Arxiv:2003.12003V3 [Math.AT] 20 Apr 2021

arXiv:2003.12003v3 [math.AT] 20 Apr 2021 ahmtclSine o upr n optlt uigt during hospitality and i support for Universitet Sciences Mathematical Stockholms Högskolan and Tekniska Institute Kungliga Research Sciences Mathematical wo the at on Foundat residence part Science in National the based is organisations: following paper the this in described alge mathematics an The fine-tune me helped Steen [DHH11]. who the Brown to with Ken working brackets; Toda about use amount to enormous an learnt I whom ii nJnay2020. January Pla in Max visit the a EP/R014604/1); number grant EPSRC by (supported .Rcletoso h terdagbaadfiiesbofa subHopf finite and algebra for relations Steenrod Wall the The on Recollections 4. manifolds for Poincaré duality modules Cyclic duality Spanier-Whitehead and Poincaré duality duality Spanier-Whitehead Duality 3. Doubling .Cl n CW and Cell of 2. theory Homotopy 1. notations & Conventions Introduction Date e od n phrases. and words Key 2010 ol iet hn h olwn o epu omnsadi and comments helpful for following the thank to like would I OOOYTER FMDLSOE COMMUTATIVE A OVER MODULES OF THEORY HOMOTOPY atudtd2/422 eso 3 version – 20/04/2021 updated last : ahmtc ujc Classification. Subject Mathematics oue vrte2lclcnetv pcrmo topologica of spectrum connective 2-local the over modules lsl eae oMhwl’ hoyof theory Mahowald’s to related closely hois n eosrt hmb osdrn oue over modules considering by them demonstrate and theories’ explored. ca be homotopy can or theory derived homotopy their which to sp passing of spectrum; categories sphere classical the T the of spectra. generalisations ring commutative are coherent which highly o study to introduction way the new allow products smash monoidal symmetric Abstract. napandsqe ewl pl hs ehiust h much the to techniques these apply will we sequel planned a In nti ae edsrb oeo h ol vial o stud for available tools the of some describe we paper this In S oenctgre fsetasc sta fEmnofe le al et Elmendorf of that as such spectra of categories Modern AGBA OETOSADEXAMPLES AND TOOLS SOME -ALGEBRA: R -modules tbehmtp hoy terdalgebra. Steenrod theory, homotopy Stable A ( R n -modules ) rmr 54;Scnay5P3 51,55S20. 55S10, 55P43, Secondary 55P42; Primary NRWBAKER ANDREW bo Contents -resolutions. 1 o ne rn o 927 0 hl h uhrwsin was author the while 000 0932078 No. Grant under ion eprogramme he nBree aionadrn h pig21 semester; 2014 Spring the during California Berkeley in pigo 08 h sa etnIsiuefor Institute Newton Isaac the 2018; of Spring n kcridotwieteato a upre by supported was author the while out carried rk cr htcrepn omdlsover modules to correspond that ectra ri eut ieHl h rwm attention my drew who Hill Mike result; braic o ler;PtrEce h agtm how me taught who Eccles Peter algebra; rod c nttt o ahmtc nBn during Bonn in Mathematics for Institute nck sgt:BbBue n onRge from Rognes John and Bruner Bob nsights: oua forms. modular l h connective the ymti ood rvdn a providing monoids symmetric f eoislast e otxsin contexts new to leads tegories igtee‘rv e homotopy new ‘brave these ying eehv aeoiso modules of categories have hese oooyHresn ihrStructures Higher Harnessing Homotopy gba 15 lgebras esfmla otx of context familiar less upe ihstrictly with quipped K ter spectrum, -theory ri:03103. arXiv:2003.12003 14 10 16 15 8 7 7 5 3 2 2 5. kO-modules and A(1)-modules 16 Realisation of cyclic A(1)-modules 22 Constructions of some more kO-modules 23 Some sample calculations and examples 24 6. Examples based on homogeneous spaces 29 References 34

Introduction

Modern categories of spectra such as that of Elmendorf et al [EKMM97] equipped with strictly symmetric monoidal smash products allow for the introduction of symmetric monoids giving a new way to study highly coherent commutative ring spectra. In turn these have categories of modules which are generalisations of the classical categories of spectra (corresponding to modules over the sphere spectrum). For example, such categories have Quillen model structures and so homotopy (or derived) categories, thus allowing the study of ‘brave new homotopy theories’. This paper provides an introduction to some of the machinery available for engaging in this version of homotopy theory and Sections 1, 2 and 3 provide an overview on homotopy theory for R-modules over a commutative S-algebra R which should be sufficient for reading the present work. Although we only discuss connective spectra, many aspects also apply to non-connective settings with suitable modifications. As an example we consider the important case of kO (the 2-local connective real K-theory spectrum) which is related to Mahowald’s theory of bo-resolutions and we review some aspects of this from the present perspective. The case of tmf (the 2-local connective spectrum of topological modular forms) is largely waiting to be developed in the spirit of Mahowald’s work and in the planned sequel we will discuss this, focusing especially on examples associated with kO considered as a tmf-module. Throughout our aim will be to exhibit interesting phenomena with connections to classical homotopy theory. Since we use make use of modules over the mod 2 Steenrod algebra A∗ and its finite subHopf algebras A(n)∗, we give some algebraic background in Section 4.

Conventions & notations. In this paper we will mainly work locally at the prime 2, so in ∗ that context H will denote the mod 2 Eilenberg-Mac Lane spectrum HF2 and A the mod 2 Steenrod algebra. To avoid excessive display of gradings we will usually suppress cohomological degrees and write V for a cohomologically graded vector space V ∗; in particular we will often write A for k −k F the Steenrod algebra. The linear dual of V is DV where (DV ) = HomF2 (V , 2), and we write V [m] for graded vector space with (V [m])k = V k−m, so for the cohomology of a spectrum X, H∗(ΣmX) = H∗(X)[m]. For a connected graded algebra B∗ we will often just write B, and denote its positive degree part by B+. When working with left modules over a Hopf algebra B over a field we will write M ⊙ N to denote the vector space tensor product of two B-modules M and N equipped with the diagonal action defined using the coproduct in B; for a vector space V , M ⊗V will denote the left module with action obtained from the action on M. The product ⊙ defines the monoidal structure on 2 the category of left B-modules. It is well known that B⊙ N =∼ B⊗ N as left B-modules, so ⊙ descends to the stable module category StModB. When discussing modules we will often follow well established precedent and use diagrams such as those in the figure below which both represent A(1) as a left A(1)-module. We will usually interpret degrees as cohomological so that Steenrod actions are displayed pointing upwards and we usually suppress arrow heads; labels on vertices denote degrees but are often omitted.

6 6

5 5

4 4

3 3 3 3

2 2

Sq2 1 Sq2 1 Sq1 Sq1 0 0

1. Homotopy theory of R-modules

We adopt the terminology and notation of [EKMM97]. Initially we do not necessarily assume spectra are localised (or completed) at some prime although this possibility would not affect the generalities described here. Later we do focus on some specifically local aspects in order to incorporate notions from [BM04]. M n Let R be a connective commutative S-algebra. In the category of R-modules R let SR denote the functorial cofibrant replacement of the suspension ΣnR. When R = S we will often suppress S from notation and for an R-module M also write M for the underlying S-module.

A morphism f : X → Y in M = MS gives rise to a morphism R ∧ X → R ∧ Y in MR, namely 1r ∧ f. If M is an R-module, a morphism g : X → M in M gives rise to a morphism R ∧ X → M.

1r∧g R ∧ X / R ∧ M /6 M

For an R-module M, ∼ D n ∼ D n πn(M) = S(S , M) = R(SR, M).

Now suppose that R admits a morphism of commutative S-algebras R → H = HFp; for Z Z example, we might have π0R = or π0R = (p). Then we can define a homology theory R H∗ (−) on DR by setting R H∗ M = π∗(H ∧R M) ∗ when M is cofibrant. This theory has a dual cohomology theory HR(−) defined by ∗ D HRM = R(M,H). 3 and which satisfies strict duality n ∼ R F HRM = HomFp (Hn M, p). One approach to calculating is by using the Künneth spectral sequence of [EKMM97],

2 H∗R F R (1.1) Es,t = Tors,t ( p,H∗M)=⇒ H∗ M which results from the isomorphism of H-modules

H ∧R M =∼ H ∧H∧R (H ∧ M). Taking M = H we obtain a spectral sequence H∗R F R Tors,t ( p,H∗H)=⇒ H∗ H which is known to be multiplicative. In a case where the induced homomorphism H∗R → H∗H = A∗ is a monomorphism, this is especially useful. For p = 2, each of the cases R =

HZ, kO, kU, tmf, tmf1(3) has this property. In such cases the dual Steenrod algebra A∗ is a free module over H∗R and we obtain R ∼ F H∗ H = A∗ ⊗H∗R p = A∗//H∗R. Dually we have ∗ D HRH = R(H,H) and a spectral sequence s,t s,t F ∗ E2 = ExtH∗R(H∗H, p)=⇒ HRH. In the situation where A(p)∗ is a free module over H∗R, this gives n ∼ R F HRH = HomFp (Hn H, p). We also have ∗ ∗ H R ∼ A(p) ⊗ ∗ F . = HRH p ∗ As in the case R = S, HRH is a cocommutative Hopf algebra and its dual is a commutative ∗ R Hopf algebra. Furthermore, HRH has a natural left coaction on H∗ M which induces a right R ∗ ∗ action of H∗ H on HRM and a left action on HRM. Using the algebra isomorphism ∼ A −−→A= ◦; θ ↔ (χθ)◦

R ∗ R we can convert H∗ M into a left HRH-module with the grading convention that Hn M is in ∗ R cohomological degree −n (so positive degree elements of HRH act on H∗ M by lowering degrees). 2 By using a geometric resolution of Fp over H∗R to compute the E -term, it can be shown ∗ ∗ that (1.1) is a spectral sequence of left HRH-comodules and right HRH-modules. In [BL01] we showed that there is an Adams spectral sequence for computing DR(L, M) for ∗ ∗ R-modules L, M with L strongly dualisable. Dualising from HRH-comodules to HRH-modules this has the form

s,t s,t R R t−s (1.2) E (L, M) = Ext ∗ (H∗ L,H∗ M)=⇒ DR(Σ L, M) , 2 HRH ∗ ∗ ∗ ∗ where (−) denotes p-adic completion. When L = S0 we will set E , (M) =E , (L, M). R r b r Assuming appropriate ﬁniteness conditions this E2-term can be rewritten to give

b s,t s,t ∗ ∗ t−s (1.3) E = Ext ∗ (H M,H L)=⇒ DR(Σ L, M) . 2 HRH R R 4 b When M = R ∧ Z, this gives

s,t s,t R s,t ∗ ∗ E (L, R ∧ Z) = Ext ∗ (H∗ L,H∗Z) =∼ Ext ∗ (H Z,H L). 2 HRH HRH R

0 Finally, if L = SR,

s,t s,t s,t ∗ E (R ∧ Z) = Ext ∗ (Fp,H∗Z) =∼ Ext ∗ (H Z, Fp). 2 HRH HRH

This agrees with the classical Adams E2-term s,t ∗ F ∼ D t−s ExtA∗ (H (R ∧ Z), p)=⇒ πt−s(R ∧ Z) = R(SR ,R ∧ Z).

∗,∗ ∗,∗ Notice that Er (R) is a bi-graded commutative algebra and for any R-module M, Er (M) ∗,∗ is a spectral sequence over it. When A is an R-ring spectrum, Er (A) is a spectral sequence of ∗,∗ Er (R)-algebras. For examples such as R = kO, kU, tmf, tmf1(3) with p = 2, we know that H∗R is isomorphic ∗ to a left A∗ = H∗H-comodule algebra and then HRH is a subHopf algebra of the Steenrod ∗ ∗ algebra A = H H. In these cases A is a free HRH-module. Now recall that for any ring homomorphism A → B, left A-module U and left B-module V , there is a Cartan-Eilenberg change of rings spectral sequence of the form

p,q q A q−p E2 = ExtB(Torp (B,U),V )=⇒ ExtA (U, V ).

If B is A-ﬂat this collapses to give

∗ ∼ ∗ ExtA(U, V ) = ExtB(B ⊗A U, V ).

∗ When A = HRH and B = A we obtain ∗ ∗ ∗ ∗ Ext ∗ (F2, F2) =∼ ExtA(A⊗ ∗ F2, F2) =∼ ExtA(H R, F2) HRH HRH which is the E2-term of the Adams spectral sequence for computing π∗R; of course, this is a standard change of rings isomorphism.

2. Cell and CW R-modules

Now we assume that R is p-local for some prime p and also (−1)-connected with R0 = π0R Z a acyclic (p)-module. This means that is a local graded ring whose maximal ideal m ⊳ π∗R consists of the ideal (p) ⊳ R0 together with all positive degree elements. The notions of cell and CW R-modules have the usual forms described in [EKMM97]. The n-skeleton X[n] of such a cell or CW R-module X is obtained from the (n − 1)-skeleton X[n−1] by forming a coﬁbre sequence of form

n−1 n−1 j [n−1] [n] (2.1) SR −−−→ X → X . _i k k Here we take SR to be the functorial coﬁbrant replacement of Σ R (recall that R is not coﬁbrant in the model category MR of [EKMM97]). Associated with such a cell R-module X there is a CW cellular chain complex of R0-modules C∗ (X) satisfying

CW F ∼ R (2.2) H∗(C∗ (X) ⊗R0 p) = H∗ X. 5 It is often useful to work with a minimal cell structure; we adapt the notion of minimal cell structure from [BM04] to the present context. A CW R-module X is minimal if for every n, the attaching map jn−1 of (2.1) satisﬁes

n−1 n−1 [n−1] [n−1] im[j∗ : π∗( SR ) → π∗X ] ⊆ mπ∗X . _i

It is easy to see that every connective R-module with ﬁnite type homotopy can be realised by a ﬁnite type connective minimal CW R-module. Furthermore, by (2.2) the cells of this give an R Fp-basis for H∗ X, and the inclusion map induces a monomorphism

R [n−1] R [n] Hn−1X → Hn−1X .

In order to describe CW modules or their (co)homology, we will often use cell diagrams or ∗ diagrams showing bases with action of HRH. Usually we will assume a minimal cell structure has been used so cells will correspond to basis elements. For example, when R = S, the mapping 1 0 0 2 ∗ cone of the Hopf map η : S → S , X = S ∪η e , and its mod 2 cohomology, H X, can be represented by diagrams such as the following.

η Sq2

Such diagrams are standard in homotopy theory, for example they are discussed by Barratt, Jones & Mahowald [BJM84]. Occasionally we will require more complicated diagrams which involve maps between more general objects than just spheres. For example, given maps g : X → Y and h: Y → Z with hg null homotopic, we can factor Σg :ΣX → ΣY through the mapping cone of h and so deﬁne an object which is the mapping cone of a factor ΣX → Z ∪h CY represented by the following diagram.

Σ2X

Σg ΣY

h Z

Here is a useful result on building objects realising such diagrams; it generalises well known results for maps between spheres. Part (a) features prominently in [BJM84] while in [Bak18] we used (b) to construct realisations of Joker-like modules.

Lemma 2.1. Suppose that f : W → X, g : X → Y and h: Y → Z where gf and hg are null homotopic; therefore the Toda bracket hh,g,fiR ⊆ DR(ΣW, Z) is deﬁned. 6 (a) It is possible to realise the diagram Σ3W

Σ2f Σ2X

Σg ΣY

h Z if and only if the Toda bracket hh,g,fiR is contains zero. (b) If there are maps k :ΣW → U and ℓ: U → Z so that ℓk ∈ hh,g,fiR then we can realise the diagram Σ3W 2 ✇ ✹ Σ f ✇✇ ✹✹ ✇✇ ✹ ✇✇ ✹ ✇✇ ✹✹ 2 ✹ Σg Σ X ✹✹ ✹✹ ✹✹ ✹✹ Σg ΣU ✡ ✡✡ ✡✡ ✡✡ ΣY ✡✡ ❍❍ ✡ ℓ ❍❍ ✡✡ ❍❍ ✡ ❍❍ ✡ h ❍❍ ✡✡ Z

Here h−, −, −iR denotes the Toda bracket calculated in the homotopy category of R-modules.

Remark 2.2. There is a diﬀerent kind of Toda bracket that we might consider for an R ring spectrum A and a left A-module X spectrum (both being R-modules of course). For example we can consider hu, v, wiR,X for u ∈ πr(A), v ∈ πs(A) and w ∈ πt(X) where uv =0in π∗(A) and vw = 0 in π∗(X) where these products are deﬁned using the evident ring structure on π∗(A) and the π∗(A)-module structure of π∗(X). The resulting bracket is a subset of πr+s+t+1(X) with indeterminacy uπs+t+1(X) + πr+s+1(A)w. A version of this theory was described by G. Whitehead [Whi70] well before modern categories of spectra were developed but the essential ideas can be found in his work. Some applications of these brackets can be found in [BM04] using examples given in [Whi70]; in these we have R = A = S and X = kO or kU.

3. Duality

Various sorts of duality occur in the categories MR and DR, generalising classical cases. Spanier-Whitehead duality. For an account of duality from a categorical viewpoint, we recommend the article of Dold & Puppe [DP80].

Following [DP80, EKMM97], the symmetric monoidal category DR has strongly dualisable objects and so there is a version of Spanier-Whitehead duality; we will denote the Spanier- 0 Whitehead of an R-module X by DRX = FR(X,SR). As usual, when X is a ﬁnite CW module we can replace it by a weakly equivalent CW module; it is well known that an R-module is 7 strongly dualisable if it is equivalent to a retract of a ﬁnite CW module. Of course, if Z is a strongly dualisable S-module (i.e., a spectrum) then R ∧ Z is a strongly dualisable R-module; more generally, if Z is a strongly dualisable R′-module where R is a commutative R′-algebra, then R ∧R′ Z is strongly dualisable. When L is strongly dualisable, the Adams spectral sequence of (1.3) can be expressed as

s,t s,t ∗ t−s t−s (3.1) E = Ext ∗ (H (DRL ∧R M), Fp)=⇒ DR(Σ , DRL ∧R M) =∼ DR(S L, M) 2 HRH R R since there are natural isomorphisms of functors b b ∗ ∗ ∗ Hom ∗ (H (D L) ⊗ (−), F ) ∼ Hom ∗ (D(H L) ⊗ (−), F ) ∼ Hom ∗ (−,H L) HRH R R p = HRH R p = HRH R ∗ on left HRH-modules extending to right derived functors. This is useful for computational purposes as it allows us to work consistently with projective resolutions and calculations with right derived functors of form Hom ∗ (−, F ). HRH p Poincaréduality and Spanier-Whitehead duality. We begin with some algebra. Let k be a field and K∗ a connected graded cocommutative Hopf algebra of finite type. We will indicate the coproduct ψ using the notation ′ ′′ ′′ ′ ψθ = θ ⊗ 1 + 1 ⊗ θ + θi ⊗ θi = θ ⊗ 1 + 1 ⊗ θ + θi ⊗ θi Xi Xi ′ ′′ where the degrees of θ ,θi are positive and smaller than the degree of θ. The action of the antipode χ will often be indicated by writing χθ = θ, so ′ ′′ ′ ′′ θ = −θ − θiθi = −θ − θiθi . i i X X ∗ Now let P∗ be a local Poincaréduality algebra of degree d and let its graded dual be P n where P = Homk(Pn, k). This means that Pn = 0 except when 0 6 n 6 d, P0 = k and there ∼ = d is a k-linear isomorphism P0 −→ P with 1 ↔ λ which induces isomorphisms ∼ = d−n Pn −→ P ; x 7→ xλ where (xλ)(y)= λ(yx) for all y ∈ Pd−n. The pairing ∗ ∗ P∗ ⊗ P → P ; x ⊗ γ 7→ xγ ∗ makes P a left P∗-module and the above duality isomorphism can be interpreted as defining an isomorphism of left P∗-modules ∼ = ∗ P∗ −→ P [−d]; x 7→ xλ. ∗ Now suppose that P∗ is a left K -module algebra. This means that there are pairings ∼ r = K ⊗ Ps −→ Ps−r; θ ⊗ x 7→ θx so that the Cartan formula holds for all x,y ∈ P∗: ′ ′′ θ(xy) = (θx)y + x(θy)+ (θix)(θi y). Xi There is also an action of K∗ on P ∗ given by pairings ∼ Kr ⊗ P s −→= P r+s; θ ⊗ γ 7→ θγ 8 where (θγ)(z)= γ(θz) and this makes P ∗ a left H∗-module. ∼ = ∗ Lemma 3.1. The duality isomorphism of left P∗-modules P∗ −→ P [−d] is also an isomorphism of left H∗-modules.

∗ Proof. We have to show that for all homogeneous elements x,y ∈ P∗ and θ ∈ K , λ(y(θx)) = ((θx)λ)(y)= λ((θy)x). We will prove this by induction on the degree of θ. It is clearly true when θ has degree 0. So assume it holds whenever θ has degree less than n> 0. Suppose that θ has degree n. Consider θ(yx); in order for this element to have degree d, yx has to be of degree n + d> 0, hence yx = 0. So 0 = ((θ(yx))λ)(1) = λ(θ(yx)) ′ ′′ = λ((θy)x)+ λ(yθx)+ λ((θiy)(θi x)) Xi ′′ ′ = λ((θy)x)+ λ(yθx)+ λ((θi θiy)x) Xi ′ ′′ = λ(yθx) − λ((θy)x)+ λ (θy)x + λ((θiθi y)x + (θy)x Xi = λ(yθx) − λ((θy)x). Thus λ(yθx)= λ((θy)x), so the result holds for all n. Now let R be a commutative S-algebra satisfying the conditions assumed earlier. In partic- Z ular, suppose that R0 is a cyclic (p)-module for some prime p. R Suppose that E is an R ring spectrum for which P∗ = H∗ E is a local Poincar´eduality algebra F ∗ ∗ over p. Taking K = HRH, the Spanier-Whitehead dual of E satisﬁes R ∼ ∗ ∼ R H∗ DRE = HRE = H∗ E[d] ∗ as HRH-modules. The next result is very useful for identifying Spanier-Whitehead stably self dual objects.

Proposition 3.2. There is a morphism of R-modules E → ΣdR inducing a non-trivial homomorphism R R H∗ E → H∗ R[−d]= Fp[−d].

The multiplication map E ∧R E → E composed with this map deﬁne a duality pairing E ∧R E → d −d Σ R. Hence E is a Spanier-Whitehead stably self dual R-module with DRE ∼ Σ E.

Proof. Choose a minimal CW R-module realisation of E; this will have a single cell in each of F R the degrees 0 and d. Inclusion of the bottom cell induces the unit p → H0 E, while collapse R F onto the top cell induces a non-trivial linear mapping Hd E → p and this gives a basis element of Hd E; by composing with a self map of Sd ∼ ΣdR we can assume this corresponds to R R ∼ R R = d 1 ∈ H0 E under the Poincar´eduality isomorphism H0 E −→ HRE. The product E ∧R E → E d d composed with the projection E → Σ R gives rise to a morphism f : E → Σ DRE and induces ∼ R R R = R a non-degenerate pairing H∗ E ⊗ H∗ E → Fp[d]. It follows that f∗ : H∗ E −→ H∗ DRE[d] and 9 ∼ d so f : E −→ Σ DRE. Notice that all the algebraic maps here are compatible with the actions ∗ of HRH.

For any compact Lie group G, E = R ∧ (G+) provides an example of such a stably self dual R ring spectrum. A generalisation to ﬁnite H-spaces was proved by Browder & Spanier [BS62], and some exotic examples can be found in the papers of Bauer, Pedersen and Rognes [Bau04, BP06, Rog08]; in all these classical cases, R = S is the base spectrum but the ideas work more generally. More recently, Beaudry et al [BHL+21] have also considered Spanier-Whitehead self duality over commutative S-algebras.

Cyclic modules. It is common to encounter cyclic modules of the form A(n) ⊗B F2 arising as cohomology of R-modules for examples such as R = kO and R = tmf. Proposition 3.2 sometimes allows us to identify the underlying R-module as stably Spanier-Whitehead self dual, but this seems not to be a purely algebraic result. For example, consider the A(1)-module 2 2 F A(1) ⊗F2(Sq ) 2 = A(1)/A(1){Sq }. Viewing this as the quotient of A(1) obtained by killing the white circles in the diagram below, we see that this is the question mark module which is clearly not self dual.

Sq2

Sq1

The situation when B is a subHopf algebra is more interesting. We will encounter many examples of this kind later arising as cohomology of kO or tmf-modules. Next we give some algebraic results that we will use. We will assume the reader is aware of basic results such as the freeness of a Hopf algebra over a subHopf algebra (Milnor-Moore or Nichols-Zoeller for the ungraded case), and Poincaréduality/Frobenius property and self injectivity for finite dimensional Hopf algebras (Browder-Spanier or Larson-Sweedler for the ungraded case). First we give a statement and proof of a result for non-graded Hopf algebras which ought to be standard; the closest reference appears to be in Fischman et al [FMS97, theorem 4.8 and corollary 4.9]. This version incorporates suggestions of Ken Brown which led to a substantial improvement of our original attempt. Of course, if K is a normal subalgebra (i.e., if HK+ = K+H) then H//K is a Hopf algebra and this result is immediate. Proposition 3.3. Let H be a finite dimensional commutative or cocommutative Hopf algebra over a field k and let K be a subHopf algebra which is also unimodular. Then + H//K = H ⊗K k =∼ H/HK 10 is a self dual left H-module.

Proof. For general results on Hopf algebras the reader is referred to Larson & Sweedler [LS69], Humphreys [Hum78, theorem 1] and Montgomery [Mon93]; the book of Lorenz [Lor18] also contains much useful material. When H is commutative or cocommutative its antipode is self inverse. i.e., χ ◦ χ = Id. Let λ be a Frobenius form for H which we take to be a left and right integral in the dual Hopf algebra DH = Homk(H, k). The Nakayama algebra automorphism ν : H → H is characterised by the identity

(3.2) λ(xy)= λ(yν(x)) for all x,y ∈ H (see [Lam99, section §16E]). If the associated bilinear pairing H ⊗k H → k is symmetric (which is true when H is unimodular) then ν = IdH . As K is also unimodular, its (1-dimensional) vector spaces of left and right integrals coincide, l r i.e., ∫ K = ∫ K, and we will just write ∫ K for this subspace. By deﬁnition, the left and right annihilators of ∫ K and any non-zero element s ∈∫ K satisfy l l r r + annK ∫ K = annK (s) = annK ∫ K = annK (s)= K , the kernel of the counit K → k which is a maximal ideal. By the Nichols-Zoeller theorem [NZ89], H is free as a left or right K-module, so the left annihilators in H satisfy

l l + (3.3) annH ∫ K = annH (s)= HK , the left ideal of H generated by K+. Similarly the right annihilators satisfy

r r + + (3.4) annH ∫ K = annH (s)= K H = χ(HK ). k Now choose a non-zero element s0 ∈ K. The -linear mapping ′ λ : HR → k; x 7→ λ(xs0) satisﬁes λ′(xz) = 0 whenever x ∈ H and z ∈ K+, so it factors through a linear mapping λ′′ : H//K → k. It follows that the left H-module homomorphism

H → H; x 7→ xs0 ∼ + = has kernel HK and so induces an isomorphism H//K −→ Hs0. Deﬁne a left H-module structure on Homk(Hs0, k) by setting (x · α)(z)= α(χ(x)z) for α ∈ Homk(Hs0, k) and x ∈ H. Now deﬁne a left H-module homomorphism ′ H → Homk(Hs0, k); x 7→ x · λ . Using the Nakayama automorphism characterised in (3.2), we obtain

(x · λ′)(z)= λ(χ(x)z)= λ(zν(χ(x))).

So when x ∈ HK+ = χ(K+H) we have (x · λ′)(z) = 0, and therefore we can factor our homomorphism through a left H-module homomorphism H//K → Homk(Hs0, k). Also, when 11 x∈ / HK+, since the kernel of the functional λ((−)ν(χ(x))): H → k cannot contain a non- trivial left ideal, the functional x · λ: Hs0 → k must be non-trivial. This shows that we have an injection

H//K → Homk(Hs0, k) =∼ Homk(H//K, k) which must be an isomorphism since the dimensions of the domain and codomain agree.

Recall that a connected cohomologically graded k-algebra A has Ai = 0 when i < 0 and A0 =∼ k. A finite dimensional connected graded Hopf algebra over a field k is a Poincaré duality algebra of some dimension d, and a basis element of the 1-dimensional k-vector space Homk(Ad, k) provides a ‘Frobenius form’ with similar properties to the ungraded case. The proof of the following involves a modification of that for Proposition 3.3 with suitable allowances for gradings.

Proposition 3.4. Let H be a ﬁnite dimensional commutative or cocommutative unimodular connected graded Hopf algebra over a ﬁeld k and let K be a subHopf algebra. Then

+ H//K = H ⊗K k =∼ H/HK and there is an isomorphism of left H-modules ∼ = Homk(H//K, k) −→ H//K[−d], hence H//K is a stably self dual left H-module.

Remark 3.5. When char k = 2, the antipode χ acts on the (necessarily 1-dimensional) top degree part of H as the identity, hence the dual Hopf algebra DH is unimodular. Hence we can apply Proposition 3.4 to pairs of ﬁnite dimensional Hopf subalgebras of the Steenrod algebra A.

In the next result we give some useful consequences of Propositions 3.3 and 3.4 (in the latter case we need to interpret modules and morphisms as being suitably graded). First we need to set up some notation.

• We will set ⊗ = ⊗k and Hom = Homk. ◦ • For a right K-module N (i.e., a left K -module), make HomK(H,N) a left H-module with action given by (h · ϕ)(x)= ϕ(χ(h)x) ◦ for h, x ∈ H and ϕ ∈ HomK◦ (H,N) where H is regarded as a left K -module through ◦ right multiplication of K. Also make H//K ⊗ N and Homk(H//K, N) left K -modules by letting K◦ act through its action on N; this makes them both H ⊗ K◦-modules. • When L and M are left H-modules make Hom(L, M) a left H-module with action given by ′′ ′ (h · θ)(x)= hi θ(χ(hix)) Xi for h ∈ H, x ∈ L and θ ∈ Hom(L, M) and the coproduct on h being

′ ′′ ψ(h)= hi ⊗ hi . Xi It is well known that θ satisﬁes h · θ = ε(h)θ for all h ∈ H if and only if θ ∈

HomH (L, M) ⊆ Hom(L, M). 12 • Viewing H and M as left K-modules, make HomK(H, M) a left H-module by setting

(h · ρ)(x)= ρ(xh).

Proposition 3.6. Let H and K be as in Proposition 3.3 or 3.4. Let L and M be left H-modules and let N be a right K-module. (a) There are natural isomorphisms of left H ⊗ K◦-modules ∼ ∼ H//K ⊗ N −→= Hom(H//K, k) ⊗ N −→= Hom(H//K, N).

(b) There are natural isomorphisms of left H-modules ∼ ∼ = = H//K ⊙ M −→ Hom(H//K, M) −→ HomK(H, M).

Proof. (a) The ﬁrst isomorphism of vector spaces uses Proposition 3.3, the second is standard; these clearly respect the H ⊗ K◦-module structures. (b) As for (a), there are isomorphisms of k-vector spaces ∼ ∼ H//K ⊗ M −→= Hom(H//K, k) ⊗ M −→= Hom(H//K, M) which are both H-linear. Deﬁne a map

Hom(H//K, M) → HomK(H, M); ϕ 7→ ϕ where e ′ ′′ + ϕ(x)= xiϕ(χ(xi )+ HK ). Xi This has inverse e ≈ HomK(H, M) → Hom(H//K, M); θ 7→ θ where ≈ + ′′ ′ θ(x + HK )= χ(xi )θ(xi). Xi These are both H-linear. (c) Using (b) and a standard adjunction result we obtain ∼ = HomH (L, H//K ⊙ M) −→ HomH (L, HomK(H, M)) ∼ = −→ HomK(H ⊗H L, M) ∼ = −→ HomK(L, M).

These identiﬁcations can be used to deduce homological results. Here is one that is useful in our work.

Proposition 3.7. Let H, K, L and M be as in Proposition 3.6. Then ∗ ∼ ∗ ExtH (L, H//K ⊙ M) = ExtK (L, M). 13 Proof. Take a projective resolution P∗ → L of the H-module L. Then for each s > 0,

Ps =∼ H ⊗H Ps is both a projective H-module and a projective K-module since H is a free left K-module. Therefore P∗ → L is also a projective resolution for L as a K-module. Also

HomH (P∗, HomK(H, M)) =∼ HomK (H ⊗H P∗, M)

=∼ HomK (P∗, M), so on taking cohomology we obtain

∗ ∼ ∗ ExtH (L, HomK(H, M)) = ExtK (L, M).

The result now follows using Proposition 3.6(b).

In particular, the case where L = H//K and M = k is useful for some of the topological examples we will see later. In the graded case, suppose that the top degrees of H and K are d and e respectively. Then the top degree of H//K is d − e and

D(H//K) =∼ H//K[e − d].

Poincaréduality for manifolds. For a commutative ring spectrum E, classical Poincaré duality in E-theory is defined using the slant product for a space X+ with disjoint base point

∗ E (X+) ⊗E∗ E∗(X+) → E∗(X+)

0 and the augmentation induced by collapse on the base point E∗(X+) → E∗(S )= E∗. Under- lying this are compositions of the form

∧ ∧ n Id ∆ n r mult Id n+r n+r Σ E ∧ X+ / E ∧ X+ ∧ X+ / Σ E ∧ Σ E ∧ X+ / Σ E ∧ X+ / Σ E and when E is a commutative S-algebra, the composition is a morphism of left E-modules. Of course this can also be formulated in terms of Spanier-Whitehead duality using Atiyah duality. This leads to the following result which gives a rich source of stably self dual R-modules; similar ideas were explored by Douglas, Henriques and Hill [DHH11].

Proposition 3.8. Let R be a commutative S-algebra. Suppose that M is a compact closed smooth n-manifold whose normal bundle admits an orientation in R-cohomology. Then R ∧M+ is a stably Spanier-Whitehead self R-module and

−n DR(R ∧ M+) ∼ R ∧ Σ (M+).

Proof. On choosing an R-orientation, Poincar´eduality for R∗(M+) gives an explicit isomorphism ∼ = n−∗ R∗(M+) −→ R (M+).

By the above remarks, this is induced by a weak equivalence of R-modules

−n DR(R ∧ M+) ∼ R ∧ Σ (M+). 14 If G is a compact Lie group, whenever A 6 G is a closed abelian subgroup (such as a maximal torus), G/A is stably frameable and so stably self dual, hence R ∧ (G/A+) is a stably self dual R-module. Since Spin manifolds are kO-orientable they satisfy the conditions; similarly when R = tmf, String manifolds are tmf-orientable. As examples of these, recall that RPn is a Spin manifold R 4k−1 if and only if n ≡ 3 mod 4, and it is a String manifold if n ≡ 7mod8. So kO ∧ ( P+ ) is a stably self dual kO-module with

R 4k−1 −4k+1 R 4k−1 DkO(kO ∧ ( P+ )) ∼ kO ∧ Σ ( P+ ) R 8ℓ−1 and tmf ∧ ( P+ ) is a stably self dual tmf-module with R 8ℓ−1 −8ℓ+1 R 8ℓ−1 Dtmf (tmf ∧ ( P+ )) ∼ tmf ∧ Σ ( P+ ).

We will discuss homogeneous spaces in more greater detail in Section 6.

4. Recollections on the Steenrod algebra and finite subHopf algebras

The book of Margolis [Mar83] provides a thorough treatment of the Steenrod algebra A and its ﬁnite subHopf algebras such as the A(n) family. For more on bases of A and its subalgebras see the survey article of Wood [Woo98]. In this appendix we highlight some important ideas that are useful for the present work.

The Wall relations for A(n). When working with A or A(n) we often describe elements by dualising the monomial basis in the dual. Recall that

A∗ = F2[ξi : i > 1], F 2n+1 2n 2n−1 2 A(n)∗ = 2[ξi : i > 1]/(ξ1 ,ξ2 ,ξ3 ,...,ξn+1,ξn+2,...).

r1 r2 rℓ Then the basis of (residue classes of) monomials ξ1 ξ2 · · · ξℓ deﬁnes a basis consisting of the elements Sq(r1, . . . , rℓ) in A or A(n). We remark that in the computer algebra system Sage, only this basis is available for working in A(n) making it the natural basis for expressing results found with its aid. Of course these basis elements can also be expressed in terms of monomials s in Sqr or indeed the indecomposables Sq2 . The usual Adem relations in A do not always restrict to a ﬁnite subalgebra A(n). For example, the following consequences of Adem relations

Sq2 Sq3 = Sq4 Sq1 +Sq5 = Sq4 Sq1 +Sq1 Sq4 are not meaningful in A(1) since Sq4 ∈A/ (1). s A minimal set of relations between indecomposable generators Sq2 of A was determined by Wall [Wal60] and these do restrict to deﬁning relations for each A(n). Incidentally, these relations can be interpreted in either the sense of algebra relations or module relations for the augmentation ideal considered as a left or right module. Consider the following elements of A∗: for 0 6 s 6 r − 2 and 1 6 t,

r s s r (A) Θ(r, s)=Sq2 Sq2 +Sq2 Sq2 ,

t t t−1 t t−1 t−1 t−1 t (B) Φ(t)=Sq2 Sq2 +Sq2 Sq2 Sq2 +Sq2 Sq2 Sq2 .

15 Then Θ(r, s) ∈ A(r − 1) and Φ(r) ∈ A(r − 1)∗, so these can be expressed as polynomial k expressions in the Sq2 for 0 6 k 6 r − 1. The elements of form

r s s r t t t−1 t t−1 t−1 t−1 t Sq2 Sq2 +Sq2 Sq2 + Θ(r, s), Sq2 Sq2 +Sq2 Sq2 Sq2 +Sq2 Sq2 Sq2 + Φ(t) give a minimal set of relations for A. In particular, such elements with r, t 6 n form a minimal set of relations for A(n). In the ﬁrst few cases the Wall relations are A(0) : Sq1 Sq1 = 0, A(1) : Sq1 Sq1 = Sq2 Sq2 +Sq1 Sq2 Sq1 = 0 A(2) : Sq1 Sq1 = Sq2 Sq2 +Sq1 Sq2 Sq1 = Sq4 Sq4 +Sq2 Sq4 Sq2 +Sq2 Sq2 Sq4 = Sq1 Sq4 +Sq4 Sq1 +Sq2 Sq1 Sq2 = 0.

s We will sometimes use the Milnor primitives P1 (s > 0) deﬁned recursively by 0 s 2s s−1 s−1 2s P1 = Sq(1), P1 = Sq P1 +P1 Sq (s > 1).

More generally, Margolis [Mar83] uses the notation Sq(r1, . . . , rℓ) for the element dual to the r1 rℓ s s s monomial ξ1 · · · ξℓ and sets Pt = Sq(0,..., 0, 2 ) where 2 occurs in the t-th place. Doubling. Doubling is discussed by Margolis [Mar83, section 15.3] and also by the present author [Bak18, section 2]. The main thing to observe is that for any n > 1 there is a grade halving homomorphism of Hopf algebras ∼ A(n) −→A= (n − 1) which induces a grade halving isomorphism ∼ A(n)//E(n) −→A= (n − 1) under which the residue class of Sqk maps to Sqk/2 if k is even and 0 otherwise. If M is a left A(n − 1)-module then it becomes A(n)//E(n)-module (1)M with all gradings doubled, hence it is also a left A(n)-module. For example, when n = 1 the A(1)-module (1)A(0) =∼ A(1)//E(1) is realised by the cohomology of the kO-module kU shown in (5.12), and when n = 2, the A(2)-module (1)A(1) =∼ A(2)//E(2) (the double of A(1)) is realised by the cohomology of the tmf-module tmf1(3) ∼ BP h2i.

5. kO-modules and A(1)-modules

In this section we review some well known results on the (co)homology of kO-modules localised at the prime 2. Such results were originally due to Adams & Priddy, Mahowald and Milgram, see [AP76,Mah81,MM76]; the book of Margolis [Mar83] also provides a thorough treatment of stable module categories for ﬁnite subHopf algebras of A. We begin by summarising some relationships between relative Steenrod algebras and their duals that involve connective K-theory; these are obtained using the ideas discussed in Section 1. These will be used heavily in what follows.

Theorem 5.1. For H = HF2 we have the following identiﬁcations of Hopf algebras: ∗ ∼ ∗ kO ∼ HkOH = A(1) = A(1),H∗ H = A(1)∗, ∗ ∼ ∗ kU ∼ HkUH = E(1) = E(1),H∗ H = E(1)∗. 16 Of course this means that the homology and cohomology of a kO-module are naturally A(1)- modules and those of a kU-module are E(1)-modules.

The kO-module H = HF2 can be given a minimal cell structure with 8 cells corresonding to the obvious basis of A(1)-module A(1).

3 3

Sq2 1 Sq1 0

−6 Since H is a kO ring spectrum, by Proposition 3.2 it is stably self dual and DkOH ∼ Σ H.

When working in the stable module category StModA(1) we will often blur the distinction between a module and its stable equivalence class. We denote the kernel of the counit ε: A(1) → −1 F2 by I(1); this module is stably invertible with stable inverse I(1) = DI(1), represented by the linear dual of I(1).

6 -1 I(1) I(1)−1 5 Sq2 -2

4 -3 -3

3 3 -4 Sq1 2 -5

(5.1) 1 -6

−1 Recall that there are two mutually inverse endofunctors Ω, Ω of StModA(1), where

ΩM = I(1) ⊙ M, Ω−1M = I(1)−1 ⊙ M.

These commute with products and preserve duals, i.e.,

(ΩM) ⊙ N = Ω(M ⊙ N), (Ω−1M) ⊙ N =Ω−1(M ⊙ N), ΩDM = DΩM, Ω−1DM = DΩ−1M.

17 Amongst the many cyclic quotient A(1)-modules is the Joker module

J = A(1)/A(1){Sq3}[−2] Sq2 0

Sq1

(5.2) −2 whose grading is arranged so that it is a self dual A(1)-module, indeed it represents the unique element of order 2 in the Picard group of StModA(1), PicA(1). In StModA(1) we have the following presentations as quotients of free modules: ΩJ = A(1)/A(1){Sq1, Sq2 Sq1 Sq2}[1] = A(1)/A(1){Sq1, Sq2 Sq3}[1], Ω−1J = A(1)/A(1){Sq2}[−4], Ω2J = (A(1)/A(1){Sq1}[2] ⊕A(1)/A(1){Sq1, Sq2 Sq1 Sq2}[6])/A(1){(Sq2 Sq1 Sq2, Sq1)}, Ω−2J = A(1)/A(1){Sq2 Sq1}[−7]. These modules have the following diagrams (the ﬁrst two are often called the question mark and upside down question mark modules). ΩJ Ω−1J Ω2J Ω−2J

4 -1 7 -2

3 6

-3 5 -4

1 -4 4 -5

-6

(5.3) 2 -7 F F We recall the structure of ExtA(1)( 2, 2), the Adams E2-term for π∗kO, part of which appears in Figure 1. We also have s,∗ F ∼ s+1,∗ F F (5.4) ExtA(1)(I(1), 2) = ExtA(1) ( 2, 2), whose chart is a shifted version of Figure 1 with the • at (0, 0) removed. For the dual DI(1), F ∗ 2 if s = 0, (5.5) Exts, (DI(1), F ) =∼ A(1) 2 s−1,∗ F F (ExtA(1) ( 2, 2) if s> 0. Adams & Priddy [AP76, tables 3.10 & 3.11] give the Ext groups of the two question mark modules ΩJ and Ω−1J . 18 ∗ Such A(1)-modules can often be realised as cohomology of kO-modules, i.e., as HkOX for some kO-module X. Indeed, in some cases X = kO ∧ Z for an S-module Z. For example, there 0 2 3 0 1 3 are S-modules S ∪η e ∪2 e and S ∪2 e ∪η e so that as A(1)-modules,

∗ 0 2 3 ∼ ∗ 0 2 3 ∼ HkO(kO ∧ S ∪η e ∪2 e ) = H (S ∪η e ∪2 e ) = ΩJ [−1], ∗ 0 1 3 ∼ ∗ 0 1 3 ∼ −1 HkO(kO ∧ S ∪2 e ∪η e ) = H (S ∪2 e ∪η e ) = Ω J [4].

In [Bak18], the Joker was shown to be realisable as H∗J for an S-module J, so as A(1)-modules, ∗ ∼ ∗ HkO(kO ∧ J) = H J. The realisability of the self dual cyclic A(1)-module A(1)//A(0) provides a more interesting question.

A(1)//A(0) = A(1)/A(1){Sq1} Sq1

Sq2

(5.6) 0

0 2 3 5 In MS, no CW complex of the form S ∪η e ∪2 e ∪η e can exist since the Toda bracket 0 hη, 2, ηi = {2ν, −2ν} ⊆ π3S does not contain 0; alternatively, its existence is contradicted by the relation Sq1 Sq2 Sq1 = Sq4 Sq1 +Sq1 Sq4 in the Steenrod algebra A. However, taken in π∗kO, the Toda bracket hη, 2, ηi ⊆ π3kO contains the images of ±2ν which are both zero, hence there is a CW kO-module with this cohomology. Of course HZ is a kO-module whose cohomology agrees with this A(1)-module, so we are describing a minimal CW structure for it. kO Since HZ is a kO ring spectrum and H∗ HZ satisfies Poincaréduality, Proposition 3.2 applies. −5 Hence the kO-module HZ is stably self dual with DkOHZ ∼ Σ HZ. The kO-modules discussed above fit into the Whitehead tower of kO shown in (5.7) as CW kO-modules, where the numbers indicate degrees of cells in a minimal CW structure; we learnt about the following from Bob Bruner and John Rognes, see for example [Bru12, Bru14, BG10]. ∗ On applying HkO(−) we obtain the A(1)-module extension of (5.8) and (5.9) which represents 4,12 F F an element of ExtA(1)( 2, 2), and in turn this represents the Bott periodicity element in π8kO in the Adams spectral sequence. This periodicity is also visible in Figure 1 which shows part of 19 ∗,∗ F F 8,4 F F ExtA(1)( 2, 2) where it is given by Yoneda product with the basis element of ExtA(1)( 2, 2).

0 (5.7)

Sq1 2

Sq2

0 (5.8)

1 1 (5.9) 0 ← F2 ←A(1)/A(1){Sq }←A(1)[2] ←A(1)[4] ←A(1)/A(1){Sq }[7] ← F2[12] ← 0 20 12

s ↑

0 0 4 8 12 16t − s → 20

s,t F F 6 6 6 6 Figure 1. ExtA(1)( 2, 2): for 0 t − s 20, 0 s 12. Removing the F generator at (0, 0) and shifting down and left gives ExtA(1)(I(1), 2).

The coﬁbres of the maps in (5.7) have the following cell structures

2 η

η 2 (5.10) and their cohomologies are the following cyclic A(1)-modules. (5.11)

A(1)/A(1){Sq2} A(1)/A(1){Sq1 Sq2} A(1)/A(1){Sq1, Sq2 Sq1 Sq2}

Sq2 Sq1

In fact the whole sequence (5.7) can be made to be Spanier-Whitehead self dual in the sense 12 that applying Σ DkO gives an equivalent sequence. 21 Realisation of cyclic A(1)-modules. There are of course other cyclic quotients of A(1) which can be realised as cohomology of kO-modules. These fall into two groups, those containing a submodule isomorphic to the quotient A(1)//A(0) and those which do not. See Example 5.3 for the ﬁrst one, the others involving A(1)//A(0) are easily realised. The others can all be constructed as kO-modules of the form kO ∧ Z.

3 3 A(1)/A(1){Sq1 Sq2 Sq1 Sq2} A(1)/A(1){Sq1 Sq2 Sq1} Sq2

Sq1

3 A(1)/A(1){Sq2 Sq1}

A(1)/A(1){Sq2 Sq1 Sq2}

3 3

A(1)/A(1){Sq1 Sq2 Sq1, Sq2 Sq1 Sq2} A(1)/A(1){Sq1 Sq2, Sq1 Sq2 Sq1}

22 3

A(1)/A(1){Sq2 Sq1, Sq2 Sq1 Sq2} A(1)/A(1){Sq1 Sq2, Sq2 Sq1}

F 1 1 A(1)// 2(P1)= A(1)/A(1){P1}

1 1 2 2 1 1 A(1)//E(1) = A(1)/A(1){Sq , P1} Sq A(1)/A(1){Sq , P1} Sq

Each of the last three has the form A(1) ⊗B F2 for some subalgebra B⊆A(1). For each of the ﬁrst two, B is a subHopf algebra which explains the self duality. In the last case, F 2 1 F 2 1 2 2 1 B = 2(Sq , P1)= 2(Sq , Sq Sq +Sq Sq ) is a 6-dimensional commutative subalgebra, but even so A(1) ⊗B F2 is a stably self dual A(1)- module. Here A(1) is not a free right B-module since for example

1 1 2 2 Sq P1 = 1(Sq Sq ). F 1 1 The self dual cyclic module A(1)// 2(P1)= A(1)/A(1){P1} is a quotient Hopf algebra since 1 P1 is central in A(1). This module can be realised, see Example 5.6 for details. Constructions of some more kO-modules. Collapsing HZ onto its top cell gives a map Z 5 5 Z H → SkO and composing with a map of degree 2 to the bottom cell of Σ H gives a kO-module map HZ → Σ5HZ whose mapping cone has cohomology as shown.

We can repeat this using a suitable map from Σ−4HZ to this by using a degree 2 map to the bottom cell and so on. Repeatedly using such constructions and their Spanier-Whitehead duals leads to a ‘daisy chain’ of copies of suspensions of A(1)//A(0) glued together by actions of Sq1. 23 Such an A(1)-module is realised as the cohomology of a kO-module. This process can either be stopped after ﬁnitely many iterations or continued indeﬁnitely in either positive or negative directions. Of course these A(1)-modules are well known. In fact HZ can itself be realised in a similar way. The kO-module kU has cohomology

∗ 1 1 2 2 HkOkU= A(1)//E(1) = A(1)/A(1){Sq , Sq Sq } Sq (5.12)

−2 so DkOkU ∼ Σ kU. A similiar construction to the above yields a complex whose cohomology has the following form.

By iterating we can obtain familiar A(1)-modules such as

which can also be extended both upwards and downwards.

Some sample calculations and examples.

Example 5.2. Consider the Adams spectral sequence

s,t s,t ∗ Z ∗ D Z s−t E2 = ExtA(1)(HkOH ,HkOH)=⇒ kO(H,H ) .

Since ∗ ∼ ∼ HkOH = A(1) = D(A(1))[6] we have

∗,∗ ∼ ∗,∗ ∗ Z E2 = ExtA(1)(HkOH , D(A(1))[6]) ∼ ∗,∗ ∗ Z F = ExtA(1)(A(1) ⊗ HkOH , 2[6]) ∼ ∗ ∗ Z F = Hom (HkOH , 2[6]) ∼ 6 ∗ Z F ∼ ∗ Z F ∼ F = Hom (HkOH , 2[6]) = Hom(HkOH , 2) = 2

0 Z Z ∗ ∼ where the generator is the dual map to the non-zero element in HkOH . SoDkO(H,H ) = F2[0] and the generator corresponds to the map H → HZ which collapses H onto its top cell composed with inclusion of the bottom cell of Σ6HZ. By Spanier-Whitehead duality,

∗ ∼ −6 −5 ∗ ∼ ∗ ∼ DkO(HZ,H) = DkO(Σ H, Σ HZ) = DkO(H, ΣHZ) = F2[5]. 24 This time the generator involves collapse of HZ onto its top cell composed with inclusion of the bottom cell of Σ5H, i.e., a composition Z 5 5 H → SkO → Σ H. The reader may like to compare and contrast this calculation with that using the classical Adams spectral sequence s,t s,t ∗ Z ∗ D Z E2 = ExtA (H H ,H H)=⇒ S(H,H ). ∗ Example 5.3. Consider DkO(H, kO) . By Spanier-Whitehead duality this is isomorphic to −6 ∗ ∼ ∗−6 DkO(kO, Σ H) = DkO(kO,H) . The Adams spectral sequence s,t s,t ∗ F D s−t E2 = ExtA(1)(HkOH, 2)=⇒ kO(H, kO) has ∗,∗ ∗ F ∼ ∗,∗ F F F F ExtA(1)(HkOH, 2) = ExtA(1)(A(1), 2) = Hom( 2, 2)= 2 whose generator detects the inclusion of the bottom cell of H whose Spanier-Whitehead dual 6 6 is collapse onto the top cell of H, i.e., a map H → SkO ∼ Σ kO. The ﬁbre of this realises the cyclic module DI(1)[5] = A(1)/A(1){Sq1 Sq2 Sq1 Sq2}, a suspension of the dual of the counit, see (5.1).

Example 5.4. Recall the Joker module J of (5.2). In [Bak18] we showed that there is a spectrum J for which the kO-module kO ∧ J has cohomology is ∗ ∼ ∗ ∼ HkO(kO ∧ J) = H (J) = J as an A(1)-module. Actually there are two inequivalent such spectra which are Spanier- Whitehead dual and their cohomology realises the A-modules with the two possible Sq4 actions. For our present purposes we may choose J ′ and J ′′ to be either of them. ′ ′′ First we will determine DkO(kO ∧ J , kO ∧ J ). Using duality we have ′ ′′ ∗ ∼ ′ ′′ ∗ DkO(kO ∧ J , kO ∧ J ) = DkO(kO, (kO ∧ DJ ) ∧kO (kO ∧ J )) ∼ ′ ′′ ∗ = DkO(kO, kO ∧ DJ ∧ J ) . Now J is stably self dual and in fact as A(1)-modules, ∼ (5.13) J ⊗J = F2[0] ⊕A(1)[−4] ⊕A(1)[−3] ⊕A(1)[−2]. There is an Adams spectral sequence s,t ∗ ′ ′′ F D ′ ′′ ∗ E2 = ExtA(1)(HkO(kO ∧ DJ ∧ J ), 2)=⇒ kO(kO, kO ∧ DJ ∧ J ) and its E2-term is given by ∗,∗ ∼ ∗,∗ F F ∗ F F ∗ F F ∗ F F E2 = ExtA(1)( 2, 2) ⊕ Hom ( 2[−4], 2) ⊕ Hom ( 2[−3], 2) ⊕ Hom ( 2[−2], 2) ∼ ∗,∗ F F F F F = ExtA(1)( 2, 2) ⊕ 2[4] ⊕ 2[3] ⊕ 2[2]. ′ ′′ The reader is invited to describe the stable maps kO ∧ J → kO ∧ J corresponding to F2- 0,0 F F summands in cellular terms. The generator of ExtA(1)( 2, 2) is an infinite cycle in the spectral sequence showing there is indeed a weak equivalence of kO-modules kO ∧ J ′ → kO ∧ J ′′, hence 25 the choices of spectra J ′ and J ′′ do not affect the kO-module up to homotopy equivalence and from now on we just write kO ∧ J for any such kO-module. 1,2 F F ∼ F Notice that ExtA(1)( 2, 2) = 2, and this corresponds to the map kO∧ΣJ → kO∧J induced by multiplication by η. This shows that the cofibre sequence

2 kO ∧ J → kO ∧ Cη ∧ J → kO ∧ Σ J cannot split since the short exact sequence

∗ ∗ ∗ 2 0 o HkO(kO ∧ J) o HkO(kO ∧ Cη ∧ J) o HkO(kO ∧ Σ J) o 0 O O O ∼ ∼ ∼ = = = ∗ ∗ ∗ 2 H (J) H (Cη ∧ J) H (Σ J) 1,2 ∼ 1,2 F represents the corresponding element of ExtA(1)(J , J ) = ExtA(1)(J ⊗J , 2). Of course this also implies the well-known fact that the coﬁbre sequence of spectra

2 J → Cη ∧ J → Σ J does not split.

Example 5.5. Consider the kO-module HZ ∧kO HZ. The cohomology of this is ∗ Z Z ∼ ∗ Z ∗ Z HkO(H ∧kO H ) = HkOH ⊙ HkOH where the factors are given in (5.6). Since HZ is a unital kO-algebra, HZ is a retract.

Notice that HZ and hence HZ ∧kO HZ are Spanier-Whitehead stably self dual with −5 −10 DkOHZ ∼ Σ HZ, DkO(HZ ∧kO HZ) ∼ Σ HZ ∧kO HZ. ∗ Z ∗ Z ∗ Z ∗ Z Moreover, HkOH ⊙ HkOH must have A(1)-module summands HkOH and HkOH [5]; in fact a routine calculation shows that ∗ Z ∗ Z ∼ ∗ Z ∗ Z HkOH ⊙ HkOH = HkOH ⊕A(1)[2] ⊕ HkOH [5]. Now it is straightforward to show that

2 5 HZ ∧kO HZ ∼ HZ ∨ Σ H ∨ Σ HZ. ZkO Z Z Z Z In particular it follows that H ∗ H is not a free (2)-module; it is known that H ∗H Z ZkO Z also has simple 2-torsion, see [Koc82]. Of course the (2)-algebra structures of H ∗ H and Z∗ Z H kOH are both trivial for degree reasons.

In the planned sequel we will consider the tmf-algebras HZ ∧tmf HZ and kO ∧tmf kO.

Example 5.6. There is a CW complex Z such that ∗ ∼ ∗ ∼ 1 HkO(kO ∧ Z) = H Z = A(1)/A(1){P1}, here is a bare handed cellular construction illustrating some of the tools available. ′ ′ Consider the complex Z for which π2(Z ) =∼ Z ⊕ Z/2.

Z′ η 2

26 Then Z is obtained by attaching a 3-cell to Z′ using the sum of the generators of the group ′ ∼ π2(Z ) = Z ⊕ Z/2 (see Figure 2). There is a map Z → H which extends to a kO-module morphism kO ∧ Z → H inducing an epimorphism of A(1)-modules making the following diagram commute. ∼ ∗ ∗ = ∗ HkOH / / HkO(kO ∧ Z) o / H Z O O ∼ ∼ = = quo 1 A(1) / / A(1)/A(1){P1}

1 The kernel of the quotient homomorphism is isomorphic to A(1)/A(1){P1}[3] and there is a coﬁbre sequence of kO-modules

kO ∧ Z → H → kO ∧ Σ3Z.

We can also splice together inﬁnitely many copies of the short exact sequence

∗ 1 0 o H Z o A(1) o A(1)/A(1){P1}[3] o 0 to obtain ∗ 0 o H Z o A(1) o A(1)/A(1)[3] o A(1)/A(1)[6] o · · · which is a periodic resolution of H∗Z by cyclic free A(1)-modules. This can be used to determine the Adams E2-term for computing π∗(kO ∧ Z), see Figure 3. The homotopy groups of kO ∧ Z are given by

π∗(kO ∧ Z)= F2[u2] 12 where u2 ∈ π2(kO ∧ Z) is represented in the Adams spectral sequence by the element of 1,3 ∗ F ExtA(1)(H Z, 2) corresponding to the algebraic extension ∗ 0 → F2[3] →A(1) → H Z → 0

1 with non-trivial P1 action.

0 0 4 8 12 16 20

s,t ∗ ′ Figure 2. ExtA (H Z , F2) for 0 6 s 6 8, 0 6 t − s 6 12

27 8

0 0 4 8 12 16

s,t ∗ F 6 6 6 6 Figure 3. ExtA(1)(H Z, 2) for 0 s 12, 0 t − s 20

F 1 Given two realisations of A(1)// 2(P1) by kO-modules W1,W2, consider the Adams spectral sequence s,t s,t F 1 F 1 D E2 = ExtA(1)(A(1)// 2(P1), A(1)// 2(P1)) =⇒ kO(W1,W2). Then by Proposition 3.7,

s,t ∼ s,t F 1 F E = ExtF 1 (A(1)// 2(P1), 2[3]). 2 2(P1) F 1 F 1 1 The 2(P1)-module structure of A(1)// 2(P1) has a non-trivial P1-action linking the generators in degrees 0 and 3. 3

2 F 1 1 A(1)// 2(P1) P1 1

Therefore F 1 ∼ F 1 F F A(1)// 2(P1) = 2(P1) ⊕ 2[1] ⊕ 2[2] and since ∗,∗ F F F ExtF 1 ( 2, 2)= 2[w] 2(P1) with w in bidegree (1, 3), we have ∗,∗ ∼ F F F E2 = 2[−3] ⊕ 2[w][−2] ⊕ 2[w][−1].

There can be no non-trivial Adams diﬀerentials, in particular, the generator of F2[−3] which corresponds to the identity homomorphism can be realised by a weak equivalence W1 → W2 of kO-modules. This shows that this kO-module is well deﬁned up to weak equivalence and also stably self dual. 28 F 1 The A(1)-module obtained by inducing up the 2(P1)-module above has the form 1 1 1 A(1) ⊗F 1 A(1)//F (P ) =∼ A(1) ⊕A(1)//F (P )[1] ⊕A(1)//F (P )[2], 2(P1) 2 1 2 1 2 1 and this is isomorphic to H∗(Z ∧ Z).

Remark: The reader may have spotted that the smash product C2 ∧ Cη is also a model for Z which is stably self dual since C2 and Cη are. However, we feel it worthwhile building it explicitly to demonstrate the techniques available. Of course we know that the actual construction of Z is irrelevant since it is well deﬁned up to weak equivalence.

6. Examples based on homogeneous spaces

In this section we consider some examples built using homogeneous spaces, our goal is to identify kO-orientable manifolds so we discuss how to determine when they admit Spin structures. The reader will ﬁnd many interesting examples in the paper of Douglas, Henriques and Hill [DHH11], and we will give some others. We refer the reader to the work of Singhof & Wemmer [Sin82,SW86,SW87] for detailed results on stable frameability of homogeneous spaces. We were inspired to consider the question of orientability for homogeneous spaces by work of Atiyah & Smith [AS74, lemma 3.1] which gave a ‘local’ consequence of the existence of tangential Spin structures for homogeneous spaces G/T where T 6 G is a maximal torus; this is based on a condition on the sum of the positive roots.

Proposition 6.1. Let G be a compact connected Lie group and H 6 G a closed subgroup.

Then the tangent bundle of G/H is TG/H = G ×H g/h where H acts on g/h by the adjoint representation.

Proof. This is well-known, for example see Brockett & Sussman [BS72] or Singhof & Wemmer [Sin82, SW86].

The vector bundle G ×H g → G/H admits a trivialisation ∼ = (6.1) G ×H g / G/H × g ■■ ♣ ■■ ♣♣ ■■ ♣♣ ■■ ♣♣♣ ■$ ♣w ♣♣ G/H

[g,x]H o / (gH, Adj(g)(x)) which is G-equivariant where G/H × g is given the diagonal left G-action. By choosing an H-invariant inner product we can ﬁnd a splitting of the adjoint representation of H on g, g =∼ g/h ⊕ h, and this induces a G-vector bundle isomorphism ∼ AdjG/H g = AdjG/H g/h ⊕ AdjG/H h, where

AdjG/H (−)= G ×H (−) → G/H denotes the associated vector bundle construction. Now combining this with Proposition 6.1 and (6.1), we obtain some useful results. 29 Proposition 6.2. Let G be a compact connected Lie group and H 6 G a closed subgroup. Then there is an isomorphism of G-vector bundles

∼ TG/H ⊕ AdjG/H h = G/H × g.

Working non-equivariantly, the right hand bundle is trivial, so TG/H and AdjG/H h are stably inverse bundles.

Corollary 6.3. For a commutative ring spectrum E, G/H is E-orientable if and only if

AdjG/H h is E-orientable.

Thus the E-orientability of G/H can be investigated either by considering the adjoint action of H on g/h in terms of the action of H on roots of G outside of h, or the adjoint representation of H h. In particular, for kO-orientability this reduces to checking that for a chosen maximal torus, 1 (sum of positive roots of H) 2 is a weight (this is always true when H is simply connected). Before giving some examples, we recall that for a compact simply connected Lie group G with chosen maximal torus T , and ﬁnite centre Z(G), there are some important relationships between the root lattice Λrt(G), the weight lattice Λwt(G), the centre and the fundamental group π1(G/Z(G)):

(6.2) Λwt(G)/Λrt(G) =∼ Z(G) =∼ π1(G/Z(G)); furthermore,

(6.3) Λwt(G/Z(G)) = Λrt(G/Z(G)).

So for example when n > 3, this gives

Λwt(Spin(n))/Λrt(Spin(n)) =∼ Z/2, Λwt(SO(n)) = Λrt(SO(n)).

It is also known that in general,

(6.4) |π1(G/Z(G))| = |Z(G)| = |Λwt(G) : Λrt(G)| = determinant of the Cartan matrix of G.

Finally we recall that if T 6 H 6 G where T is a maximal torus of G then the cohomology of G/H is concentrated in even degrees so all odd degree elements of the Steenrod algebra act trivially. Now we discuss some examples where we can determine whether Spin structures exist on the adjoint representations of some standard Lie groups.

Example 6.4. Let n > 1. Then the adjoint representation of U(n) on u(n) (the space of n × n skew-Hermitian matrices) lifts to a Spin representation if and only if n is odd.

To see this, recall that the diagonal matrices diag(z1,...,zn) ∈ U(n) form a maximal torus Tn 6 U(n). Let Ers be the matrix with all entries 0 except 1 in the (r, s) place. For each pair r, s with 1 6 r

rs −1 rs −1 diag(z1,...,zn)E diag(z1,...,zn) = zrE zs 30 It follows that the weights are the ωr − ωs where ωk is the fundamental weight given by to the projection of T n onto the k-th factor. The sum of these is

~~(ωr − ωs)= (n − r)ωr − (r − 1)ωr 16r~~

~~Example 6.5. The root system for the rank 2 group G2 is shown in Figure 4. The Cartan~~

O✤ 3α + 2α ✤ 1 2 ✤ ✤ ✤ ✤ α2 ✤ ▼f ▼ X✶ ✤ F q8 ▼▼ α + α ✌ q ▼▼ 1 2 ✶ ✤ 2α1 + α2 q 3α1 + α2 ▼▼ ✌ q ▼▼ ✶ ✤ ✌ q ▼▼ ✶ q ▼▼ ✤ ✌ q ▼▼ ✶ q ▼▼ ✌ q ▼▼ ✶ ✤ q ▼▼ ✤ ✌ q ▼▼▼ ✶ q −α1 ▼▼ ✤ ✌q α1 o❴❴❴❴❴❴❴❴❴▼✶•q✤▼✌✶ / q✌ ▼ q ✤ ✶ ▼ q ✌ ▼ q ✤ ✶ ▼ q ✌ ▼ q ✤ ✶ ▼ q ✌ ▼ q ✤ ✶ ▼ q ✌ ✶ ▼ q ✌ ✤ ▼ q ✶ ▼ −3α1 − α2 q −2α − α ✌ ✤ −α − α ▼ −α2 q 1 2 ✶ 1 2 ▼ xq Ö✌ ✤ & ✤ ✤ ✤ ✤ ✤ ✤ −3α1 − 2α2

Figure 4. The roots of G2 matrix has determinant 1, so the weight and root lattices agree and the centre is trivial. One half of the sum of the positive roots is 1 10α + 6α = 5α + 3α , 2 1 2 1 2 which is a weight and therefore the adjoint representation on the Lie algebra g2 lifts to Spin. So for any compact connected Lie group G which contains a subgroup isomorphic to G2, G/G2 admits a Spin structure. One example of this is F4, so the homogeneous space F4/G2 of 2 2 dimension 36 has a Spin structure. We also know that for any maximal torus T 6 G2, G2/T has a Spin structure. See [DHH11, ﬁgure 4] for the cohomology of these examples as A-modules.

Example 6.6. For n > 1 the Lie group SU(n + 1) corresponds to the root system An and Z(SU(n + 1)) is cyclic of order n + 1. 31 The positive roots have the form αr + αr+1 + · · · + αs (1 6 r 6 s 6 n) and half the sum of these is k(n + 1 − k) (6.5) α . 2 k 6 6 1 Xk n When n is even, k and n + 1 − k have opposite parity, so the coeﬃcients are all integers so this is not in the weight lattice of SU(n + 1)/Z(SU(n + 1)). When n is odd, for each odd value of k the coeﬃcient of αk is odd, so this is not in the weight lattice of SU(n + 1)/Z(SU(n + 1)). So the adjoint representation of SU(n + 1)/Z(SU(n + 1)) admits a Spin structure if and only if n is even.

Example 6.7. For n > 2, Bn is the root system of Spin(2n + 1) and the centre has order 2 where Spin(2n + 1)/Z(Spin(n + 1)) =∼ SO(2n + 1). It turns out that in the root lattice the sum of the positive roots is congruent to

~~kαk mod 2, 6 6 1 Xk n which implies that the adjoint representation of SO(2n + 1) does not have a Spin structure.~~

~~Example 6.8. The exceptional Lie group F4 has rank 4 and its Cartan matrix has determinant 1 so the root and weight lattices coincide. One half of the sum of the positive roots is the weight~~

~~8α1 + 15α2 + 21α3 + 11α4, so the adjoint representation on the Lie algebra f4 has a Spin structure.~~

Example 6.9. For the exceptional Lie groups E6, E7 and E8, the Cartan matrices have deter- minants 3, 2 and 1 respectively and these are the orders of their centres. Here are the formulae for half the sums of the positive roots:

~~E6 : 8α1 + 11α2 + 15α3 + 21α4 + 15α5 + 8α6, 49 75 27 E : 17α + α + 33α + 48α + α + 26α + α , 7 1 2 2 3 4 2 5 6 2 7 E8 : 46α1 + 68α2 + 91α3 + 135α4 + 110α5 + 84α6 + 57α7 + 29α8.~~

~~For E6/Z(E6)andE8/Z(E8), the adjoint representations admit Spin structures, but for E7/Z(E7) the expression is not a weight so the adjoint representation does not admit a Spin structure.~~

Example 6.10. Consider the 18-dimensional homogeneous space SO(8)/Sp(2). The cohomology is easily determined using for example the Eilenberg-Moore spectral sequence ∗ 2 H (BSO(8)) ∗ F t−s Es,t = Tors,t (H (BSp(4))), 2)=⇒ H (SO(8)/Sp(2)). Here the inclusion Sp(2) ֒→ SO(8) induces the algebra homomorphism ∗ ∗ H (BSO(8)) / H (BSp(4))

~~F2[w2, w3, w4, w5, w6, w7, w8] w4k 7→ ℘k F2[℘1,℘2]~~

4k 2 where ℘k ∈ H (BSp(4)) is the k-th symplectic Pontrjagin class. The E -term is the exterior algebra 2 E∗,∗ = Λ(U1,U2,U4,U5,U6) 32 where Uk has bidegree (1, k + 1) and is represented in the bar construction by the class [wk+1]. This spectral sequence has Steenrod operations and using the action on the Stiefel-Whitney classes we ﬁnd that 1 2 Sq U1 = U2, Sq U2 = U4. The spectral sequence collapses and so H∗(SO(8)/Sp(2)) is generated as an algebra by the elements uk represented by Uk, where 2 3 4 u2 = u1, u3 = u1, u4 = u1. Therefore it has a basis consisting of the monomials a b c u1u5u6 (0 6 a 6 7, b, c ∈ {0, 1}) and the self dual A(1)-module structure shown below. The kO-module kO ∧ (SO(8)/Sp(2)+) ∗ is Spanier-Whitehead stably self dual and HkO(kO ∧ (SO(8)/Sp(2)+)) agrees with this A(1)- module. 7 u1u5u6 6 u1u5u6 5 u1u5u6 4 u1u5u6 3 u1u5u6 7 2 u1u6 u1u5u6 7 6 u u u u1u5 u1u6 1 5 6 6 5 u u u1u5 u1u6 5 6 5 4 u1u5 u1u6 4 3 u1u5 u1u6 3 2 u1u5 u1u6 7 2 u u u1 u1u5 1 6 6 u u u u1 1 5 6 5 u u1 5 4 u1 3 u1 2 u1

~~33 References~~

[AP76] J. F. Adams and S. B. Priddy, Uniqueness of BSO, Math. Proc. Camb. Phil. Soc. 80 (1976), 475–509. [AS74] M. F. Atiyah and L. Smith, Compact Lie groups and the stable homotopy of spheres, Topology 13 (1974), 135–142. [Bak18] A. Baker, Iterated doubles of the Joker and their realisability, Homology, Homotopy Appl. 20 (2018), 341–360. [BL01] A. Baker and A. Lazarev, On the Adams spectral sequence for R-modules, Algebr. Geom. Topol. 1 (2001), 173–199. [BM04] A. J. Baker and J. P. May, Minimal atomic complexes, Topology 43 (2004), 645–665. [BJM84] M. G. Barratt, J. D. S. Jones, and M. E. Mahowald, Relations amongst Toda brackets and the Kervaire invariant in dimension 62, J. London Math. Soc. (2) 30 (1984), 533–550. + [BHL 21] A. Beaudry, M. A. Hill, T. Lawson, X. D. Shi, and M. Zeng, Quotient rings of HF2 ∧ HF2 (2021), available at arXiv:2103.14707. [Bau04] T. Bauer, p-compact groups as framed manifolds, Topology 43 (2004), no. 3, 569–597. [BP06] T. Bauer and E. K. Pedersen, The realizability of local loop spaces as manifolds, K-Theory 37 (2006), no. 3, 329–339. [BS72] R. W. Brockett and H. J. Sussmann, Tangent bundles of homogeneous spaces are homogeneous spaces, Proc. Amer. Math. Soc. 35 (1972), 550–551. [BS62] W. Browder and E. Spanier, H-spaces and duality, Pacific J. Math. 12 (1962), 411–414. [Bru12] R. R. Bruner, On the Postnikov towers for real and complex connective K-theory (2012), available at arXiv:1208.2232. [Bru14] , Idempotents, localizations and Picard groups of A(1)-modules, Contemp. Math. 617 (2014), 81–108. [BG10] R. R. Bruner and J. P. C. Greenlees, Connective Real K-theory of Finite Groups, Mathematical Surveys and Monographs, vol. 169, Amer. Math. Soc., 2010. [DP80] A. Dold and D. Puppe, Duality, trace, and transfer, Proceedings of the International Conference on Geometric Topology (Warsaw, 1978), PWN, Warsaw, 1980, pp. 81–102. [DHH11] C. L. Douglas, A. G. Henriques, and M. A. Hill, Homological obstructions to string orientations, Int. Math. Res. Not. IMRN 18 (2011), 4074–4088. [EKMM97] A. D. Elmendorf, I. Kriz, M. A. Mandell, and J. P. May, Rings, Modules and Algebras in Stable Homotopy Theory, Vol. 47, 1997. With an appendix by M. Cole. [FMS97] D. Fischman, S. Montgomery, and H.-J. Schneider, Frobenius extensions of subalgebras of Hopf algebras, Trans. Amer. Math. Soc. 349 (1997), 4857–4895. [GLM92] P. Goerss, J. Lannes, and F. Morel, Vecteurs de Witt non commutatifs et représentabilité de l’homologie modulo p, Invent. Math. 108 (1992), 163–227 (French). [Hum78] J. E. Humphreys, Symmetry for finite dimensional Hopf algebras, Proc. Amer. Math. Soc. 68 (1978), no. 2, 143–146. [Koc82] S. O. Kochman, Integral cohomology operations, Current Trends in Algebraic Topology, Part 1 (London, Ont., 1981), CMS Conf. Proc., vol. 2, Amer. Math. Soc., 1982, pp. 437–478. [Lam99] T. Y. Lam, Lectures on Modules and Rings, Graduate Texts in Mathematics, vol. 189, Springer- Verlag, 1999. [LS69] R. G. Larson and M. E. Sweedler, An associative orthogonal bilinear form for Hopf algebras, Amer. J. Math. 91 (1969), 75–94. [Lor11] M. Lorenz, Some applications of Frobenius algebras to Hopf algebras, Contemp. Math. 537 (2011), 269–289. [Lor18] , A Tour of Representation Theory, Graduate Studies in Mathematics, vol. 193, American Mathematical Society, 2018. [Mah81] M. Mahowald, bo-resolutions, Pacific J. Math. 92 (1981), 365–383. [MM76] M. Mahowald and R. J. Milgram, Operations which detect Sq4 in connective K-theory and their applications, Quart. J. Math. Oxford Ser. (2) 27 (1976), 415–432.

34 [Mar83] H. R. Margolis, Spectra and the Steenrod Algebra: Modules over the Steenrod algebra and the stable homotopy category, North-Holland, 1983. [Mon93] S. Montgomery, Hopf Algebras and their Actions on Rings, CBMS Regional Conference Series in Mathematics, vol. 82, 1993. [NZ89] W. D. Nichols and M. B. Zoeller, A Hopf algebra freeness theorem, Amer. J. Math. 111 (1989), 381–385. [Pow15] G. Powell, Truncated projective spaces, Brown-Gitler spectra and indecomposable A(1)-modules, Topology Appl. 183 (2015), 45–85. [Rog08] J. Rognes, Galois Extensions of Structured Ring Spectra. Stably Dualizable Groups, Mem. Amer. Math. Soc. 192 (2008), no. 898. [Sin82] W. Singhof, Parallelizability of homogeneous spaces. I, Math. Ann. 260 (1982), 101–116. [SW86] W. Singhof and D. Wemmer, Parallelizability of homogeneous spaces. II, Math. Ann. 274 (1986), 157–176. [SW87] , Erratum: Parallelizability of homogeneous spaces. II, Math. Ann. 276 (1987), 699–700. [Wal60] C. T. C. Wall, Generators and relations for the Steenrod algebra, Annals Math. 72 (1960), 429–444. [Whi70] G. W. Whitehead, Recent Advances in Homotopy Theory, Vol. 5, American Mathematical Society, 1970. Conference Board of the Mathematical Sciences Regional Conference Series in Mathematics. [Woo98] R. M. W. Wood, Problems in the Steenrod algebra, Bull. London Math. Soc. 30 (1998), no. 5, 449–517.

~~School of Mathematics & Statistics, University of Glasgow, Glasgow G12 8QQ, Scotland. Email address: [email protected] URL: http://www.maths.gla.ac.uk/∼ajb~~