Major problems 2 — The generating hypothesis, which asserts that the stable homotopy functor is faithful on the category…

Major problems 2 — The generating hypothesis, which asserts that the stable homotopy functor is faithful on the category…

Definitions and mathematical statement

A finite spectrum is a spectrum built from finitely many cells, or equivalently an object of the finite stable homotopy category, with the usual retract convention. Write $[X,Y]$ for stable homotopy classes of maps and $\pi_nX=[\Sigma^nS,X]$. The graded group $\pi_*X$ is a module over the stable homotopy ring $\pi_*S$.

Freyd's generating hypothesis says that for finite spectra $X,Y$, \[ \left(\pi_n(f)=0\text{ for every }n\in{\mathbb{Z}}\right) \Longrightarrow f=0\text{ in }[X,Y]. \] Equivalently, $[X,Y]\to\operatorname{Hom}_{\pi_*S}(\pi_*X,\pi_*Y)$ is injective, with degree-zero graded homomorphisms on the right. The target is faithfulness, not surjectivity of this map. Allowing arbitrary infinite spectra changes the statement.

Formulation and concept references: [S1].

Short English statement

Is a map between finite spectra necessarily null-homotopic when it acts as zero on every stable homotopy group?

Research attempt

Scope and literature check.

This attempt by GPT-6 Astra Ultra was prepared on 27 September 2026. Call a map $f$ a ghost when $\pi_*(f)=0$. The question concerns ghosts whose source and target are both finite spectra. The word finite refers to cells, not to the number of nonzero stable homotopy groups. Even a sphere has infinitely many possible degrees to consider.

Hovey [S1] gives the precise faithfulness statement and consequences of it, including restrictions on finite generation over the sphere's homotopy ring. Lockridge [E1] studies related questions in derived categories; the ambient category is an essential part of those results. The elementary arguments below isolate two tempting approaches: induction over cells and construction of a universal ghost. The first gives nilpotence of sufficiently long composites, while the second generally leaves the finite category. An explicit algebraic example shows that these are substantive obstructions, and a Moore-spectrum calculation tests why that example does not automatically transfer to topology.

Lemma 482.1: the statement holds for retracts of sphere wedges.

Let $W=\bigvee_{j=1}^r\Sigma^{n_j}S$, and let $f:W\to Y$ be a ghost, where $Y$ need not be finite. Its composite with the inclusion of the $j$th summand is the image under $\pi_{n_j}(f)$ of that inclusion. Every such composite is zero. By the coproduct property, $f=0$.

More generally let $X$ be a retract of $W$, with maps $i:X\to W$ and $r:W\to X$ satisfying $ri=1_X$. If $f:X\to Y$ is a ghost, then $fr$ is a ghost out of $W$, hence $fr=0$ and $f=fri=0$. This proves an actual finite subcase without putting any restriction on the target. It does not cover arbitrary cell attachments: a cofiber of a map of spheres need not split as a wedge of spheres.

Lemma 482.2: finite cell filtrations bound lengths of ghost composites.

Suppose a spectrum $X$ has a filtration \[ 0=X_0\longrightarrow X_1\longrightarrow\cdots\longrightarrow X_r=X \] with cofiber triangles $X_{j-1}\to X_j\to W_j\to\Sigma X_{j-1}$, where each $W_j$ is a finite wedge of suspensions of $S$. Then every composite of $r$ ghosts starting at $X$ is zero, even if its intermediate targets are not finite.

Proof. For $r=1$ this is Lemma 482.1. Write the last triangle as \[ A\xrightarrow{i}X\xrightarrow{q}W\longrightarrow\Sigma A. \] Given ghosts $f_j:Y_{j-1}\to Y_j$ with $Y_0=X$, set $F=f_{r-1}\cdots f_1$. The map $f_1i$ is a ghost, so the induction hypothesis for $A$ gives $Fi=0$. Exactness of $[-,Y_{r-1}]$ on the triangle produces $h:W\to Y_{r-1}$ such that $F=hq$. The last ghost $f_r$ kills every map from a sphere, hence $f_rh=0$ and $f_rF=0$. This completes the induction.

If $X$ is only a retract of such a filtered spectrum $Z$, precompose the first ghost with the retraction $Z\to X$ and then postcompose the resulting zero relation with the inclusion $X\to Z$. The same conclusion follows. Thus every finite spectrum has a finite bound of this kind. In particular, a ghost endomorphism of a fixed finite spectrum is nilpotent.

This lemma does not prove that a single ghost is zero. The ring $\mathbb Z[\epsilon]/(\epsilon^2)$ has a nonzero nilpotent element $\epsilon$. More concretely, a two-cell source gives a bound of two ghosts, while the conjecture asks for a bound of one. Repeating the filtration argument supplies no mechanism for removing that last factor.

The obstruction in an attempted one-ghost induction.

Suppose $A$ and $W$ in the last triangle are already known to have no nonzero ghosts, and let $f:X\to Y$ be a ghost. One obtains $fi=0$, so $f=hq$ for some $h:W\to Y$. It is tempting to claim that $h$ is also a ghost. The homotopy exact sequence only says \[ \operatorname{im}\pi_n(q) =\ker\left(\pi_n(W)\longrightarrow\pi_{n-1}(A)\right). \] The condition on $f$ says that $\pi_n(h)$ kills this kernel. It does not say that $\pi_n(h)$ kills the rest of $\pi_n(W)$. Surjectivity of $\pi_n(q)$ in all degrees would suffice, but that is an additional splitting-type condition, not a property of an arbitrary cell attachment. The successful composite argument used a new ghost after $h$; there is no such map available when proving faithfulness of $f$ alone.

Lemma 482.3: universal ghosts exist, usually with infinite targets.

For each integer $n$ and each element $a\in\pi_nX$, choose a representing map $a:\Sigma^nS\to X$. Form the wedge and evaluation map \[ W_X=\bigvee_{n\in\mathbb Z}\ \bigvee_{a\in\pi_nX}\Sigma^nS, \qquad e:W_X\longrightarrow X. \] Complete it to a cofiber triangle \begin{equation}\label{wsref-0-61-34-38-32-3a-75-6e-69-76-65-72-73-61-6c} W_X\xrightarrow{e}X\xrightarrow{u}G_X\longrightarrow\Sigma W_X. \end{equation} The map $\pi_n(e)$ is onto for every $n$, because its summands include a representative of every element. Exactness therefore gives $\pi_n(u)=0$: $u$ is a ghost. If $f:X\to Y$ is any other ghost, all its composites with summands of $W_X$ vanish. Thus $fe=0$, and exactness gives a factorization $f=hu$ for some $h:G_X\to Y$.

This is a universal factorization property, not a finite counterexample. The wedge $W_X$ is generally infinite, and the construction does not make $G_X$ finite. Applying the desired conjecture to $u$ at this point would change its hypotheses. Conversely, proving that every map $hu$ with finite target is zero would prove the conjecture, but the universal construction itself supplies no such vanishing theorem.

A conditional finite-generation consequence.

Write $R=\pi_*S$. Suppose the graded $R$-module $\pi_*X$ is generated by finitely many homogeneous elements $a_1,\ldots,a_s$. Evaluation from the finite wedge \[ W=\bigvee_{j=1}^s\Sigma^{|a_j|}S\longrightarrow X \] is surjective on all homotopy groups: the $R$-action is realized by precomposing with stable maps of spheres. Its cofiber $G$ is finite if $X$ is finite, and $u:X\to G$ is a ghost. If Freyd's hypothesis holds, then $u=0$. Applying $[X,-]$ to the triangle shows that $1_X$ lifts to a map $X\to W$, so evaluation has a right inverse and $X$ is a retract of a finite sphere wedge.

Equivalently, a finite spectrum whose homotopy module is finitely generated but which is not such a retract would supply a counterexample through this construction. No example satisfying both requirements is constructed here. This is a rederivation of the basic finite-generation obstruction discussed in [S1], in the retract form sufficient for this argument, not a new consequence claimed independently of the literature.

Finitely many cells do not imply finitely generated homotopy as an $R$-module. In a cofiber long exact sequence, kernels of module maps occur as well as cokernels. Finite generation of kernels would require additional ring-theoretic hypotheses; none has been established for the modules in this argument. Counting attaching maps only counts the cell presentation, not generators in all subsequent stable degrees.

A finite algebraic ghost that exposes the danger of analogy.

Consider homological chain complexes of free abelian groups. Let $m\ge2$ and let $P$ have $P_1=P_0=\mathbb Z$, differential $d_1=m$, and all other terms zero. Use the shift convention $(\Sigma P)_j=P_{j-1}$, with shifted differential $-d$. Define \[ f:P\longrightarrow\Sigma P, \qquad f_1=1:\mathbb Z\to\mathbb Z, \quad f_j=0\ (j\ne1). \] This is a chain map: the only possible differential condition has both sides zero. The sole homology of $P$ is $H_0(P)=\mathbb Z/m$, while the sole homology of $\Sigma P$ is $H_1(\Sigma P)=\mathbb Z/m$. The degree-zero map $f$ therefore induces zero in every homology degree.

Nevertheless $f$ is not chain null-homotopic. A chain homotopy $s_j$ from $f$ to zero would satisfy $f=ds+sd$. In degree one this requires \begin{equation}\label{wsref-0-61-34-38-32-3a-68-6f-6d-6f-74-6f-70-79} 1=-m s_1+m s_0, \end{equation} where $s_0,s_1$ are integers. This is impossible modulo $m$. More generally, replacing $f_1=1$ by $f_1=c$ gives a null-homotopic map exactly when $m\mid c$: necessity follows from the same equation, and sufficiency follows by choosing $s_1=0$, $s_0=c/m$.

These bounded free complexes compute their own morphisms in the derived category. One can justify this fact by filtering a bounded projective complex and using projectivity to see that maps into an acyclic complex form an acyclic Hom complex. Hence localizing at quasi-isomorphisms does not turn the displayed nonzero homotopy class into zero. We have exhibited a ghost between perfect complexes in $D(\mathbb Z)$. Its composite with its shift is zero already as a chain map, since $P$ and $\Sigma^2P$ have no common nonzero degree. It therefore fits the length-two nilpotence conclusion exactly.

This is not a counterexample between finite sphere spectra. The generator in this derived category has endomorphism ring $\mathbb Z$ concentrated in degree zero. The sphere has nonzero positive stable stems. Passing to Eilenberg--Mac Lane module spectra changes the generator and the meaning of finite cells. The algebraic example diagnoses why a purely formal triangulated-category induction cannot settle the original question without special information about the sphere.

Testing the analogous map on an actual Moore spectrum.

Let $M(m)$ be the finite cofiber of multiplication by $m$ on $S$: \[ S\xrightarrow{m}S\xrightarrow{i}M(m)\xrightarrow{q}\Sigma S. \] Define the candidate Bockstein map \begin{equation}\label{wsref-0-61-34-38-32-3a-62-6f-63-6b-73-74-65-69-6e} \beta=(\Sigma i)q:M(m)\longrightarrow\Sigma M(m). \end{equation} Ordinary integral homology is concentrated in degree zero for $M(m)$ and degree one for $\Sigma M(m)$, so $H_*(\beta;\mathbb Z)=0$. That test alone is weaker than being a ghost.

Put $A_n=\pi_nS$. The exact sequence gives \[ \operatorname{im}\left(\pi_nq\right)=A_{n-1}[m],\qquad \ker\left(\pi_n(\Sigma i)\right)=mA_{n-1}, \] where $A[m]=\{a:ma=0\}$. Consequently \begin{equation}\label{wsref-0-61-34-38-32-3a-63-72-69-74-65-72-69-6f-6e} \pi_*(\beta)=0 \quad\Longleftrightarrow\quad A_j[m]\subseteq mA_j\quad\hbox{for every integer }j. \end{equation} This is an exact criterion obtained from the cofiber sequence; it is not a truncation of the homotopy test.

For $m=2$, the first stable stem is $A_1=\mathbb Z/2$, generated by the stable Hopf class (the standard calculation recalled in [E2]). Thus $A_1[2]=A_1$ and $2A_1=0$, violating the inclusion. More explicitly, $\pi_2M(2)\to A_1$ is onto, while $A_1\to\pi_1M(2)$ is injective. Choose $x\in\pi_2M(2)$ lifting its nonzero element. Then $\pi_2(\beta)(x)\ne0$. The candidate is therefore not a ghost, despite being zero on ordinary integral homology. This concrete low-degree failure prevents the algebraic chain-complex construction from providing the proposed topological counterexample.

An affirmative algebraic comparison and its limitation.

Over a field $K$, a bounded complex of finite-dimensional vector spaces splits into its homology, with zero differentials, and contractible two-term complexes. To see this, choose complements \[ B_n\subset Z_n\subset C_n, \qquad Z_n=B_n\oplus H_n',\qquad C_n=Z_n\oplus L_n. \] The differential identifies $L_n$ with $B_{n-1}$. Those pairs are contractible; $H_n'$ represents the homology. A map inducing zero on homology is consequently zero in the derived category. This proves faithfulness for perfect complexes over a field.

The contrast with the explicit $\mathbb Z$-complex is explained by the existence of linear complements, not by finiteness of the complexes alone. Sphere spectra do not have an analogous decomposition supplied by choosing a basis of their graded homotopy groups. A proposed argument that tacitly selects such complements has assumed precisely the sort of splitting it needs to prove.

Finite checks and what an actual certificate would require.

The chain-map calculation reduces exactly to the residue class of $c$ modulo $m$; finite checks of small $m$ verify this arithmetic but are not evidence that the stable homotopy category behaves the same way. The Moore-spectrum calculation uses the known first stable stem and full exactness, not a computer claim that higher stems vanish.

For a proposed finite-spectrum counterexample, testing a finite range of degrees is only a rejection test: a single nonzero induced map rules it out, as above. Passing finitely many degrees does not establish the universal quantifier. A successful counterexample would need an all-degree mechanism forcing $\pi_*(f)=0$, together with an independent proof $f\ne0$. A successful proof would need to show that the residual factor $h$ in the one-ghost induction cannot produce a nonzero $hq$, or otherwise force every universal-ghost factorization with finite target to vanish.

UNRESOLVED. The argument proves the sphere-wedge retract subcase, a finite bound on composite ghost length for every finite source, and an exact diagnostic for a Moore-spectrum candidate. It neither proves faithfulness for all finite spectra nor constructs a finite topological ghost. The algebraic counterexample is explicitly confined to its own category; no general solution or novelty claim is made.

Sources

• S1 Mark Hovey, On Freyd's Generating Hypothesis, Quarterly Journal of Mathematics 58 $2007$, 31–45. https://math.uchicago.edu/~may/PEOPLE/AMB/Hovey_OnFreydsGenHyp.pdf. Formulation source.
• E1 Keir H. Lockridge, The generating hypothesis in the derived category of $R$-modules, arXiv:math/0511534. https://arxiv.org/abs/math/0511534. Context for algebraic analogues; the explicit complex calculation here is given in full and is not transferred to finite sphere spectra.
• E2 Allen Hatcher, Algebraic Topology, Chapter 4, homotopy groups of spheres and stabilization, especially Section 4.2. https://pi.math.cornell.edu/~hatcher/AT/ATch4.pdf. Reference for the standard first stable stem $\pi_1S=\mathbb Z/2$.