one can compute the cohomology of the elementary sub-Hopf algebras and the induced maps on cohomology; second, because every Hopf algebra involved here has a finite profile function, it is an inverse limit of finite-dimensional Hopf algebras; and third, a theorem of Hopkins and Smith [7 Theorem 4.13] says that if Λ ⊆ Γ \Lambda\subseteq\Gamma Λ ⊆ Γ is an inclusion of finite-dimensional graded connected Hopf algebras, then the induced map on cohomology is an F F F -epimorphism. See [20 Theorem 5.3.4] for details.
In our situation we have all of these components except for the computation of the cohomology of each algebra. If we could do that, we presumably would get an F F F -isomorphism. The cohomology of an arbitrary object in Bdd ( C ′ ) \operatorname{Bdd}(C^\prime) Bdd ( C ′ ) is likely to be quite complicated, but we hope that it would be good enough to work with the quasi-elementary sub-Hopf algebras of C C C . Recall from Theorem 5.4 that every quasi-elementary sub-Hopf algebra of A A A is contained in D D D and has a bounded profile function, so the category Bdd \operatorname{Bdd} Bdd contains the quasi-elementary sub-Hopf algebras. (Note, though, that there are Hopf algebras with bounded profile function that are not quasi-elementary, such as the Hopf algebra with profile function ( 1 , 2 , 3 , 3 , … ) (1,2,3,3, \ldots) ( 1 , 2 , 3 , 3 , … ) .)
If Conjecture 7.1 holds, then we would get an F F F -epimorphism
H ∗ ( C ′ ) → ( lim QElem ( C ′ ) H ∗ ( Q ) ) C , H^*(C^\prime) \to\left(\lim_{\operatorname{QElem}(C^\prime)} H^*(Q)\right)^C, H ∗ ( C ′ ) → ( QElem ( C ′ ) lim H ∗ ( Q ) ) C ,
where the inverse limit is taken over the quasi-elementary sub-Hopf algebras of C ′ C^\prime C ′ . Given the conjecture, we would also likely get a version of Theorem C: see the last paragraph of this section. So we are led to this conjecture:
Conjecture 6.1. For C ⊆ A C \subseteq A C ⊆ A , the quasi-elementary sub-Hopf algebras of C C C detect nilpotence in Stable ( C ∨ ) \operatorname{Stable}(C^\vee) Stable ( C ∨ ) , and there is an F F F -isomorphism
H ∗ ( C ) → ( lim QElem ( C ′ ) H ∗ ( Q ) ) C , H^*(C) \to\left(\lim_{\operatorname{QElem}(C^\prime)} H^*(Q)\right)^C, H ∗ ( C ) → ( QElem ( C ′ ) lim H ∗ ( Q ) ) C ,
where the limit is taken over the category whose objects are quasi-elementary sub-Hopf algebras Q Q Q of C C C , and whose morphisms are inclusions.
Anyway, back to our task, proving Theorem C . We do this by imitating the p = 2 p = 2 p = 2 proof from [20 Section 5.3] . We work in the category Stable( C ∨ ) (C^{\vee}) ( C ∨ ) . Let’s start by recalling [20 1.5.1–2] .
Lemma 6.2. (a) Given a map f : X → X f : X \to X f : X → X (omitting suspensions) with cofiber X / f X/f X / f and telescope f − 1 X f^{-1}X f − 1 X , we have ⟨ X ⟩ = ⟨ X / f ⟩ ∨ ⟨ f − 1 X ⟩ \langle X\rangle= \langle X/f\rangle\vee\langle f^{-1}X\rangle ⟨ X ⟩ = ⟨ X / f ⟩ ∨ ⟨ f − 1 X ⟩ and X / f ∧ f − 1 X = 0 X/f \wedge f^{-1}X = 0 X / f ∧ f − 1 X = 0 .
(b) Therefore given Hopf algebras B ′ ⊆ B ⊆ C B' \subseteq B \subseteq C B ′ ⊆ B ⊆ C with B ′ B' B ′ normal in B B B and B / / B ′ ≅ D ( P t s ) B//B' \cong D(P_t^s) B // B ′ ≅ D ( P t s ) , we have ⟨ H B ⟩ = ⟨ H B ′ ⟩ ∨ ⟨ b s t − 1 H B ⟩ \langle HB\rangle= \langle HB'\rangle\vee\langle b_{st}^{-1}HB\rangle ⟨ H B ⟩ = ⟨ H B ′ ⟩ ∨ ⟨ b s t − 1 H B ⟩ and H B ′ ∧ b s t − 1 H B = 0 HB' \wedge b_{st}^{-1}HB = 0 H B ′ ∧ b s t − 1 H B = 0 .
(c) Similarly, given Hopf algebras B ′ ⊆ B ⊆ C B' \subseteq B \subseteq C B ′ ⊆ B ⊆ C with B ′ B' B ′ normal in B B B and B / / B ′ ≅ Λ ( Q n ) B//B' \cong\Lambda(Q_n) B // B ′ ≅ Λ ( Q n ) , we have ⟨ H B ⟩ = ⟨ H B ′ ⟩ ∨ ⟨ a n − 1 H B ⟩ \langle HB\rangle= \langle HB'\rangle\vee\langle a_n^{-1}HB\rangle ⟨ H B ⟩ = ⟨ H B ′ ⟩ ∨ ⟨ a n − 1 H B ⟩ and H B ′ ∧ a n − 1 H B = 0 HB' \wedge a_n^{-1}HB = 0 H B ′ ∧ a n − 1 H B = 0 .
In the ensuing discussion, we will often start with a Hopf algebra quotient map B → D ( P t s ) B \to D(P_t^s) B → D ( P t s ) ; note that this is equivalent to having a Hopf algebra extension B ′ → B → D ( P t s ) B' \to B \to D(P_t^s) B ′ → B → D ( P t s ) , and similarly for a quotient map B → Λ ( Q n ) B \to\Lambda(Q_n) B → Λ ( Q n ) . We abuse notation and write B / b t s B/b_{ts} B / b t s for the Hopf algebra kernel: this is the sub-Hopf algebra of B B B whose dual is B ∨ / ( ξ t 2 s ) B^{\vee}/(\xi_{t}^{2^{s}}) B ∨ / ( ξ t 2 s ) . According to [20 1.5.2] , H ( B / b t s ) H(B/b_{ts}) H ( B / b t s ) is not the cofiber of H B → b t s H B HB \xrightarrow{b_{ts}} HB H B b t s H B , that is, H ( B / b t s ) ≠ ( H B ) / b t s H(B/b_{ts}) \ne(HB)/b_{ts} H ( B / b t s ) = ( H B ) / b t s . They are however Bousfield equivalent, and that’s all we will need below. As a result, in Bousfield class computations we will omit parentheses, writing ⟨ H B / b t s ⟩ \langle HB/b_{ts}\rangle ⟨ H B / b t s ⟩ for ⟨ H ( B / b t s ) ⟩ = ⟨ ( H B ) / b t s ⟩ \langle H(B/b_{ts})\rangle= \langle(HB)/b_{ts}\rangle ⟨ H ( B / b t s )⟩ = ⟨( H B ) / b t s ⟩ . Similarly we write B / a n B/a_n B / a n for the algebra dual to B ∨ / ( τ n ) B^{\vee}/(\tau_n) B ∨ / ( τ n ) , and for this one we do have H ( B / a n ) = ( H B ) / a n H(B/a_n) = (HB)/a_n H ( B / a n ) = ( H B ) / a n .
For example, B ′ B' B ′ in part (b) above could be written as B / b t s B/b_{ts} B / b t s , and we have
⟨ H B ⟩ = ⟨ H B / b t s ⟩ ∨ ⟨ b t s − 1 H B ⟩ . \langle HB\rangle= \langle HB/b_{ts}\rangle\vee\langle b_{ts}^{-1}HB\rangle. ⟨ H B ⟩ = ⟨ H B / b t s ⟩ ∨ ⟨ b t s − 1 H B ⟩ .
We also define B / 0 : = B B/0 := B B /0 := B , so B / ( 0 , x , y , z , … ) = B / ( x , y , z , … ) B/(0, x, y, z, \ldots) = B/(x, y, z, \ldots) B / ( 0 , x , y , z , … ) = B / ( x , y , z , … ) . This convention will be useful in Lemma 6.6 .
Corollary 6.3. Suppose that B ⊆ C B \subseteq C B ⊆ C has profile function ( ( b i ) ; ( e j ) ) ((b_i); (e_j)) (( b i ) ; ( e j ))
(a) Suppose that b r , b q < ∞ b_r, b_q < \infty b r , b q < ∞ for some r , q r, q r , q , and let i = b r − 1 i = b_r - 1 i = b r − 1 , j = b q − 1 j = b_q - 1 j = b q − 1 . Assume that we have Hopf algebra quotient maps B → D ( P r i ) B \to D(P_r^i) B → D ( P r i ) and B → D ( P q j ) B \to D(P_q^j) B → D ( P q j ) . Then we have
⟨ H B ⟩ = ⟨ H B / b r i ⟩ ∨ ⟨ b r i − 1 H B / b q j ⟩ ∨ ⟨ b q j − 1 b r i − 1 H B ⟩ . \langle HB\rangle= \langle HB/b_{ri}\rangle\vee\langle b_{ri}^{-1}HB/b_{qj}\rangle\vee\langle b_{qj}^{-1}b_{ri}^{-1}HB\rangle. ⟨ H B ⟩ = ⟨ H B / b r i ⟩ ∨ ⟨ b r i − 1 H B / b q j ⟩ ∨ ⟨ b q j − 1 b r i − 1 H B ⟩ . (b) Suppose that b r < ∞ b_r < \infty b r < ∞ and e n = 2 e_n = 2 e n = 2 for some r , n r, n r , n , and let i = b r − 1 i = b_r - 1 i = b r − 1 . Assume that we have Hopf algebra quotient maps B → D ( P r i ) B \to D(P_r^i) B → D ( P r i ) and B → Λ ( Q n ) B \to\Lambda(Q_n) B → Λ ( Q n ) . Then we have
⟨ H B ⟩ = ⟨ H B / b r i ⟩ ∨ ⟨ b r i − 1 H B / a n ⟩ ∨ ⟨ a n − 1 b r i − 1 H B ⟩ . \langle HB\rangle= \langle HB/b_{ri}\rangle\vee\langle b_{ri}^{-1}HB/a_n\rangle\vee\langle a_n^{-1}b_{ri}^{-1}HB\rangle. ⟨ H B ⟩ = ⟨ H B / b r i ⟩ ∨ ⟨ b r i − 1 H B / a n ⟩ ∨ ⟨ a n − 1 b r i − 1 H B ⟩ .
Of course if b q j b r i b_{qj}b_{ri} b q j b r i is nilpotent, the last term in part (a) drops out, and similarly if a n b r i a_nb_{ri} a n b r i is nilpotent for part (b). This is how we will apply the result.
Proof. The lemma tells us that
⟨ H B ⟩ = ⟨ H B / b r i ⟩ ∨ ⟨ b r i − 1 H B ⟩ . \langle HB\rangle= \langle HB/b_{ri}\rangle\vee\langle b_{ri}^{-1}HB\rangle. ⟨ H B ⟩ = ⟨ H B / b r i ⟩ ∨ ⟨ b r i − 1 H B ⟩ .
For part (a), the lemma also tells us that
⟨ H B ⟩ = ⟨ H B / b q j ⟩ ∨ ⟨ b q j − 1 H B ⟩ . \langle HB\rangle= \langle HB/b_{qj}\rangle\vee\langle b_{qj}^{-1}HB\rangle. ⟨ H B ⟩ = ⟨ H B / b q j ⟩ ∨ ⟨ b q j − 1 H B ⟩ .
Smash this with ⟨ b r i − 1 H B ⟩ \langle b_{ri}^{-1}HB\rangle ⟨ b r i − 1 H B ⟩ and use ⟨ H B ∧ H B ⟩ = ⟨ H B ⟩ \langle HB \wedge HB\rangle= \langle HB\rangle ⟨ H B ∧ H B ⟩ = ⟨ H B ⟩ to get
⟨ b r i − 1 H B ⟩ = ⟨ b r i − 1 H B / b q j ⟩ ∨ ⟨ b r i − 1 b q j − 1 H B ⟩ . \langle b_{ri}^{-1}HB\rangle= \langle b_{ri}^{-1}HB/b_{qj}\rangle\vee\langle b_{ri}^{-1}b_{qj}^{-1}HB\rangle. ⟨ b r i − 1 H B ⟩ = ⟨ b r i − 1 H B / b q j ⟩ ∨ ⟨ b r i − 1 b q j − 1 H B ⟩ .
Part (b) is similar.
Here are the main players in the proof of Theorem C . Recall from Subsection 2.1 that for a non-negative integer r r r , D ≥ r D_{\ge r} D ≥ r has profile function
( ( 0 , … , 0 , r , r + 1 , r + 2 , … ) ; ( 1 , … , 1 , 2 , 2 , 2 , … ) ) . ((0,\ldots,0,r,r+1,r+2,\ldots);(1,\ldots,1,2,2,2,\ldots)). (( 0 , … , 0 , r , r + 1 , r + 2 , … ) ; ( 1 , … , 1 , 2 , 2 , 2 , … )) .
If i ≥ 0 i\ge0 i ≥ 0 and q ≥ r + i q\ge r+i q ≥ r + i , we let D ≥ r , i , q D_{\ge r,i,q} D ≥ r , i , q be the sub-Hopf algebra of A A A with profile function
( ( 0 , … , 0 , r , r + 1 , … , r + i , r + i + 1 , … , r + i + 1 , q , q + 1 , q + 2 , … ) ; ( 1 , … , 1 , 2 , 2 , 2 , … ) ) . ((0,\ldots,0,r,r+1,\ldots,r+i,r+i+1,\ldots,r+i+1,q,q+1,q+2,\ldots);(1,\ldots,1,2,2,2,\ldots)). (( 0 , … , 0 , r , r + 1 , … , r + i , r + i + 1 , … , r + i + 1 , q , q + 1 , q + 2 , … ) ; ( 1 , … , 1 , 2 , 2 , 2 , … )) .
If i ≥ 0 i\ge0 i ≥ 0 , we define B r , i B_{r,i} B r , i by the (bounded) profile function
( ( 0 , … , 0 , r , r + 1 , … , r + i , r + i + 1 , r + i + 1 , r + i + 1 , … ) ; ( 1 , … , 1 , 2 , 2 , 2 , … ) ) . ((0,\ldots,0,r,r+1,\ldots,r+i,r+i+1,r+i+1,r+i+1,\ldots);(1,\ldots,1,2,2,2,\ldots)). (( 0 , … , 0 , r , r + 1 , … , r + i , r + i + 1 , r + i + 1 , r + i + 1 , … ) ; ( 1 , … , 1 , 2 , 2 , 2 , … )) .
Here are pictures of the polynomial components of the profile functions for D ≥ r , i , q D_{\ge r,i,q} D ≥ r , i , q and B r , i B_{r,i} B r , i :
Figure 3 (PDF p. 24).
Lemma 6.4. (a) In Stable ( A ∨ ) \operatorname{Stable}(A^\vee) Stable ( A ∨ ) we have lim r H D ≥ r = H F p \lim_{r} HD_{\ge r}=HF_p lim r H D ≥ r = H F p , and for fixed r , i r,i r , i , lim q H D ≥ r , i , q = H B r , i \lim_{q} HD_{\ge r,i,q}=HB_{r,i} lim q H D ≥ r , i , q = H B r , i .
(b) Given C ⊆ A C\subseteq A C ⊆ A , then in Stable ( C ∨ ) \operatorname{Stable}(C^\vee) Stable ( C ∨ ) we have lim r H ( C ∩ D ≥ r ) = H F p \lim_{r} H(C\cap D_{\ge r})=HF_p lim r H ( C ∩ D ≥ r ) = H F p , and for fixed r , i r,i r , i , lim q H ( C ∩ D ≥ r , i , q ) = H ( C ∩ B r , i ) \lim_{q} H(C\cap D_{\ge r,i,q})=H(C\cap B_{r,i}) lim q H ( C ∩ D ≥ r , i , q ) = H ( C ∩ B r , i ) .
Proof. This is immediate. (See [20 Lemma 5.3.2] for the p = 2 p=2 p = 2 version, with a similarly detailed proof.)
Lemma 6.5. Fix r ≥ 0 r\ge0 r ≥ 0 .
(a) In Stable ( A ∨ ) \operatorname{Stable}(A^\vee) Stable ( A ∨ ) we have
⟨ H D ≥ r ⟩ = ⟨ a r − 1 H D ≥ r ⟩ ∨ ⋁ i = 0 r − 1 ⟨ b r i − 1 H D ≥ r / ( a r , b r , 0 , … , b r , i − 1 ) ⟩ ∨ ⟨ H D ≥ r + 1 ⟩ . \langle HD_{\ge r}\rangle=\langle a_r^{-1}HD_{\ge r}\rangle\vee\bigvee_{i=0}^{r-1}\langle b_{r}^{i-1}HD_{\ge r}/(a_r,b_{r,0},\ldots,b_{r,i-1})\rangle\vee\langle HD_{\ge r+1}\rangle. ⟨ H D ≥ r ⟩ = ⟨ a r − 1 H D ≥ r ⟩ ∨ i = 0 ⋁ r − 1 ⟨ b r i − 1 H D ≥ r / ( a r , b r , 0 , … , b r , i − 1 )⟩ ∨ ⟨ H D ≥ r + 1 ⟩ . (b) Given C ⊆ A C\subseteq A C ⊆ A with profile function ( ( c i ) ; ( e j ) ) ((c_i);(e_j)) (( c i ) ; ( e j )) , in Stable ( C ∨ ) \operatorname{Stable}(C^\vee) Stable ( C ∨ ) we have a similar formula for ⟨ H ( C ∩ D ≥ r ) ⟩ \langle H(C\cap D_{\ge r})\rangle ⟨ H ( C ∩ D ≥ r )⟩ , but it depends on whether τ r \tau_r τ r is zero in C ∨ C^\vee C ∨ :
⟨ H ( C ∩ D ≥ r ) ⟩ = ⋁ i = 0 min ( c r , r ) − 1 ⟨ b r i − 1 H ( C ∩ D ≥ r ) / ( b r , 0 , … , b r , i − 1 ) ⟩ ∨ ⟨ H ( C ∩ D ≥ r + 1 ) ⟩ . \langle H(C\cap D_{\ge r})\rangle=\bigvee_{i=0}^{\min(c_r,r)-1}\langle b_r^{i-1}H(C\cap D_{\ge r})/(b_{r,0},\ldots,b_{r,i-1})\rangle\vee\langle H(C\cap D_{\ge r+1})\rangle. ⟨ H ( C ∩ D ≥ r )⟩ = i = 0 ⋁ m i n ( c r , r ) − 1 ⟨ b r i − 1 H ( C ∩ D ≥ r ) / ( b r , 0 , … , b r , i − 1 )⟩ ∨ ⟨ H ( C ∩ D ≥ r + 1 )⟩ .
⟨ H ( C ∩ D ≥ r ) ⟩ = ⟨ a r − 1 H ( C ∩ D ≥ r ) ⟩ ∨ ⋁ i = 0 min ( c r , r ) − 1 ⟨ b r i − 1 H ( C ∩ D ≥ r ) / ( a r , b r 0 , … , b r , i − 1 ) ⟩ ∨ ⟨ H ( C ∩ D ≥ r + 1 ) ⟩ \langle H(C \cap D_{\ge r})\rangle= \langle a_r^{-1}H(C \cap D_{\ge r})\rangle
\vee\bigvee_{i=0}^{\min(c_r,r)-1}\langle b_{ri}^{-1}H(C \cap D_{\ge r})/(a_r,b_{r0},\ldots,b_{r,i-1})\rangle
\vee\langle H(C \cap D_{\ge r+1})\rangle ⟨ H ( C ∩ D ≥ r )⟩ = ⟨ a r − 1 H ( C ∩ D ≥ r )⟩ ∨ i = 0 ⋁ m i n ( c r , r ) − 1 ⟨ b r i − 1 H ( C ∩ D ≥ r ) / ( a r , b r 0 , … , b r , i − 1 )⟩ ∨ ⟨ H ( C ∩ D ≥ r + 1 )⟩
Proof. (a) By the previous lemma we have
⟨ H D ≥ r ⟩ = ⟨ a r − 1 H D ≥ r ⟩ ∨ ⟨ H D ≥ r / a r ⟩ . \langle HD_{\ge r}\rangle= \langle a_r^{-1}HD_{\ge r}\rangle\vee\langle HD_{\ge r}/a_r\rangle. ⟨ H D ≥ r ⟩ = ⟨ a r − 1 H D ≥ r ⟩ ∨ ⟨ H D ≥ r / a r ⟩ .
Now we apply the lemma repeatedly to the last term: we find that
⟨ H D ≥ r / a r ⟩ = ⟨ b r 0 − 1 H D ≥ r / a r ⟩ ∨ ⟨ H D ≥ r / ( a r , b r 0 ) ⟩ = ⟨ b r 0 − 1 H D ≥ r / a r ⟩ ∨ ⟨ b r 1 − 1 H D ≥ r / ( a r , b r 0 ) ⟩ ∨ ⟨ H D ≥ r / ( a r , b r 0 , b r 1 ) ⟩ = ⋯ \begin{aligned}
\langle HD_{\ge r}/a_r\rangle
&= \langle b_{r0}^{-1}HD_{\ge r}/a_r\rangle\vee\langle HD_{\ge r}/(a_r,b_{r0})\rangle\\
&= \langle b_{r0}^{-1}HD_{\ge r}/a_r\rangle\vee\langle b_{r1}^{-1}HD_{\ge r}/(a_r,b_{r0})\rangle\vee\langle HD_{\ge r}/(a_r,b_{r0},b_{r1})\rangle\\
&= \cdots
\end{aligned} ⟨ H D ≥ r / a r ⟩ = ⟨ b r 0 − 1 H D ≥ r / a r ⟩ ∨ ⟨ H D ≥ r / ( a r , b r 0 )⟩ = ⟨ b r 0 − 1 H D ≥ r / a r ⟩ ∨ ⟨ b r 1 − 1 H D ≥ r / ( a r , b r 0 )⟩ ∨ ⟨ H D ≥ r / ( a r , b r 0 , b r 1 )⟩ = ⋯
To finish, observe that
D ≥ r / ( a r , b r 0 , b r 1 , … , b r , r − 1 ) = D ≥ r + 1 . D_{\ge r}/(a_r,b_{r0},b_{r1},\ldots,b_{r,r-1})=D_{\ge r+1}. D ≥ r / ( a r , b r 0 , b r 1 , … , b r , r − 1 ) = D ≥ r + 1 .
Part (b) is similar; one just has to keep track of whether τ r \tau_r τ r is present in C ∨ C^\vee C ∨ , along with which powers of ξ r \xi_r ξ r are present.
Lemma 6.6. Fix q > R ≥ 1 q>R\ge1 q > R ≥ 1 .
(a) In Stable( A ∨ ) (A^\vee) ( A ∨ ) we have
⟨ H D ⟩ = ⟨ H D ≥ R + 1 ⟩ ∨ ⋁ r = 0 R ( ⟨ a r − 1 H D ≥ r ⟩ ∨ ⋁ i = 0 r − 1 ⟨ b r i − 1 H D ≥ r / ( a r , b r 0 , … , b r , i − 1 ) ⟩ ) \langle HD\rangle=\langle HD_{\ge R+1}\rangle\vee\bigvee_{r=0}^{R}\left(\langle a_r^{-1}HD_{\ge r}\rangle\vee\bigvee_{i=0}^{r-1}\langle b_{ri}^{-1}HD_{\ge r}/(a_r,b_{r0},\ldots,b_{r,i-1})\rangle\right) ⟨ H D ⟩ = ⟨ H D ≥ R + 1 ⟩ ∨ r = 0 ⋁ R ( ⟨ a r − 1 H D ≥ r ⟩ ∨ i = 0 ⋁ r − 1 ⟨ b r i − 1 H D ≥ r / ( a r , b r 0 , … , b r , i − 1 )⟩ ) = ⟨ H D ≥ R + 1 ⟩ ∨ ⋁ r = 0 R ( ⟨ a r − 1 H D ≥ r , 0 , q ⟩ ∨ ⋁ i = 0 r − 1 ⟨ b r i − 1 H D ≥ r , i + 1 , q / ( a r , b r 0 , … , b r , i − 1 ) ⟩ ) . =\langle HD_{\ge R+1}\rangle\vee\bigvee_{r=0}^{R}\left(\langle a_r^{-1}HD_{\ge r,0,q}\rangle\vee\bigvee_{i=0}^{r-1}\langle b_{ri}^{-1}HD_{\ge r,i+1,q}/(a_r,b_{r0},\ldots,b_{r,i-1})\rangle\right). = ⟨ H D ≥ R + 1 ⟩ ∨ r = 0 ⋁ R ( ⟨ a r − 1 H D ≥ r , 0 , q ⟩ ∨ i = 0 ⋁ r − 1 ⟨ b r i − 1 H D ≥ r , i + 1 , q / ( a r , b r 0 , … , b r , i − 1 )⟩ ) . (b) Given C ⊆ A C\subseteq A C ⊆ A with profile function ( ( c i ) ; ( e j ) ) ((c_i);(e_j)) (( c i ) ; ( e j )) , in Stable( C ∨ ) (C^\vee) ( C ∨ ) we have
⟨ H ( C ∩ D ) ⟩ = ⟨ H ( C ∩ D ≥ R + 1 ) ⟩ ∨ ⋁ r = 0 R ( ⟨ a ~ r − 1 H ( C ∩ D ≥ r ) ⟩ ∨ ⋁ i = 0 min ( c r , r ) − 1 ⟨ b r i − 1 H ( C ∩ D ≥ r ) / ( a ~ r , b r 0 , … , b r , i − 1 ) ⟩ ) \langle H(C\cap D)\rangle=\langle H(C\cap D_{\ge R+1})\rangle\vee\bigvee_{r=0}^{R}\left(\langle\widetilde a_r^{-1}H(C\cap D_{\ge r})\rangle\vee\bigvee_{i=0}^{\min(c_r,r)-1}\langle b_{ri}^{-1}H(C\cap D_{\ge r})/(\widetilde a_r,b_{r0},\ldots,b_{r,i-1})\rangle\right) ⟨ H ( C ∩ D )⟩ = ⟨ H ( C ∩ D ≥ R + 1 )⟩ ∨ r = 0 ⋁ R ⟨ a r − 1 H ( C ∩ D ≥ r )⟩ ∨ i = 0 ⋁ m i n ( c r , r ) − 1 ⟨ b r i − 1 H ( C ∩ D ≥ r ) / ( a r , b r 0 , … , b r , i − 1 )⟩ = ⟨ H ( C ∩ D ≥ R + 1 ) ⟩ ∨ ⋁ r = 0 R ( ⟨ a ~ r − 1 H ( C ∩ D ≥ r , 0 , q ) ⟩ ∨ ⋁ i = 0 min ( c r , r ) − 1 ⟨ b r i − 1 H ( C ∩ D ≥ r , i + 1 , q ) / ( a ~ r , b r 0 , … , b r , i − 1 ) ⟩ ) , =\langle H(C\cap D_{\ge R+1})\rangle\vee\bigvee_{r=0}^{R}\left(\langle\widetilde a_r^{-1}H(C\cap D_{\ge r,0,q})\rangle\vee\bigvee_{i=0}^{\min(c_r,r)-1}\langle b_{ri}^{-1}H(C\cap D_{\ge r,i+1,q})/(\widetilde a_r,b_{r0},\ldots,b_{r,i-1})\rangle\right), = ⟨ H ( C ∩ D ≥ R + 1 )⟩ ∨ r = 0 ⋁ R ⟨ a r − 1 H ( C ∩ D ≥ r , 0 , q )⟩ ∨ i = 0 ⋁ m i n ( c r , r ) − 1 ⟨ b r i − 1 H ( C ∩ D ≥ r , i + 1 , q ) / ( a r , b r 0 , … , b r , i − 1 )⟩ , where a ~ r = a r \widetilde a_r=a_r a r = a r if τ r ≠ 0 \tau_r\ne0 τ r = 0 in C ∨ C^\vee C ∨ , a ~ r = 0 \widetilde a_r=0 a r = 0 if τ r = 0 \tau_r=0 τ r = 0 in C ∨ C^\vee C ∨ . (That is, when τ r = 0 \tau_r=0 τ r = 0 in C ∨ C^\vee C ∨ , omit terms involving a ~ r − 1 \widetilde a_r^{-1} a r − 1 , and omit mention of a ~ r \widetilde a_r a r in other terms.)
This is an odd primary version of [20 ] , Lemma 5.3.3.
Proof. We will prove part (a). Part (b) is proved almost identically; the only difference is that if some τ r \tau_r τ r or ξ r e i \xi_r^{e_i} ξ r e i is zero in C ∨ C^\vee C ∨ , then we omit the corresponding self-maps and wedge summands.
The first equality follows from Lemma 6.5. For the second equality, we appeal to Corollary 6.3 and the Nilpotence Lemma 2.5 : if k ≥ r + i + 1 k \ge r+i+1 k ≥ r + i + 1 and if ξ n p k = 0 \xi_n^{p^k}=0 ξ n p k = 0 while ξ n p n ≠ 0 \xi_n^{p^n}\ne0 ξ n p n = 0 , then b r i b n k b_{ri}b_{nk} b r i b nk is nilpotent, so in the above equality of Bousfield classes, as long as k ≥ r + i + 1 k \ge r+i+1 k ≥ r + i + 1 , then ⟨ b r i − 1 H B ⟩ = ⟨ b r i − 1 H B / b n k ⟩ \langle b_{ri}^{-1}HB\rangle=\langle b_{ri}^{-1}HB/b_{nk}\rangle ⟨ b r i − 1 H B ⟩ = ⟨ b r i − 1 H B / b nk ⟩ . We apply this repeatedly to the “boxes” ξ n p k \xi_n^{p^k} ξ n p k which form the difference between D ≥ r D_{\ge r} D ≥ r and D ≥ r , i + 1 , q D_{\ge r,i+1,q} D ≥ r , i + 1 , q , to get
⟨ b r i − 1 H D ≥ r ⟩ = ⟨ b r i − 1 H D ≥ r / b r + i + 2 , r + i + 1 ⟩ = ⟨ b r i − 1 H D ≥ r / ( b r + i + 2 , r + i + 1 , b r + i + 3 , r + i + 2 , b r + i + 3 , r + i + 1 ) ⟩ = ⋯ = ⟨ b r i − 1 H D ≥ r , i + 1 , q ⟩ \begin{aligned}
\langle b_{ri}^{-1}HD_{\ge r}\rangle
&=\langle b_{ri}^{-1}HD_{\ge r}/b_{r+i+2,r+i+1}\rangle\\
&=\langle b_{ri}^{-1}HD_{\ge r}/(b_{r+i+2,r+i+1},b_{r+i+3,r+i+2},b_{r+i+3,r+i+1})\rangle\\
&=\cdots=\langle b_{ri}^{-1}HD_{\ge r,i+1,q}\rangle
\end{aligned} ⟨ b r i − 1 H D ≥ r ⟩ = ⟨ b r i − 1 H D ≥ r / b r + i + 2 , r + i + 1 ⟩ = ⟨ b r i − 1 H D ≥ r / ( b r + i + 2 , r + i + 1 , b r + i + 3 , r + i + 2 , b r + i + 3 , r + i + 1 )⟩ = ⋯ = ⟨ b r i − 1 H D ≥ r , i + 1 , q ⟩
for any q q q . To allow a uniform choice of q q q independent of r r r , we choose q > R q>R q > R .
Similarly, since a r b n k a_rb_{nk} a r b nk is nilpotent if k ≥ r k\ge r k ≥ r , then we see that ⟨ a r − 1 H D ≥ r ⟩ = ⟨ a r − 1 H D ≥ r , 0 , q ⟩ \langle a_r^{-1}HD_{\ge r}\rangle=\langle a_r^{-1}HD_{\ge r,0,q}\rangle ⟨ a r − 1 H D ≥ r ⟩ = ⟨ a r − 1 H D ≥ r , 0 , q ⟩ for any q > R q>R q > R .
Proof of Theorem C . For part (a), as in the p = 2 p=2 p = 2 case (the proof of [20 Theorem 5.1.6] ) we can reduce to the smash nilpotence case, and then we can reduce to a Bousfield class argument. Summarizing the first paragraph of the p = 2 p=2 p = 2 proof, we end up with an object X ∞ X^\infty X ∞ , and we want to know whether H ( C ∩ D ) ∧ X ∞ = 0 H(C\cap D)\wedge X^\infty=0 H ( C ∩ D ) ∧ X ∞ = 0 , and we may assume that H B ∧ X ∞ = 0 HB\wedge X^\infty=0 H B ∧ X ∞ = 0 for every sub-Hopf algebra B ⊆ C ∩ D B\subseteq C\cap D B ⊆ C ∩ D with bounded profile function.
Since F p \mathbb{F}_p F p is a (trivial) Hopf algebra with a bounded profile function, we know that H F p ∧ X ∞ = 0 H\mathbb{F}_p\wedge X^\infty=0 H F p ∧ X ∞ = 0 . Thus any map
S 0 ⟶ lim R H ( C ∩ D ≥ R + 1 ) ∧ X ∞ = H F p ∧ X ∞ S^0\longrightarrow\lim_{R}H(C\cap D_{\ge R+1})\wedge X^\infty=H\mathbb{F}_p\wedge X^\infty S 0 ⟶ R lim H ( C ∩ D ≥ R + 1 ) ∧ X ∞ = H F p ∧ X ∞
is null; we have applied Lemma 6.4 here. Since homotopy commutes with colimits, then for some R R R , a particular map S 0 → H ( C ∩ D ≥ R + 1 ) ∧ X ∞ S^0\to H(C\cap D_{\ge R+1})\wedge X^\infty S 0 → H ( C ∩ D ≥ R + 1 ) ∧ X ∞ is null, and by arguing as in the p = 2 p=2 p = 2 proof, this means that H ( C ∩ D ≥ R + 1 ) ∧ X ∞ = 0 H(C\cap D_{\ge R+1})\wedge X^\infty=0 H ( C ∩ D ≥ R + 1 ) ∧ X ∞ = 0 . Similarly, since each Hopf algebra C ∩ B r , i + 1 C\cap B_{r,i+1} C ∩ B r , i + 1 has bounded profile function, we know that H ( C ∩ B r , i + 1 ) ∧ X ∞ = 0 H(C\cap B_{r,i+1})\wedge X^\infty=0 H ( C ∩ B r , i + 1 ) ∧ X ∞ = 0 , so we can repeat the colimit argument to see that
b r i − 1 H ( C ∩ D ≥ r , i + 1 , q ) / ( a r , b r 0 , … , b r , i − 1 ) ∧ X ∞ = 0 , b_{ri}^{-1}H(C\cap D_{\ge r,i+1,q})/(a_r,b_{r0},\ldots,b_{r,i-1})\wedge X^\infty=0, b r i − 1 H ( C ∩ D ≥ r , i + 1 , q ) / ( a r , b r 0 , … , b r , i − 1 ) ∧ X ∞ = 0 ,
a r − 1 H D ≥ r , 0 , q ∧ X ∞ = 0 , a_r^{-1}HD_{\ge r,0,q}\wedge X^\infty=0, a r − 1 H D ≥ r , 0 , q ∧ X ∞ = 0 ,
for all r , i r,i r , i . So by Lemma 6.6 , we see that H ( C ∩ D ) ∧ X ∞ = 0 H(C\cap D)\wedge X^\infty=0 H ( C ∩ D ) ∧ X ∞ = 0 , as desired.
Presumably a classification of quasi-elementary sub-Hopf algebras of A A A as in Conjecture 7.1 (see also Remark 7.4) would allow one to establish Bousfield equivalences between B r B_r B r and some quasi-elementary Q Q Q , and so we hope that we could then prove that the quasi-elementary sub-Hopf algebras detect nilpotence.
Discussion
Quasi-elementary sub-Hopf algebras of P P P
One of main open questions here is the classification of quasi-elementary sub-Hopf algebras of A A A . Let’s focus on quasi-elementary sub-Hopf algebras of P P P : if one could classify those, it seems likely that this would carry over to a classification of the quasi-elementary sub-Hopf algebras of A A A : see Remark 7.4. Here is at least some of what we know about the situation with P P P :
(a) Every elementary Hopf algebra is quasi-elementary. The maximal elementary sub-Hopf algebras of A A A were described in Subsection 2.1, and it is easy to figure out the elementary sub-Hopf algebras of P P P from there: the maximal ones have profile function e ( r ) : = ( 0 , … , 0 , r + 1 , r + 1 , … ) \operatorname{e}(r):=(0,\ldots,0,r+1,r+1,\ldots) e ( r ) := ( 0 , … , 0 , r + 1 , r + 1 , … ) , and every list ( k 1 , k 2 , k 3 , … ) (k_1,k_2,k_3,\ldots) ( k 1 , k 2 , k 3 , … ) of non-negative integers which is term-wise no larger than some e ( r ) \operatorname{e}(r) e ( r ) is a valid profile function and corresponds to an elementary Hopf algebra. That is, the profile function fits inside a rectangle like this: (b) Fix r ≥ 2 r \ge2 r ≥ 2 . The computation given in [23 6.3] , and see also [8 Theorem 1.1] , generalizes to show that Hopf algebras with profile function
( 0 , … , 0 , 1 , n r , n r + 1 , … ) (0,\ldots,0,1,n_r,n_{r+1},\ldots) ( 0 , … , 0 , 1 , n r , n r + 1 , … )
with each n i ≤ r n_i \le r n i ≤ r are quasi-elementary: the profile function fits inside
ξ r p r ξ r p r − 1 ⋮ ξ r \begin{array}{c}
\xi_r^{p^r}\\
\xi_r^{p^{r-1}}\\
\vdots\\
\xi_r
\end{array} ξ r p r ξ r p r − 1 ⋮ ξ r
These Hopf algebras include the ones in (a).
Conjecture 7.1. The quasi-elementary sub-Hopf algebras of P P P are those described in (b).
(This was stated as a theorem in [16 ] .)
Note that one can compute the cohomology of a Hopf algebra as in (b) — call it Q Q Q — by imitating Wilkerson’s work in [23 6.3] . There is a spectral sequence converging to the cohomology with E 2 E_2 E 2 -page of the form
Λ ( h t s : ξ t p s ≠ 0 in Q ∨ ) ⊗ F p [ b t s : ξ t p s ≠ 0 in Q ∨ ] . \Lambda(h_{ts}:\xi_t^{p^s}\ne0\text{ in }Q^\vee)\otimes\mathbb{F}_p[b_{ts}:\xi_t^{p^s}\ne0\text{ in }Q^\vee]. Λ ( h t s : ξ t p s = 0 in Q ∨ ) ⊗ F p [ b t s : ξ t p s = 0 in Q ∨ ] .
The differentials only affect the exterior part, so the cohomology ends up being the tensor product of the polynomial part with a subquotient of the exterior part. That is, there is an F F F -isomorphism
H ∗ ( Q ) → F -iso F p [ b t s : ξ t p s ≠ 0 in Q ∨ ] . H^*(Q)\xrightarrow{F\text{-iso}}\mathbb{F}_p[b_{ts}:\xi_t^{p^s}\ne0\text{ in }Q^\vee]. H ∗ ( Q ) F -iso F p [ b t s : ξ t p s = 0 in Q ∨ ] .
Any inclusion Q ↪ Q ′ Q\hookrightarrow Q' Q ↪ Q ′ induces the obvious map on the polynomial part of the cohomology, so one can then compute the limit of the cohomology algebras of these, at least up to F F F -isomorphism. So if the conjecture holds, one would get a close analog to what happens at the prime 2, computationally as well as theoretically. In particular, one should be able to deduce the following from the previous conjecture:
Conjecture 7.2. Let S = F p [ b t s : s < t ] S=\mathbb{F}_p[b_{ts}:s<t] S = F p [ b t s : s < t ] . Let J ⊆ S J\subseteq S J ⊆ S be the ideal
J = ( b t s b v u : u ≥ t unless s = 0 and u = t ) . J=(b_{ts}b_{vu}:u\ge t\text{ unless }s=0\text{ and }u=t). J = ( b t s b v u : u ≥ t unless s = 0 and u = t ) . Then there is an F F F -isomorphism lim H ∗ ( Q ) → S / J \lim H^*(Q)\to S/J lim H ∗ ( Q ) → S / J , where the limit is taken over the quasi-elementary sub-Hopf algebras of P P P . There is an action of P P P on S / J S/J S / J ; this F F F -isomorphism is a maps of P P P -algebras, and the induced map on the invariant subalgebras is again an F F F -isomorphism.
Combined with Conjecture 6.1 , one should be able to see that there is an F F F -isomorphism
H ∗ ( D ∩ P ) → lim H ∗ ( Q ) , H^*(D\cap P)\to\lim H^*(Q), H ∗ ( D ∩ P ) → lim H ∗ ( Q ) ,
which is a map of P P P -algebras. Compare this to the p = 2 p = 2 p = 2 computations of H ∗ ( D ) H^*(D) H ∗ ( D ) and lim H ∗ ( E ) \lim H^*(E) lim H ∗ ( E ) up to F F F -isomorphism in [20 Proposition 6.5.1] .
Here is some of what we know about the status of Conjecture 7.1 . First, if for some r r r the profile function of a Hopf algebra B B B has the form
( 0 , … , 0 ⏟ r − 2 , 1 , b r , b r + 1 , … ) (\underbrace{0,\ldots,0}_{r-2},1,b_r,b_{r+1},\ldots) ( r − 2 0 , … , 0 , 1 , b r , b r + 1 , … )
and b i > r b_i > r b i > r for some i i i , then B B B cannot be quasi-elementary by Lemmas 2.5 and 5.6 . As a result, if 1 is the first nonzero entry in the profile function, then the Hopf algebra is quasi-elementary if and only if it is of the form given in (b). So to address the classification question, we need to consider Hopf algebras where the first nonzero entry is larger than 1. It turns out — see the paragraph after Question 7.3 — that the following case seems to be difficult: the case when the profile function looks like
( 0 , … , 0 , s t , 0 , … , 0 , s + t n , … ) , (0,\ldots,0,\underset{t}{s},0,\ldots,0,\underset{n}{s+t},\ldots), ( 0 , … , 0 , t s , 0 , … , 0 , n s + t , … ) ,
with 1 < s < t 1 < s < t 1 < s < t , so for example ( 0 , 2 , 0 , 4 , 0 , 2 ) (0,2,0,4,0,2) ( 0 , 2 , 0 , 4 , 0 , 2 ) . The author is not able to determine whether the corresponding Hopf algebra is quasi-elementary. If it is not, it lends support to the claimed classification. If it is, then note that it has a sub-Hopf algebra with profile function ( 0 , 1 , 0 , 4 , 0 , 2 ) (0,1,0,4,0,2) ( 0 , 1 , 0 , 4 , 0 , 2 ) which is not quasi-elementary, and it would perhaps be surprising to have B ⊆ Q ⊆ P B \subseteq Q \subseteq P B ⊆ Q ⊆ P with Q Q Q quasi-elementary and B B B not quasi-elementary.
The Nilpotence Lemma
As discussed, the Nilpotence Lemma 2.5 is an odd primary analog of a result of Lin’s [8 Theorem 3.1] . If one frames Lin’s result the right way, it can be stated as an if-and-only-if statement. Can one do the same thing here?
To set this up, let’s fix integers s s s , t t t , n n n , k k k with s < t s < t s < t and k < n k < n k < n — we want to consider whether b t s b n k b_{t s}b_{n k} b t s b nk is nilpotent in the cohomology of an appropriate Hopf algebra, and by Theorem A we know that b t s b_{t s} b t s is nilpotent if s ≥ t s \ge t s ≥ t , and similarly for b n k b_{n k} b nk . Let’s also assume that t < n t < n t < n . To keep things simple, let’s work with a Hopf algebra C C C whose profile function looks like
( 0 , … , 0 , s t , 0 , … , 0 , c n n , … ) (0,\ldots,0,\underset{t}{s},0,\ldots,0,\underset{n}{c_n},\ldots) ( 0 , … , 0 , t s , 0 , … , 0 , n c n , … )
where c n > k c_n > k c n > k . Therefore ξ s t p s \xi_{s t}^{p^s} ξ s t p s and ξ n p k \xi_{n}^{p^k} ξ n p k are nonzero and primitive in C ∨ C^\vee C ∨ .
At the prime 2, Lin’s theorem says that in this situation, h n k h t s h_{n k}h_{t s} h nk h t s is nilpotent if k ≥ t k \ge t k ≥ t , while if k < t k < t k < t , by restricting to an appropriate elementary sub-Hopf algebra one can see that h n k h t s h_{n k}h_{t s} h nk h t s is non-nilpotent. That is, with s s s , t t t , and n n n fixed, h t s h n k h_{t s}h_{n k} h t s h nk is nilpotent if and only if k ≥ t k \ge t k ≥ t .
When p p p is odd, here are two cases to consider:
(a) c n = ∞ c_n = \infty c n = ∞ : no power of ξ n \xi_n ξ n is zero in C ∨ C^\vee C ∨ ,
(b) c n > k c_n > k c n > k : ξ n p k ≠ 0 \xi_n^{p^k} \ne0 ξ n p k = 0 while some power of ξ n \xi_n ξ n is zero.
Unfortunately, in case (a) the author can say nothing beyond what is already in the lemma. For case (b), though, we can say more. We have the reduced coproduct ξ n + t ↦ ξ t p t ⊗ ξ n \xi_{n+t} \mapsto\xi_t^{p^t} \otimes\xi_n ξ n + t ↦ ξ t p t ⊗ ξ n , so we get the relation h n t h t 0 = 0 h_{n t}h_{t 0} = 0 h n t h t 0 = 0 in cohomology. If we apply Steenrod operations, we get more relations:
β P ~ 0 : b n t h t 1 = h n , t + 1 b n 0 , β P ~ 1 : b n t p b t 1 = b n , t + 1 b n 0 p , P ~ p : b n , t + 1 p 2 b t , 2 = b n , t + 2 b n 1 p 2 , \begin{aligned}
\beta\widetilde{\mathcal{P}}^0 &: \quad b_{n t}h_{t 1}=h_{n,t+1}b_{n0},\\
\beta\widetilde{\mathcal{P}}^1 &: \quad b_{n t}^{p}b_{t1}=b_{n,t+1}b_{n0}^{p},\\
\widetilde{\mathcal{P}}^p &: \quad b_{n,t+1}^{p^2}b_{t,2}=b_{n,t+2}b_{n1}^{p^2},
\end{aligned} β P 0 β P 1 P p : b n t h t 1 = h n , t + 1 b n 0 , : b n t p b t 1 = b n , t + 1 b n 0 p , : b n , t + 1 p 2 b t , 2 = b n , t + 2 b n 1 p 2 ,
and then apply P ~ p 2 \widetilde{\mathcal{P}}^{p^2} P p 2 , P ~ p 3 \widetilde{\mathcal{P}}^{p^3} P p 3 , etc. so in general for d ≥ 1 d \ge1 d ≥ 1 we get
b n t p d b t d = b n , t + d b n 0 p d . b_{n t}^{p^d}b_{t d}=b_{n,t+d}b_{n0}^{p^d}. b n t p d b t d = b n , t + d b n 0 p d .
Then with i ≥ 0 i \ge0 i ≥ 0 we apply ( P ~ 0 ) i (\widetilde{P}^{0})^{i} ( P 0 ) i to get
b n , t + i p d b t , d + i = b n , t + d + i b n , i p d . b_{n,t+i}^{p^{d}}b_{t,d+i}=b_{n,t+d+i}b_{n,i}^{p^{d}}. b n , t + i p d b t , d + i = b n , t + d + i b n , i p d .
If we let s = d + i s=d+i s = d + i , this becomes
b n , s + t − d p d b t , s = b n , s + t b t , s − d p d . b_{n,s+t-d}^{p^{d}}b_{t,s}=b_{n,s+t}b_{t,s-d}^{p^{d}}. b n , s + t − d p d b t , s = b n , s + t b t , s − d p d .
If ξ n p s + t = 0 \xi_{n}^{p^{s+t}}=0 ξ n p s + t = 0 , then the right side is zero, so we see that b n , s + t − d b t , s b_{n,s+t-d}b_{t,s} b n , s + t − d b t , s is nilpotent whenever s ≥ d ≥ 1 s\ge d\ge1 s ≥ d ≥ 1 . Letting k = s + t − d k=s+t-d k = s + t − d , we conclude that if ξ n p s + t = 0 \xi_{n}^{p^{s+t}}=0 ξ n p s + t = 0 , then b n k b t s b_{nk}b_{ts} b nk b t s is nilpotent whenever t ≤ k ≤ s + t − 1 t\le k\le s+t-1 t ≤ k ≤ s + t − 1 . The Nilpotence Lemma tells us that if k ≥ s + t + 1 k\ge s+t+1 k ≥ s + t + 1 (and ξ n p k ≠ 0 \xi_{n}^{p^{k}}\ne0 ξ n p k = 0 is nonzero), then b n k b t s b_{nk}b_{ts} b nk b t s is nilpotent. This leaves one open case: what happens if k = s + t k=s+t k = s + t ?
Question 7.3. Suppose that the sub-Hopf algebra C C C of P P P has a finite profile function, and assume that ξ t p s \xi_{t}^{p^{s}} ξ t p s and ξ n p s + t \xi_{n}^{p^{s+t}} ξ n p s + t are nonzero and primitive in C ∨ C^{\vee} C ∨ . Suppose that s < t s<t s < t and s + t < n s+t<n s + t < n . Suppose further that ξ t p s + t = 0 \xi_{t}^{p^{s+t}}=0 ξ t p s + t = 0 and ξ n p s + t + 1 = 0 \xi_{n}^{p^{s+t+1}}=0 ξ n p s + t + 1 = 0 . Is b t s b n , s + t b_{ts}b_{n,s+t} b t s b n , s + t nilpotent?
For example, is b 21 b 43 b_{21}b_{43} b 21 b 43 nilpotent in the Hopf algebra with profile function ( 0 , 2 , 0 , 4 , 0 , 2 ) (0,2,0,4,0,2) ( 0 , 2 , 0 , 4 , 0 , 2 ) ? Answering this should determine whether this Hopf algebra is quasi-elementary. In any case, answering the question would presumably lead to a classification of quasi-elementary sub-Hopf algebras of P P P .
Note that the situation with products of the form a t b n k a_{t}b_{nk} a t b nk is simpler. Assume that k < n k<n k < n . Then a t b n k a_{t}b_{nk} a t b nk is nilpotent if and only if k ≥ t k\ge t k ≥ t : the k ≥ t k\ge t k ≥ t case comes from the Nilpotence Lemma, while if k < t k<t k < t , then a t b n k a_{t}b_{nk} a t b nk is detected in the cohomology of an elementary sub-Hopf algebra of A A A and hence is non-nilpotent.
Remark 7.4. Therefore as far as quasi-elementary sub-Hopf algebras of A A A go, if we can understand the quasi-elementary sub-Hopf algebras of P P P , we should also be able to understand those in A A A . If Conjecture 7.1 is correct, for example, then the maximal quasi-elementary sub-Hopf algebras of A A A should be the ones with profile function
( ( 0 , … , 0 ⏟ r − 1 , 1 , r + 1 , r + 1 , r + 1 , … ) ; ( 1 , … , 1 ⏟ r + 1 , 2 , 2 , 2 , … ) ) \left((\underbrace{0,\ldots,0}_{r-1},1,r+1,r+1,r+1,\ldots);(\underbrace{1,\ldots,1}_{r+1},2,2,2,\ldots)\right) ( r − 1 0 , … , 0 , 1 , r + 1 , r + 1 , r + 1 , … ) ; ( r + 1 1 , … , 1 , 2 , 2 , 2 , … )
for some r ≥ 1 r\ge1 r ≥ 1 , along with the elementary Hopf algebras E ( − 1 ) E(-1) E ( − 1 ) and E ( 0 ) E(0) E ( 0 ) from Subsection 2.1.
Computations in H ∗ ( D ) H^{*}(D) H ∗ ( D )
We note that D D D is the sub-Hopf algebra generated by the elementary sub-Hopf algebras of A A A , and also generated by the quasi-elementary sub-Hopf algebras of A A A . It is also the sub-Hopf algebra generated by the Margolis elements P s t P_{s}^{t} P s t with s < t s<t s < t along with the Milnor primitives Q n Q_{n} Q n , and those elements are precisely the ones which determine projectivity and vanishing lines in cohomology [14 ] . So it is natural that D D D is so prominent in this work.
It would be good to understand something about the cohomology of H ∗ ( D ) H^{*}(D) H ∗ ( D ) in order to provide examples and applications of Theorems A A A and B B B . We recall an old observation, that for a graded connected Hopf algebra B B B , the vector space H 1 ( B ) H^{1}(B) H 1 ( B ) is isomorphic to the primitives in B ∨ B^{\vee} B ∨ , with a primitive x ∈ B ∨ x\in B^{\vee} x ∈ B ∨ corresponding to the cohomology class represented by [ x ] [x] [ x ] in the cobar complex. This was originally probably a folk theorem. It can be deduced by dualizing the results from Wall [22 p. 442] , together with Milnor and Moore’s observation [13 Proposition 3.10] that the indecomposables in B B B are dual to the primitives in B ∨ B^{\vee} B ∨ .
Proposition 7.5. ξ t p s \xi_{t}^{p^{s}} ξ t p s is primitive in D ∨ D^{\vee} D ∨ if and only if s ≥ ⌊ t / 3 ⌋ s\ge\lfloor t/3\rfloor s ≥ ⌊ t /3 ⌋ . τ n \tau_{n} τ n is primitive in D ∨ D^{\vee} D ∨ if and only if n = 0 n=0 n = 0 .
The condition on ξ t p s \xi_{t}^{p^{s}} ξ t p s being primitive is the same as given by Harker in the p = 2 p=2 p = 2 case [4 2.7–2.8] .
Proof. This is just a computation in D ∨ = A / ( ξ k p k : k ≥ 1 ) D^{\vee}=A/(\xi_{k}^{p^{k}}:k\ge1) D ∨ = A / ( ξ k p k : k ≥ 1 ) . The primitivity of τ n \tau_{n} τ n if and only if n = 0 n=0 n = 0 is clear. It remains to show that ξ 3 n p s \xi_{3n}^{p^{s}} ξ 3 n p s , ξ 3 n + 1 p s \xi_{3n+1}^{p^{s}} ξ 3 n + 1 p s , and ξ 3 n + 2 p s \xi_{3n+2}^{p^{s}} ξ 3 n + 2 p s are primitive in D ∨ D^{\vee} D ∨ if and only if s ≥ n s\ge n s ≥ n . There are three cases and they are very similar, so we do the last one. The reduced coproduct on ξ 3 n + 2 \xi_{3n+2} ξ 3 n + 2 is
∑ j = 1 3 n + 1 ξ 3 n + 2 − j p j ⊗ ξ j . \sum_{j=1}^{3n+1} \xi_{3n+2-j}^{p^j} \otimes\xi_j. j = 1 ∑ 3 n + 1 ξ 3 n + 2 − j p j ⊗ ξ j .
We need to show that the coproduct on ξ 3 n + 2 p n − 1 \xi_{3n+2}^{p^{n-1}} ξ 3 n + 2 p n − 1 is nonzero, while the coproduct on ξ 3 n + 2 p n \xi_{3n+2}^{p^n} ξ 3 n + 2 p n is zero. If j = n , n + 1 j=n,n+1 j = n , n + 1 , we get the terms
ξ 2 n + 2 p n ⊗ ξ n + ξ 2 n + 1 p n + 1 ⊗ ξ n + 1 . \xi_{2n+2}^{p^n} \otimes\xi_n + \xi_{2n+1}^{p^{n+1}} \otimes\xi_{n+1}. ξ 2 n + 2 p n ⊗ ξ n + ξ 2 n + 1 p n + 1 ⊗ ξ n + 1 .
If we raise these to the power p n − 1 p^{n-1} p n − 1 , the terms are nonzero:
ξ 2 n + 2 p 2 n − 1 ⊗ ξ n p n − 1 + ξ 2 n + 1 p 2 n ⊗ ξ n + 1 p n − 1 . \xi_{2n+2}^{p^{2n-1}} \otimes\xi_n^{p^{n-1}} + \xi_{2n+1}^{p^{2n}} \otimes\xi_{n+1}^{p^{n-1}}. ξ 2 n + 2 p 2 n − 1 ⊗ ξ n p n − 1 + ξ 2 n + 1 p 2 n ⊗ ξ n + 1 p n − 1 .
Therefore ξ 3 n + 2 p n − 1 \xi_{3n+2}^{p^{n-1}} ξ 3 n + 2 p n − 1 is not primitive.
Now consider the coproduct on ξ 3 n + 2 p n \xi_{3n+2}^{p^n} ξ 3 n + 2 p n : the j j j th term is
ξ 3 n + 2 − j p j + n ⊗ ξ j p n . \xi_{3n+2-j}^{p^{j+n}} \otimes\xi_j^{p^n}. ξ 3 n + 2 − j p j + n ⊗ ξ j p n .
If j ≤ n j \le n j ≤ n , then the second factor is zero. If j ≥ n + 1 j \ge n+1 j ≥ n + 1 , then the first factor is zero: if j ≥ n + 1 j \ge n+1 j ≥ n + 1 , then 3 n + 2 − j ≤ 2 n + 1 3n+2-j \le2n+1 3 n + 2 − j ≤ 2 n + 1 while j + n ≥ 2 n + 1 j+n \ge2n+1 j + n ≥ 2 n + 1 . The computations for ξ 3 n \xi_{3n} ξ 3 n and ξ 3 n + 1 \xi_{3n+1} ξ 3 n + 1 are similar and a bit simpler than this one.
Corollary 7.6. H 1 ( D ) H^1(D) H 1 ( D ) is the vector space spanned by { h t s : s ≥ ⌊ t / 3 ⌋ } ∪ { a 0 } \{h_{ts}:s \ge\lfloor t/3 \rfloor\} \cup\{a_0\} { h t s : s ≥ ⌊ t /3 ⌋} ∪ { a 0 } .
The author believes that the following is true, and we provide a sketch of a proof. Some details are lacking, though.
Guess 7.7. For each n ≥ 1 n \ge1 n ≥ 1 , the element b n , n − 1 ∈ H 2 ( D ) b_{n,n-1} \in H^2(D) b n , n − 1 ∈ H 2 ( D ) is invariant under the A A A -action. Therefore some p p p th power of b n , n − 1 b_{n,n-1} b n , n − 1 lifts to H ∗ ( A ) H^*(A) H ∗ ( A ) .
Sketch of proof. We first note that b n , n − 1 ∈ H 2 ( D ) b_{n,n-1} \in H^2(D) b n , n − 1 ∈ H 2 ( D ) is detected by the restriction map to H ∗ ( D ( P n − 1 ) ) H^*(D(P^{n-1})) H ∗ ( D ( P n − 1 )) , and so is non-nilpotent.
The action of A A A on H ∗ ( D ) H^*(D) H ∗ ( D ) comes from the action of A A A on D D D itself. It is convenient to recast this as a coaction of A ∨ A^\vee A ∨ on H ∗ ( D ) H^*(D) H ∗ ( D ) . As at the prime 2 — see Lemmas A.3 and A.5 of [17 ] — the coaction on an element h t s h_{ts} h t s comes from the coproduct on ξ t p s \xi_t^{p^s} ξ t p s and so has the form 1 ⊗ h t s + ∑ i , j x ⊗ h i j 1 \otimes h_{ts}+\sum_{i,j} x \otimes h_{ij} 1 ⊗ h t s + ∑ i , j x ⊗ h ij where x ∈ A x \in A x ∈ A and i < t i<t i < t , j ≥ s j \ge s j ≥ s . In particular, the elements h n , n − 1 h_{n,n-1} h n , n − 1 are invariant. We would like at this point to appeal to the naturality of May’s Steenrod operations, but it is not clear how to do that. Naturality should imply that this coaction commutes with the Steenrod operations on Hopf algebra cohomology, so the same sort of coaction formula should hold for b t s b_{ts} b t s , and therefore b n , n − 1 b_{n,n-1} b n , n − 1 should be invariant for each n n n . That is, the subalgebra of A A A -invariants in H ∗ ( D ) H^*(D) H ∗ ( D ) should be closed under the action of May’s Steenrod operations, so since each h n , n − 1 h_{n,n-1} h n , n − 1 is invariant, each b n , n − 1 b_{n,n-1} b n , n − 1 should also be.
In addition, if Conjecture 7.2 is correct (see also Remark 7.4), then for any n ≥ 2 n \ge2 n ≥ 2 one would get invariant monomials of the form
b n − 1 , 0 j n − 1 b n , n − 1 j n b n + 1 , n j n + 1 ⋯ b 2 n − 1 , n − 2 j 2 n − 1 , b_{n-1,0}^{j_{n-1}}b_{n, n-1}^{j_n}b_{n+1,n}^{j_{n+1}}\cdots b_{2n-1,n-2}^{j_{2n-1}}, b n − 1 , 0 j n − 1 b n , n − 1 j n b n + 1 , n j n + 1 ⋯ b 2 n − 1 , n − 2 j 2 n − 1 ,
as long as j n − 1 > 0 j_{n-1}>0 j n − 1 > 0 .
Other results
If one would like results that do not rely on the unpublished work [3 ] , then we note that this was only used in § 4 . In particular nothing in § 5 relies on it, so we have:
Theorem 7.8. Suppose that C ⊆ A C \subseteq A C ⊆ A has a finite profile function. Then
ι ˉ ∗ : Ext C ∗ ( F p , F p ) ⟶ ( Ext C ∩ D ∗ ( F p , F p ) ) C / / C ∩ D \bar{\iota}^{*}: \operatorname{Ext}^{*}_{C}(\mathbb{F}_{p},\mathbb{F}_{p}) \longrightarrow\left(\operatorname{Ext}^{*}_{C\cap D}(\mathbb{F}_{p},\mathbb{F}_{p})\right)^{C//C\cap D} ι ˉ ∗ : Ext C ∗ ( F p , F p ) ⟶ ( Ext C ∩ D ∗ ( F p , F p ) ) C // C ∩ D
is an F F F -isomorphism.
We prove the Burklund-Levy-Pstrągowski theorem when p = 3 p = 3 p = 3 in the appendix, so we have:
Theorem 7.9. Fix p = 3 p = 3 p = 3 and suppose that C C C is a sub-Hopf algebra of A A A .
ι ˉ ∗ : Ext C ∗ ( F p , F p ) ⟶ ( Ext C ∩ D ∗ ( F p , F p ) ) C / / C ∩ D \bar{\iota}^{*}: \operatorname{Ext}^{*}_{C}(\mathbb{F}_{p},\mathbb{F}_{p}) \longrightarrow\left(\operatorname{Ext}^{*}_{C\cap D}(\mathbb{F}_{p},\mathbb{F}_{p})\right)^{C//C\cap D} ι ˉ ∗ : Ext C ∗ ( F p , F p ) ⟶ ( Ext C ∩ D ∗ ( F p , F p ) ) C // C ∩ D is an F F F -isomorphism.
Nilpotence of b n n b_{nn} b nn when p = 3 p = 3 p = 3
We fix p = 3 p = 3 p = 3 for this section. The goal is to provide a proof of the nilpotence of b n n b_{nn} b nn in H ∗ ( P ≥ n ) H^{*}(P_{\geq n}) H ∗ ( P ≥ n ) , independent of Theorem 4.1.
We start by mentioning that Nakamura proved that b 11 ∈ H ∗ ( P ) b_{11} \in H^{*}(P) b 11 ∈ H ∗ ( P ) is nilpotent. More precisely, he showed [15 Proposition 1.1c)] that b 11 2 = h 11 x b_{11}^{2} = h_{11}x b 11 2 = h 11 x for some x x x , and since h 11 2 = 0 h_{11}^{2} = 0 h 11 2 = 0 , one immediately deduces that b 11 4 = 0 b_{11}^{4} = 0 b 11 4 = 0 . His proof is a calculation in the cobar complex, and it is not easily generalizable (at least not by the author) to either higher primes or to b n n b_{nn} b nn for larger values of n n n . His element x x x looks like a Massey product, though, and that observation inspired the proof of the following result.
Theorem A.1. If p = 3 p = 3 p = 3 , then b n n ∈ H 2 ( P ≥ n ) b_{nn} \in H^{2}(P_{\geq n}) b nn ∈ H 2 ( P ≥ n ) is nilpotent.
The proof here follows the sketch in [20 Section B.3.1] .
Proof. We work with P ≥ n P_{\geq n} P ≥ n and its cohomology. The reduced coproduct ξ 2 n ↦ ξ n p ⊗ ξ n \xi_{2n} \mapsto\xi_{n}^{p} \otimes\xi_{n} ξ 2 n ↦ ξ n p ⊗ ξ n yields the relation
h n n h n 0 = 0. h_{nn}h_{n0}=0. h nn h n 0 = 0.
If we apply Steenrod operations to this, we get the following, where each line is labeled by the Steenrod operation which has been applied:
β P ~ 0 : b n n h n 1 − h n , n + 1 b n 0 = 0 , P ~ 1 : b n n p h n 2 − h n , n + 2 b n 0 p = 0 , P ~ p : b n n p 2 h n 3 − h n , n + 3 b n 0 p 2 = 0 , ⋮ P ~ p k − 2 : b n n p k − 1 h n k − h n , n + k b n 0 p k − 1 = 0. \begin{aligned}
\beta\widetilde{P}^{0}:&\ b_{nn}h_{n1}-h_{n,n+1}b_{n0}=0,\\
\widetilde{P}^{1}:&\ b_{nn}^{p}h_{n2}-h_{n,n+2}b_{n0}^{p}=0,\\
\widetilde{P}^{p}:&\ b_{nn}^{p^{2}}h_{n3}-h_{n,n+3}b_{n0}^{p^{2}}=0,\\
&\ \vdots\\
\widetilde{P}^{p^{k-2}}:&\ b_{nn}^{p^{k-1}}h_{nk}-h_{n,n+k}b_{n0}^{p^{k-1}}=0.
\end{aligned} β P 0 : P 1 : P p : P p k − 2 : b nn h n 1 − h n , n + 1 b n 0 = 0 , b nn p h n 2 − h n , n + 2 b n 0 p = 0 , b nn p 2 h n 3 − h n , n + 3 b n 0 p 2 = 0 , ⋮ b nn p k − 1 h nk − h n , n + k b n 0 p k − 1 = 0.
That is, b n n p k − 1 h n k = h n , n + k b n 0 p k − 1 b_{nn}^{p^{k-1}}h_{nk}=h_{n,n+k}b_{n0}^{p^{k-1}} b nn p k − 1 h nk = h n , n + k b n 0 p k − 1 for any k ≥ 1 k \geq1 k ≥ 1 . So letting k = n k=n k = n , we have
h n n b n n p n − 1 = h n , 2 n b 10 p n − 1 , (1) h_{nn}b_{nn}^{p^{n-1}}=h_{n,2n}b_{10}^{p^{n-1}}, \tag*{(1)} h nn b nn p n − 1 = h n , 2 n b 10 p n − 1 , (1)
and letting k = 2 n k = 2n k = 2 n , we have
b n 0 p 2 n − 1 h n , 2 n = h n , 3 n b n 0 p 2 n − 1 . b_{n0}^{p^{2n-1}}h_{n,2n}=h_{n,3n}b_{n0}^{p^{2n-1}}. b n 0 p 2 n − 1 h n , 2 n = h n , 3 n b n 0 p 2 n − 1 .
We apply both of these relations to get
h n n b n n p n − 1 + p 2 n − 1 = h n , 2 n b n 0 p n − 1 b n n p 2 n − 1 = h n , 3 n b n 0 p n − 1 + p 2 n − 1 . (2) \begin{aligned}
h_{nn}b_{nn}^{p^{n-1}+p^{2n-1}}&=h_{n,2n}b_{n0}^{p^{n-1}}b_{nn}^{p^{2n-1}}\\
&=h_{n,3n}b_{n0}^{p^{n-1}+p^{2n-1}}. \tag*{(2)}
\end{aligned} h nn b nn p n − 1 + p 2 n − 1 = h n , 2 n b n 0 p n − 1 b nn p 2 n − 1 = h n , 3 n b n 0 p n − 1 + p 2 n − 1 . (2)
(More generally, h n n b n n M = h n , m n b n 0 M h_{nn}b_{nn}^{M}=h_{n,mn}b_{n0}^{M} h nn b nn M = h n , mn b n 0 M , where M = p n − 1 + p 2 n − 1 + ⋯ + p ( m − 1 ) n − 1 M=p^{n-1}+p^{2n-1}+\cdots+p^{(m-1)n-1} M = p n − 1 + p 2 n − 1 + ⋯ + p ( m − 1 ) n − 1 , but we don’t need this.) Now we have the Massey product ⟨ h n , 3 n , h n , 2 n , h n n ⟩ \langle h_{n,3n},h_{n,2n},h_{nn}\rangle ⟨ h n , 3 n , h n , 2 n , h nn ⟩ , defined because h n , n + k h n k = 0 h_{n,n+k}h_{nk}=0 h n , n + k h nk = 0 for all k k k : apply the algebra map ( P 0 ) k (\mathcal{P}^{0})^{k} ( P 0 ) k to the relation h n n h n 0 = 0 h_{nn}h_{n0}=0 h nn h n 0 = 0 . We have the reduced coproduct
ξ 3 n ↦ ∑ i = 1 3 n − 1 ξ 3 n − i p i ξ i = ∑ i = n 2 n ξ 3 n − i p i ξ i = ξ 2 n p n ξ n + ξ n p 2 n ⊗ ξ 2 n + ∑ i = n + 1 2 n − 1 ξ 3 n − i p i ⊗ ξ i . (3) \begin{aligned}
\xi_{3n}&\mapsto\sum_{i=1}^{3n-1}\xi_{3n-i}^{p^{i}}\xi_i\\
&=\sum_{i=n}^{2n}\xi_{3n-i}^{p^{i}}\xi_i\\
&=\xi_{2n}^{p^n}\xi_n+\xi_n^{p^{2n}}\otimes\xi_{2n}+\sum_{i=n+1}^{2n-1}\xi_{3n-i}^{p^i}\otimes\xi_i. \tag*{(3)}
\end{aligned} ξ 3 n ↦ i = 1 ∑ 3 n − 1 ξ 3 n − i p i ξ i = i = n ∑ 2 n ξ 3 n − i p i ξ i = ξ 2 n p n ξ n + ξ n p 2 n ⊗ ξ 2 n + i = n + 1 ∑ 2 n − 1 ξ 3 n − i p i ⊗ ξ i . (3)
We note that if n ≤ i ≤ 2 n − 1 n\leq i\leq2n-1 n ≤ i ≤ 2 n − 1 , then ξ i \xi_i ξ i is primitive in P ≥ n ∨ P^\vee_{\geq n} P ≥ n ∨ . Hence the tensor factors in the sum give cohomology classes, and the above coproduct on ξ 3 n \xi_{3n} ξ 3 n gives us this relation in cohomology:
0 = ⟨ h n , 2 n , h n n , h n 0 ⟩ + ∑ i = n + 1 2 n − 1 h 3 n − i , i h i , 0 . 0=\langle h_{n,2n},h_{nn},h_{n0}\rangle+\sum_{i=n+1}^{2n-1}h_{3n-i,i}h_{i,0}. 0 = ⟨ h n , 2 n , h nn , h n 0 ⟩ + i = n + 1 ∑ 2 n − 1 h 3 n − i , i h i , 0 .
Apply P 0 \mathcal{P}^{0} P 0 to get
⟨ h n , 3 n , h n , 2 n , h n n ⟩ = − ∑ i = n + 1 2 n − 1 h 3 n − i , i + 1 h i , 1 . \langle h_{n,3n},h_{n,2n},h_{nn}\rangle=-\sum_{i=n+1}^{2n-1}h_{3n-i,i+1}h_{i,1}. ⟨ h n , 3 n , h n , 2 n , h nn ⟩ = − i = n + 1 ∑ 2 n − 1 h 3 n − i , i + 1 h i , 1 .
Note that the right side is a product of exterior elements, and so it is nilpotent; thus ⟨ h n , 3 n , h n , 2 n , h n n ⟩ \langle h_{n,3n},h_{n,2n},h_{nn}\rangle ⟨ h n , 3 n , h n , 2 n , h nn ⟩ is nilpotent.
Multiply this Massey product by b n 0 N b_{n0}^{N} b n 0 N where N = p 2 n − 1 + p n − 1 + p n − 1 N=p^{2n-1}+p^{n-1}+p^{n-1} N = p 2 n − 1 + p n − 1 + p n − 1 : first we get
b n 0 N ⟨ h n , 3 n , h n , 2 n , h n n ⟩ ∈ ⟨ h n , 3 n b n 0 p 2 n − 1 + p n − 1 , h n , 2 n b n 0 p n − 1 , h n n ⟩ . b_{n0}^{N}\langle h_{n,3n},h_{n,2n},h_{nn}\rangle\in\langle h_{n,3n}b_{n0}^{p^{2n-1}+p^{n-1}},h_{n,2n}b_{n0}^{p^{n-1}},h_{nn}\rangle. b n 0 N ⟨ h n , 3 n , h n , 2 n , h nn ⟩ ∈ ⟨ h n , 3 n b n 0 p 2 n − 1 + p n − 1 , h n , 2 n b n 0 p n − 1 , h nn ⟩ .
The Massey product on the left side consists of a single element, for degree reasons. The Massey product on the right is a set, containing the single element on the left. Now we apply (A.2) and (A.3):
b n 0 N ⟨ h n , 3 n , h n , 2 n , h n n ⟩ ∈ ⟨ h n , 3 n b n 0 p 2 n − 1 + p n − 1 , h n , 2 n b n 0 p n − 1 , h n n ⟩ = ⟨ h n n b n n p 2 n − 1 + p n − 1 , h n n b n n p n − 1 , h n n ⟩ ∋ b n n N + 1 ⟨ h n n , h n n , h n n ⟩ = b n n N + 1 . \begin{aligned}
b_{n0}^{N}\langle h_{n,3n},h_{n,2n},h_{nn}\rangle
&\in\langle h_{n,3n}b_{n0}^{p^{2n-1}+p^{n-1}},h_{n,2n}b_{n0}^{p^{n-1}},h_{nn}\rangle\\
&=\langle h_{nn}b_{nn}^{p^{2n-1}+p^{n-1}},h_{nn}b_{nn}^{p^{n-1}},h_{nn}\rangle\\
&\ni b_{nn}^{N+1}\langle h_{nn},h_{nn},h_{nn}\rangle=b_{nn}^{N+1}.
\end{aligned} b n 0 N ⟨ h n , 3 n , h n , 2 n , h nn ⟩ ∈ ⟨ h n , 3 n b n 0 p 2 n − 1 + p n − 1 , h n , 2 n b n 0 p n − 1 , h nn ⟩ = ⟨ h nn b nn p 2 n − 1 + p n − 1 , h nn b nn p n − 1 , h nn ⟩ ∋ b nn N + 1 ⟨ h nn , h nn , h nn ⟩ = b nn N + 1 .
Therefore
b n n N + 1 − b n 0 N ⟨ h n , 3 n , h n , 2 n , h n n ⟩ b_{nn}^{N+1}-b_{n0}^{N}\langle h_{n,3n},h_{n,2n},h_{nn}\rangle b nn N + 1 − b n 0 N ⟨ h n , 3 n , h n , 2 n , h nn ⟩
is in the indeterminacy of the Massey product
⟨ h n n b n n p 2 n − 1 + p n − 1 , h n n b n n p n − 1 , h n n ⟩ . \langle h_{nn}b_{nn}^{p^{2n-1}+p^{n-1}},h_{nn}b_{nn}^{p^{n-1}},h_{nn}\rangle. ⟨ h nn b nn p 2 n − 1 + p n − 1 , h nn b nn p n − 1 , h nn ⟩ .
We claim that every element in the indeterminacy of this Massey product is nilpotent.
Lemma A.5. Suppose that Ω \Omega Ω is a differential graded algebra over a field whose characteristic is not two, and assume that the cohomology algebra H ∗ Ω H^{*}\Omega H ∗ Ω is graded commutative. If x x x and z z z are odd-dimensional classes in H ∗ Ω H^{*}\Omega H ∗ Ω and if ⟨ x , y , z ⟩ \langle x,y,z\rangle ⟨ x , y , z ⟩ is defined, then every element in the indeterminacy of ⟨ x , y , z ⟩ \langle x,y,z\rangle ⟨ x , y , z ⟩ is nilpotent.
Proof. The indeterminacy of a length 3 Massey product has an explicit formula, given by May in [10 ] : every element in the indeterminacy of ⟨ x , y , z ⟩ \langle x,y,z\rangle ⟨ x , y , z ⟩ has the form x a + b z xa+bz x a + b z , where a a a and b b b are cohomology classes in the appropriate dimensions. Since x x x and z z z are odd-dimensional cohomology classes in a graded commutative algebra and since we are not working in characteristic two, both x x x and z z z are nilpotent and hence so is every element in the indeterminacy.
Since ⟨ h n , 3 n , h n , 2 n h n n ⟩ \langle h_{n,3n},h_{n,2n}h_{nn}\rangle ⟨ h n , 3 n , h n , 2 n h nn ⟩ is nilpotent, we see that b n N + 1 b_{n}^{N+1} b n N + 1 is the sum of two nilpotent elements, and hence is nilpotent itself.
This proof does not easily generalize to larger primes for two reasons: first, at larger primes the various terms in (A.4) are not all primitive and so do not correspond to cohomology classes. Second, Lemma A.5 is not known to hold at larger primes. See [20 Conjectures B.3.9–10] for a possible replacement for the lemma.