Categories and Cohomology Theories

Graeme Segal. Topology, Vol. 13, pp. 293–312. Pergamon Press, 1974.

Relations

Builds on: (Quillen, unpublished; ideas on algebraic K-theory), papers/papers/boardman-vogt-homotopy-everything|Boardman–Vogt (1968) (no note yet), (Milnor, geometric realization of semi-simplicial complexes), (Barratt–Priddy 1972) (no note yet) Extended by: (Bousfield–Friedlander 1978 homotopy theory of Γ-spaces) (no note yet), (May, \(E_\infty\) operads and Γ-spaces comparison) (no note yet) Concepts used: concepts/category-theory/foundations/01-categories-functors-natural-transformations|Categories, Functors, and Natural Transformations, concepts/category-theory/foundations/03-limits-colimits|Limits and Colimits, concepts/category-theory/foundations/05-kan-extensions|Kan Extensions

Table of Contents

• #Overview|Overview
  • #Historical Significance|Historical Significance
  • #Main Themes|Main Themes
  • #What to Get Out of This Paper|What to Get Out of This Paper
• #1. The Category Γ|1. The Category Γ
• #2. Γ-Spaces: Definition and the Segal Condition|2. Γ-Spaces: Definition and the Segal Condition
  • #2.1 The Segal Condition|2.1 The Segal Condition
  • #2.2 Special and Very Special Γ-Spaces|2.2 Special and Very Special Γ-Spaces
  • #2.3 Γ-Spaces as Simplicial Spaces|2.3 Γ-Spaces as Simplicial Spaces
• #3. The Classifying-Space Construction and Spectra|3. The Classifying-Space Construction and Spectra
  • #3.1 The Delooping Machine|3.1 The Delooping Machine
  • #3.2 Proposition 1.4 and Its Significance|3.2 Proposition 1.4 and Its Significance
• #4. Γ-Categories from Symmetric Monoidal Categories|4. Γ-Categories from Symmetric Monoidal Categories
  • #4.1 Definition of a Γ-Category|4.1 Definition of a Γ-Category
  • #4.2 Construction from Sums|4.2 Construction from Sums
  • #4.3 Key Examples|4.3 Key Examples
• #5. Γ-Spaces and Spectra: The Adjunction|5. Γ-Spaces and Spectra: The Adjunction
  • #5.1 The Spectrum Associated to a Γ-Space|5.1 The Spectrum Associated to a Γ-Space
  • #5.2 The Γ-Space Associated to a Spectrum|5.2 The Γ-Space Associated to a Spectrum
  • #5.3 Adjointness and the Main Equivalence|5.3 Adjointness and the Main Equivalence
• #6. The Barratt–Priddy–Quillen Theorem|6. The Barratt–Priddy–Quillen Theorem
  • #6.1 The Γ-Space BC|6.1 The Γ-Space BC
  • #6.2 Proof Sketch via Adjointness|6.2 Proof Sketch via Adjointness
  • #6.3 Stable Cohomotopy as K-Theory|6.3 Stable Cohomotopy as K-Theory
• #7. Group Completion and the Grothendieck Construction|7. Group Completion and the Grothendieck Construction
  • #7.1 The Group Completion Problem|7.1 The Group Completion Problem
  • #7.2 The Space A’|7.2 The Space A’
  • #7.3 Quillen’s Plus-Construction and K-Theory|7.3 Quillen’s Plus-Construction and K-Theory
• #8. Ring Spectra|8. Ring Spectra
• #9. Relationship with Operads (Boardman–Vogt–May)|9. Relationship with Operads (Boardman–Vogt–May)
• #10. Realization of Simplicial Spaces|10. Realization of Simplicial Spaces
• #References|References

Overview 🗺️

Historical Significance

By the early 1970s, algebraic topology faced a pressing structural question: which spaces admit the structure of infinite loop spaces — that is, which spaces \(X\) arise as \(X \simeq \Omega^\infty Y\) for some spectrum \(Y\)? The answer matters because infinite loop spaces are exactly the zeroth spaces of connective spectra, and spectra represent (generalized) cohomology theories. So the question amounts to: which spaces “see” a cohomology theory?

Prior to Segal’s paper, the best available tools were the operadic machines of Boardman–Vogt and May (\(E_\infty\) operads), which characterize infinite loop spaces in terms of higher coherence homotopies for the multiplication. These approaches were powerful but technically formidable — the coherence data lives in a tower of spaces with complex interrelations.

Segal’s 1974 paper introduced a strikingly cleaner alternative: the Γ-space machine. The key insight is that the combinatorics of “commutative addition up to homotopy” are already fully encoded in the category \(\Gamma\) of finite sets and partial maps. A Γ-space is simply a functor \(A: \Gamma^{\mathrm{op}} \to \mathbf{Top}\) satisfying a homotopy-coherence condition (the Segal condition). From any such functor, Segal extracts a full spectrum automatically, with no additional coherence data required. The machine is adjoint-theoretic at its core, and its outputs are canonical.

The paper’s impact has been enormous. It:

• gave the first clean, categorical proof of the Barratt–Priddy–Quillen theorem (the sphere spectrum \(\mathbb{S}\) is the K-theory of finite sets);
• provided a general machine for constructing the K-theory spectrum of any symmetric monoidal category, subsuming Quillen’s plus-construction as a special case;
• established the precise adjoint relationship between Γ-spaces and connective spectra, showing these two worlds are equivalent;
• seeded decades of subsequent work: Bousfield–Friedlander’s model structure on Γ-spaces, Schwede–Shipley’s comparison with symmetric spectra, and the modern \(\infty\)-categorical perspective via Lurie’s \(\mathbb{E}_\infty\)-spaces.

In retrospect, Segal’s paper is one of the founding documents of higher algebra — the study of ring- and module-like structures in homotopy theory.

Main Themes

The paper is organized around three interlocking ideas:

1. The Segal condition as homotopy commutativity. The category \(\Gamma\) encodes all the combinatorics of abelian-group-like structure. The Segal condition \(A(\mathbf{n}) \simeq A(\mathbf{1})^n\) is a homotopy version of the statement “\(A\) is a commutative monoid.” When the condition is strengthened so that \(\pi_0 A(\mathbf{1})\) is a group (the very special condition), the space \(A(\mathbf{1})\) is an infinite loop space. This hierarchy — Γ-space → special → very special → infinite loop space — is a prototype for the hierarchy of \(E_n\)-algebras central to modern homotopy theory.

2. The Γ-category construction. Any symmetric monoidal category \((\mathcal{C}, \oplus, 0)\) gives rise to a Γ-space \(A_\mathcal{C}\) by letting \(A_\mathcal{C}(S)\) parametrize “\(S\)-indexed sums” in \(\mathcal{C}\). This is a categorification of the observation that an abelian group \(A\) assigns to each finite set \(S\) the product \(A^S\), functorially. The resulting spectrum \(\{B_n\}\) deloops \(B|\mathcal{N}\mathcal{C}|\), the classifying space of the nerve, producing the K-theory spectrum of \(\mathcal{C}\) without any ad hoc construction.

3. Adjointness as the organizing principle. The relationship between Γ-spaces and spectra is not merely a correspondence but an adjunction \(A \dashv B\), which restricts to an equivalence on the subcategory of very special Γ-spaces and connective spectra. This adjoint-theoretic framing is characteristic of Segal’s style: rather than constructing things by hand, he identifies the universal property and reads off the structure. The same philosophy recurs in his later work on conformal field theory, loop groups, and \(K\)-homology.

What to Get Out of This Paper

Reading Segal (1974) rewards attention at several levels:

1. The Category Γ 📐

The central organizing object of Segal’s theory is a small category \(\Gamma\) whose morphisms encode all possible “multi-valued” maps between finite sets — exactly the combinatorial data required to parametrize associative, commutative composition laws up to homotopy.

Definition (The Category Γ). Let \(\Gamma\) be the category whose: - objects are all finite sets (including the empty set \(\mathbf{0} = \emptyset\)); - morphisms from \(S\) to \(T\) are functions \(\theta: S \to \mathcal{P}(T)\) (the power set of \(T\)) such that \(\theta(\alpha)\) and \(\theta(\beta)\) are disjoint whenever \(\alpha \neq \beta\).

Composition of \(\theta: S \to \mathcal{P}(T)\) and \(\phi: T \to \mathcal{P}(U)\) is \(\psi: S \to \mathcal{P}(U)\) defined by \[\psi(\alpha) = \bigcup_{\beta \in \theta(\alpha)} \phi(\beta).\]

A concrete Γ-morphism \(\phi: \mathbf{3} \to \mathbf{6}\) (left): each element of \(\mathbf{3}\) maps to a disjoint subset of \(\mathbf{6}\), here \(1 \mapsto \{1\}\), \(2 \mapsto \{3,6\}\), \(3 \mapsto \{4,5\}\). On the right, the same data is redrawn as the dual morphism \(\phi^{\mathrm{op}}: \mathbf{6}^+ \to \mathbf{3}^+\) in \(\Gamma^{\mathrm{op}} \simeq \mathbf{Fin}_*\), the category of finite pointed sets, where each element of \(\mathbf{6}^+\) is sent to the unique element of \(\mathbf{3}^+\) whose fiber contains it (or to the basepoint \(0\) if it is in no fiber). (From the Machine Appreciation blog, 2021.)

[!TIP]- Solution to Exercise 1 (a) Morphisms \(\mathbf{1} \to \mathbf{2}\): a morphism is \(\theta: \{1\} \to \mathcal{P}(\{1,2\})\). The value \(\theta(1)\) can be any subset of \(\{1,2\}\): namely \(\emptyset\), \(\{1\}\), \(\{2\}\), or \(\{1,2\}\) — giving four morphisms. (The disjointness condition on distinct inputs is vacuous since \(|\mathbf{1}| = 1\).)

Morphisms \(\mathbf{2} \to \mathbf{1}\): a morphism is \(\theta: \{1,2\} \to \mathcal{P}(\{1\})\). The disjointness condition requires \(\theta(1) \cap \theta(2) = \emptyset\). Since \(\mathcal{P}(\{1\}) = \{\emptyset, \{1\}\}\), the possible assignments are:

\[(\theta(1),\theta(2)) \in \{(\emptyset,\emptyset),\,(\{1\},\emptyset),\,(\emptyset,\{1\}),\,(\{1\},\{1\})\}.\]

Wait — \((\{1\},\{1\})\) violates disjointness. So there are exactly three valid morphisms: \((\emptyset,\emptyset)\), \((\{1\},\emptyset)\), \((\emptyset,\{1\})\).

(b) Suppose \(\theta: S \to \mathcal{P}(T)\) is an isomorphism, with inverse \(\phi: T \to \mathcal{P}(S)\). The composition \(\psi = \phi \circ \theta: S \to \mathcal{P}(S)\) must equal the identity, meaning \(\psi(s) = \{s\}\) for all \(s\). Expanding: \(\psi(s) = \bigcup_{t \in \theta(s)} \phi(t)\). For this to equal \(\{s\}\), each \(\theta(s)\) must be non-empty (otherwise \(\psi(s) = \emptyset\)), and the union of \(\phi(t)\) over \(t \in \theta(s)\) collapses to \(\{s\}\). A clean sufficient condition is \(|\theta(s)| = 1\) for all \(s\) and the singletons \(\theta(s) = \{t_s\}\) are pairwise distinct (so they partition \(T\)). Then \(\phi(t_s) = \{s\}\) defines the inverse and all conditions are satisfied. Conversely if any \(\theta(s)\) has \(|\theta(s)| \geq 2\) or the \(\theta(s)\) are not disjoint singletons, the inverse \(\phi\) cannot recover \(s\) uniquely.

(c) By (b), \(\mathrm{Aut}_\Gamma(\mathbf{n})\) consists of maps \(\theta: \{1,\ldots,n\} \to \mathcal{P}(\{1,\ldots,n\})\) where each \(\theta(i) = \{\sigma(i)\}\) for a bijection \(\sigma: \{1,\ldots,n\} \to \{1,\ldots,n\}\). The correspondence \(\theta \leftrightarrow \sigma\) is a bijection, and composition of morphisms in \(\Gamma\) corresponds exactly to composition of permutations. Hence \(\mathrm{Aut}_\Gamma(\mathbf{n}) \cong \Sigma_n\). \(\square\)

The key morphisms to single out are the projections \(i_k: \mathbf{1} \to \mathbf{n}\) defined by \(i_k(1) = \{k\}\) for \(1 \leq k \leq n\). Their duals \(i_k^*: A(\mathbf{n}) \to A(\mathbf{1})\) are the components of the Segal map.

[!TIP]- Solution to Exercise 2 (a) Given \(\theta: S \to \mathcal{P}(T)\) and \(\theta': S' \to \mathcal{P}(T')\) with \(S \cap S' = T \cap T' = \emptyset\), define \((\theta \sqcup \theta'): S \sqcup S' \to \mathcal{P}(T \sqcup T')\) by \((\theta \sqcup \theta')(s) = \theta(s)\) for \(s \in S\) and \((\theta \sqcup \theta')(s') = \theta'(s')\) for \(s' \in S'\). The disjointness condition is preserved since the two components live in disjoint subsets of \(T \sqcup T'\). Functoriality: for composable pairs \((\theta,\phi)\) and \((\theta',\phi')\), the composition distributes over \(\sqcup\) because the two summands do not interact. The symmetry isomorphism \(S \sqcup S' \xrightarrow{\sim} S' \sqcup S\) is the evident relabeling, and the unit axiom holds since \(\theta \sqcup \emptyset_\emptyset = \theta\) where \(\emptyset_\emptyset: \emptyset \to \mathcal{P}(T)\) is the empty function.

(b) For any \(S\), the unique morphism \(\emptyset \to S\) is the empty function \(\theta: \emptyset \to \mathcal{P}(S)\), which is vacuously a valid Γ-morphism. Hence \(\mathbf{0} = \emptyset\) is an initial object.

(c) A terminal object \(T\) would require a unique morphism \(\mathbf{n} \to T\) for every \(n\). A morphism \(\mathbf{n} \to T\) is \(\theta: \{1,\ldots,n\} \to \mathcal{P}(T)\) satisfying pairwise disjointness. From \(\mathbf{n}\) to \(\mathbf{0} = \emptyset\), the only map sends every element to \(\emptyset \in \mathcal{P}(\emptyset)\) — exactly one such map exists for every \(n\). So \(\mathbf{0}\) is in fact the terminal object. Surprisingly, \(\mathbf{0}\) is both initial and terminal, making it a zero object in \(\Gamma\). The problem statement’s claim of “no terminal object” is incorrect: \(\mathbf{0}\) serves as both.

(d) Since \(\mathbf{0}\) is terminal, any Γ-space \(A\) with \(A(\mathbf{0}) \simeq *\) is asking that \(A\) send the terminal object to the terminal space — the normalization condition. Although \(\Gamma\) is not a category with finite products (the coproduct \(\sqcup\) plays the role of both), the condition \(A(\mathbf{0}) \simeq *\) ensures that \(A\) “preserves” the zero object in a homotopy-coherent sense, providing the unit for the H-space structure on \(A(\mathbf{1})\). \(\square\)

2. Γ-Spaces: Definition and the Segal Condition 📐

2.1 The Segal Condition

Definition (Γ-Space). A Γ-space is a contravariant functor \(A: \Gamma \to \mathbf{Top}\) satisfying: 1. \(A(\mathbf{0})\) is contractible; 2. for each \(n \geq 1\), the map \[\varphi_n \;=\; (i_1^*, \ldots, i_n^*) : A(\mathbf{n}) \longrightarrow A(\mathbf{1}) \times \cdots \times A(\mathbf{1}) \qquad (n\text{ factors})\] induced by the projections \(i_k: \mathbf{1} \to \mathbf{n}\), is a homotopy equivalence.

Condition (2) is the Segal condition (also called the Segal map condition). It asserts that \(A(\mathbf{n})\) is, up to homotopy, the \(n\)-fold Cartesian power of the single space \(A(\mathbf{1})\). The functor \(\Gamma\) provides higher coherence: the structure maps \(A(\theta)\) for all \(\theta\) encode an associative and commutative composition on \(A(\mathbf{1})\) up to all higher coherent homotopies.

[!TIP]- Solution to Exercise 3 (a) For \(\theta: \mathbf{m} \to \mathcal{P}(\mathbf{n})\) and \(\phi: \mathbf{k} \to \mathcal{P}(\mathbf{m})\), the composed morphism in \(\Gamma\) is \(\psi(i) = \bigcup_{j \in \phi(i)} \theta(j)\). We compute:

\[\hat{G}(\psi)(g_1,\ldots,g_n)_i = \sum_{k \in \psi(i)} g_k = \sum_{k \in \bigcup_{j \in \phi(i)}\theta(j)} g_k = \sum_{j \in \phi(i)} \sum_{k \in \theta(j)} g_k\]

where the last step uses the fact that the \(\theta(j)\) are pairwise disjoint (so the double union is a disjoint union and commutativity of \(+\) in \(G\) allows regrouping). The right-hand side is \(\hat{G}(\phi)\bigl(\hat{G}(\theta)(g_1,\ldots,g_n)\bigr)_i\). Hence \(\hat{G}(\psi) = \hat{G}(\phi) \circ \hat{G}(\theta)\), confirming functoriality.

(b) The Segal map \(\varphi_n: G^n \to G^n\) sends \((g_1,\ldots,g_n) \mapsto (i_1^*(g_1,\ldots,g_n),\ldots,i_n^*(g_1,\ldots,g_n))\) where \(i_k: \mathbf{1} \to \mathbf{n}\) with \(i_k(1) = \{k\}\). Then \(i_k^*(g_1,\ldots,g_n) = \sum_{j \in \{k\}} g_j = g_k\). So \(\varphi_n(g_1,\ldots,g_n) = (g_1,\ldots,g_n)\), the identity map on \(G^n\). This is trivially a homeomorphism.

(c) The map \(m_2: \mathbf{1} \to \mathbf{2}\) sends \(1 \mapsto \{1,2\}\). Then \(\hat{G}(m_2): G^2 \to G^1\) is \(\hat{G}(m_2)(g_1,g_2) = \sum_{j \in \{1,2\}} g_j = g_1 + g_2\). Via the Segal isomorphism (which is the identity), the induced binary operation on \(G \times G\) sends \((g_1,g_2) \mapsto g_1 + g_2\), which is exactly the original group operation \(+\) on \(G\). \(\square\)

2.2 Special and Very Special Γ-Spaces

The distinction between several levels of the Segal condition governs exactly what algebraic structure \(A(\mathbf{1})\) carries.

Definition (Special Γ-Space). A Γ-space \(A\) is special if \(\varphi_n\) is a homotopy equivalence for all \(n\) — i.e., the standard Segal condition as stated in Definition above. (Segal himself calls this simply a Γ-space satisfying (1.2).)

Definition (Very Special Γ-Space). A special Γ-space \(A\) is very special if, additionally, the monoid \(\pi_0(A(\mathbf{1}))\) is a group.

The hierarchy is: \[\{\text{topological abelian groups}\} \subsetneq \{\text{very special Γ-spaces}\} \subsetneq \{\text{special Γ-spaces}\} \subsetneq \{\text{Γ-spaces}\}\]

A topological abelian monoid \(M\) defines a Γ-space \(A\) with \(A(\mathbf{n}) = M^n\) and the projection maps being honest homeomorphisms (not just homotopy equivalences) — this is the case where the Segal condition holds strictly.

[!TIP]- Solution to Exercise 4 (a) If \(A\) is discrete and special, then \(A(\mathbf{n}) \cong A(\mathbf{1})^n\) as sets (the Segal condition becomes a bijection). The binary operation \(\mu: A(\mathbf{1}) \times A(\mathbf{1}) \cong A(\mathbf{2}) \xrightarrow{m_2^*} A(\mathbf{1})\) is defined by the unique Γ-morphism \(m_2: \mathbf{1} \to \mathbf{2}\) sending \(1 \mapsto \{1,2\}\). The unit is the image of the unique element of \(A(\mathbf{0}) \cong \{*\}\) under the map induced by \(\mathbf{0} \hookrightarrow \mathbf{1}\). Associativity follows from the Γ-morphism \(\mathbf{1} \to \mathbf{3}\) sending \(1 \mapsto \{1,2,3\}\) — both bracketings factor through \(A(\mathbf{3}) \cong A(\mathbf{1})^3\) by functoriality and must agree. Commutativity: the swap \(\sigma: \mathbf{2} \to \mathbf{2}\) with \(\sigma(1) = \{2\}\) and \(\sigma(2) = \{1\}\) is a Γ-isomorphism. Since \(m_2 \circ \sigma = m_2\) (as \(m_2: \mathbf{1} \to \mathbf{2}\) sends \(1 \mapsto \{1,2\} = \{2,1\}\) — sets are unordered), functoriality gives \(\mu(a,b) = A(m_2)(a,b) = A(m_2 \circ \sigma)(a,b) = A(m_2)(b,a) = \mu(b,a)\).

(b) Very special adds the condition that \(\pi_0(A(\mathbf{1})) = A(\mathbf{1})\) (discrete) is a group. The shear map \(A(\mathbf{2}) \to A(\mathbf{1}) \times A(\mathbf{1})\) sending \((a,b) \mapsto (a, \mu(a,b))\) must be a bijection. This means: for every \(a, c \in A(\mathbf{1})\) there is a unique \(b\) with \(\mu(a,b) = c\), i.e., every element has a right-inverse. Combined with the monoid structure from (a), \(A(\mathbf{1})\) is a group — and since it is commutative, an abelian group.

(c) The functor sending a discrete very special Γ-space \(A\) to the abelian group \(A(\mathbf{1})\) (with operation \(\mu\)) is well-defined and sends Γ-space maps to group homomorphisms. The inverse sends an abelian group \(G\) to the discrete Γ-space \(\hat{G}\) from Exercise 3. These functors are mutually inverse, establishing the equivalence of categories. \(\square\)

2.3 Γ-Spaces as Simplicial Spaces

There is a covariant functor \(\Delta \to \Gamma\) taking \([m] \mapsto \mathbf{m}\) and a non-decreasing map \(f: [m] \to [n]\) to the morphism \(\theta_f: \mathbf{m} \to \mathcal{P}(\mathbf{n})\) defined by \[\theta_f(i) = \{ j \in \mathbf{n} : f(i-1) < j \leq f(i) \}.\] Using this functor, every Γ-space \(A\) can be regarded as a simplicial space. The simplicial structure refines the Γ-structure and is the tool used to form realizations.

[!TIP]- Solution to Exercise 5 (a) Apply the formula \(\theta_f(i) = \{j \in \mathbf{n} : f(i-1) < j \leq f(i)\}\) with \(f = d^0: [1] \to [2]\), \(d^0(0)=1\), \(d^0(1)=2\), and \(\mathbf{n} = \mathbf{2}\). For \(i = 1\):

\[\theta_{d^0}(1) = \{j \in \{1,2\} : d^0(0) < j \leq d^0(1)\} = \{j : 1 < j \leq 2\} = \{2\}.\]

So \(\theta_{d^0}: \mathbf{1} \to \mathcal{P}(\mathbf{2})\) sends \(1 \mapsto \{2\}\).

(b) For \(f = s^0: [1] \to [0]\), \(s^0(0) = s^0(1) = 0\), and \(\mathbf{n} = \mathbf{0} = \emptyset\). For \(i = 1\):

\[\theta_{s^0}(1) = \{j \in \emptyset : s^0(0) < j \leq s^0(1)\} = \{j \in \emptyset : 0 < j \leq 0\} = \emptyset.\]

So \(\theta_{s^0}: \mathbf{1} \to \mathcal{P}(\mathbf{0})\) sends \(1 \mapsto \emptyset\).

(c) The map \(A(\theta_{d^0}): A(\mathbf{2}) \to A(\mathbf{1})\) is induced by the Γ-morphism \(1 \mapsto \{2\}\). Via the Segal equivalence \(A(\mathbf{2}) \simeq A(\mathbf{1}) \times A(\mathbf{1})\), this projects onto the second factor (the one indexed by \(\{2\} \subseteq \mathbf{2}\)) — it is the face map \(d_0: A_2 \to A_1\) in the associated simplicial space.

The map \(A(\theta_{s^0}): A(\mathbf{0}) \to A(\mathbf{1})\) is induced by \(1 \mapsto \emptyset\): it sends the unique point of \(A(\mathbf{0}) \simeq *\) to the element of \(A(\mathbf{1})\) corresponding to the “empty sum,” which is the unit/basepoint.

(d) Since \(A(\mathbf{0}) \simeq *\), the degeneracy \(A(\theta_{s^0}): A(\mathbf{0}) \to A(\mathbf{1})\) selects a canonical basepoint in \(A(\mathbf{1})\) — the image of the unique element of \(A(\mathbf{0})\). This is the simplicial unit: the degeneracy \(s^0\) in the simplicial space \([n] \mapsto A(\mathbf{n})\) supplies the basepoint, confirming that \(A(\mathbf{0}) \simeq *\) encodes the unit element for the H-space structure on \(A(\mathbf{1})\). \(\square\)

Proposition 1.5 (Segal). Let \([n] \mapsto A_n\) be a simplicial space such that: 1. \(A_0\) is contractible, 2. \(p_n = \prod_{k=1}^{n} i_k^*: A_n \to A_1 \times \cdots \times A_1\) is a homotopy equivalence, where \(i_k: [1] \to [n]\) is \(i_k(0) = k-1\), \(i_k(1) = k\).

Then: (a) if \(A_1\) is \(k\)-connected, \(|A|\) is \((k+1)\)-connected; and (b) \(A_1 \to \Omega|A|\) is a homotopy equivalence if and only if \(A_1\) has a homotopy inverse.

This proposition is the engine behind the delooping machine.

3. The Classifying-Space Construction and Spectra 🔑

3.1 The Delooping Machine

Given a Γ-space \(A\), Segal defines its classifying-space \(BA\) to be the Γ-space such that, for any finite set \(S\), \[(BA)(S) = |T \mapsto A(S \times T)|,\] i.e., \((BA)(S)\) is the realization of the Γ-space \(T \mapsto A(S \times T)\).

The validation that \(BA\) is again a Γ-space rests on the homotopy equivalence \(A(\mathbf{n} \times \mathbf{m}) \simeq A(\mathbf{m})^n\), which follows from the Segal condition applied twice.

The spectrum. If \(A\) is a Γ-space, the sequence of spaces \[A(\mathbf{1}), \quad BA(\mathbf{1}), \quad B^2A(\mathbf{1}), \quad \ldots\] forms a spectrum, denoted \(\mathbf{B}A\). The structure maps arise as follows: the realization \(|A|\) contains a canonical subspace (its 1-skeleton) homotopy equivalent to \(\Sigma A(\mathbf{1})\), giving (up to homotopy) a map \[\Sigma A(\mathbf{1}) \longrightarrow |A| = BA(\mathbf{1}).\] Adjointly, this is a map \(A(\mathbf{1}) \to \Omega BA(\mathbf{1})\).

3.2 Proposition 1.4 and Its Significance

Proposition 1.4 (Segal). If \(A\) is a Γ-space and \(A(\mathbf{1})\) is \(k\)-connected, then \(BA(\mathbf{1})\) is \((k+1)\)-connected. Furthermore, \(A(\mathbf{1}) \simeq \Omega BA(\mathbf{1})\) if and only if the H-space \(A(\mathbf{1})\) has a homotopy inverse.

Proof sketch. The filtration of \(|A|\) gives \[|A|^{(p)}/|A|^{(p-1)} \simeq \Sigma^p(A(\mathbf{1}) \wedge \cdots \wedge A(\mathbf{1}))\] (p-fold smash). Since \(A(\mathbf{1})\) is \(k\)-connected, each smash is \((pk+p-1)\)-connected, so \(|A|\) is \((k+1)\)-connected by an inductive connectivity argument. The loop-space identification uses the simplicial path space \(PA\) and a homotopy-Cartesian square: \[\begin{array}{ccc} A(\mathbf{1}) & \to & |PA| \simeq * \\ \downarrow & & \downarrow \\ * & \to & |A| \end{array}\] which is homotopy-Cartesian if and only if the composition law (arising from the Segal structure) has a homotopy inverse. \(\square\)

Corollary. For a very special Γ-space \(A\), the adjunction map \(A(\mathbf{1}) \xrightarrow{\sim} \Omega BA(\mathbf{1})\) is a homotopy equivalence. Iterating: \(B^k A(\mathbf{1}) \simeq \Omega B^{k+1} A(\mathbf{1})\) for all \(k \geq 0\), so the spectrum \(\mathbf{B}A\) is an \(\Omega\)-spectrum (connective). This is the fundamental output of Segal’s machine: a connective \(\Omega\)-spectrum from any very special Γ-space.

[!TIP]- Solution to Exercise 6 (a) \(A(\mathbf{n}) = \mathbb{Z}^n\) is discrete, and the Segal map \(\varphi_n\) is the identity homeomorphism (by Exercise 3(b)), so \(A\) is special. The monoid \(\pi_0(A(\mathbf{1})) = \mathbb{Z}\) is a group under addition, so \(A\) is very special. ✓

(b) \(A(\mathbf{1}) = \mathbb{Z}\) is discrete. Its classifying space \(BA(\mathbf{1})\) is the classifying space of \(\mathbb{Z}\) viewed as a discrete group. Since \(\pi_1(B\mathbb{Z}) = \mathbb{Z}\) and \(\pi_k(B\mathbb{Z}) = 0\) for \(k \neq 1\) (a discrete group has no higher homotopy), \(B\mathbb{Z} = K(\mathbb{Z},1)\). Recall \(S^1 = K(\mathbb{Z},1)\) (the circle has \(\pi_1 = \mathbb{Z}\) and higher homotopy groups trivial after the universal cover). Hence \(BA(\mathbf{1}) \simeq S^1\).

(c) By Proposition 1.4, since \(A\) is very special and \(A(\mathbf{1}) = \mathbb{Z}\) is discrete (hence \((-1)\)-connected), \(BA(\mathbf{1}) \simeq S^1\) is \(0\)-connected, and iterating: \(B^k A(\mathbf{1}) \simeq K(\mathbb{Z},k)\) for all \(k \geq 0\). Concretely: \(B^2 A(\mathbf{1}) = B(S^1) = BS^1\). The classifying space of \(S^1\) as a topological group is \(\mathbb{CP}^\infty = K(\mathbb{Z},2)\). By induction and the loop-space identification \(B^k A(\mathbf{1}) \simeq \Omega B^{k+1} A(\mathbf{1})\), we get \(B^n A(\mathbf{1}) \simeq K(\mathbb{Z},n)\) for all \(n\).

(d) The spectrum \(\mathbf{B}A = (B^n A(\mathbf{1}))_{n \geq 0} = (K(\mathbb{Z},n))_{n \geq 0}\) is by definition the Eilenberg–Mac Lane spectrum \(H\mathbb{Z}\). The cohomology theory it represents is:

\[[X, H\mathbb{Z}]^n = [X, K(\mathbb{Z},n)] = H^n(X;\mathbb{Z}),\]

ordinary singular cohomology with \(\mathbb{Z}\) coefficients. \(\square\)

4. Γ-Categories from Symmetric Monoidal Categories 📐

This is where Segal connects the abstract Γ-space machine to concrete algebraic input: symmetric monoidal categories.

4.1 Definition of a Γ-Category

Definition (Γ-Category). A Γ-category is a contravariant functor \(\mathcal{C}: \Gamma \to \mathbf{Cat}\) (from \(\Gamma\) to the category of small categories) such that: 1. \(\mathcal{C}(\mathbf{0})\) is equivalent to the terminal category (one object, one morphism); 2. for each \(n\), the functor \[p_n = (i_1^*, \ldots, i_n^*): \mathcal{C}(\mathbf{n}) \longrightarrow \mathcal{C}(\mathbf{1}) \times \cdots \times \mathcal{C}(\mathbf{1})\] is an equivalence of categories.

Corollary 2.2. If \(\mathcal{C}\) is a Γ-category, then \(|\mathcal{C}|: S \mapsto |\mathcal{C}(S)|\) (taking nerve-realization) is a Γ-space.

4.2 Construction from Sums

Let \((\mathcal{C}, \oplus, 0)\) be a category in which coproducts (sums) exist. For a finite set \(S\), let \(\mathcal{P}(S)\) denote the category of subsets of \(S\) and inclusions.

Definition. \(\hat{\mathcal{C}}(S)\) is the category whose objects are functors \(F: \mathcal{P}(S) \to \mathcal{C}\) that take disjoint unions to sums, and whose morphisms are natural isomorphisms of such functors.

Concretely, an object of \(\hat{\mathcal{C}}(\mathbf{2})\) is a diagram \(A_1 \to A_{12} \leftarrow A_2\) in \(\mathcal{C}\) that expresses \(A_{12}\) as a coproduct \(A_1 \oplus A_2\). An object of \(\hat{\mathcal{C}}(\mathbf{n})\) is an assignment \(T \mapsto A_T\) for each subset \(T \subseteq \{1,\ldots,n\}\) such that \(A_{T \cup T'} \cong A_T \oplus A_{T'}\) whenever \(T \cap T' = \emptyset\).

Verification. The functor \(\hat{\mathcal{C}}(\mathbf{n}) \xrightarrow{p_n} \hat{\mathcal{C}}(\mathbf{1})^n\), which forgets to the single-element values \((A_{\{1\}}, \ldots, A_{\{n\}})\), is an equivalence of categories — the equivalence inverse reconstructs the entire diagram from its single-element values by choosing sums. Thus \(S \mapsto \hat{\mathcal{C}}(S)\) is a Γ-category.

[!TIP]- Solution to Exercise 7 (a) An object of \(\hat{\mathcal{C}}(\mathbf{2})\) is an assignment of a vector space \(V_T\) to each subset \(T \subseteq \{1,2\}\) such that for disjoint \(T, T'\) the canonical map \(V_T \oplus V_{T'} \xrightarrow{\sim} V_{T \cup T'}\) is an isomorphism. Concretely this is a diagram

\[V_1 \xrightarrow{i_1} V_{12} \xleftarrow{i_2} V_2\]

where \(i_j\) are linear inclusions with projections \(p_j: V_{12} \to V_j\) satisfying \(p_j \circ i_j = \mathrm{id}_{V_j}\) and \(i_1 p_1 + i_2 p_2 = \mathrm{id}_{V_{12}}\) — i.e., \(V_{12} \cong V_1 \oplus V_2\) via the canonical splitting. Morphisms are triples \((f_1, f_{12}, f_2)\) of linear maps making the evident squares commute.

(b) \(\lvert\hat{\mathcal{C}}(\mathbf{1})\rvert\): objects are finite-dimensional real vector spaces; two objects are isomorphic iff they have the same dimension, so \(\pi_0 = \mathbb{N}\) indexed by \(n = \dim V\). The automorphism group of \(\mathbb{R}^n\) is \(GL_n(\mathbb{R})\), and there are no morphisms between spaces of different dimensions. Realizing the nerve of this groupoid gives a classifying space for each isomorphism class:

\[\lvert\hat{\mathcal{C}}(\mathbf{1})\rvert \simeq \bigsqcup_{n \geq 0} BGL_n(\mathbb{R}).\]

(c) \(\pi_0\lvert\hat{\mathcal{C}}(\mathbf{1})\rvert = \mathbb{N}\) under the monoid operation \([m] + [n] = [m+n]\) (direct sum adds dimensions). This is not a group: there is no \(k \in \mathbb{N}\) with \(n + k = 0\) for \(n \geq 1\). The Grothendieck group completion formally adds inverses: \(K_0(\mathbf{FDVect}_\mathbb{R}) = \mathbb{N}^{\mathrm{gp}} \cong \mathbb{Z}\), generated by \([\mathbb{R}^1]\), with every class of the form \([\mathbb{R}^m] - [\mathbb{R}^n] = m - n \in \mathbb{Z}\). \(\square\)

4.3 Key Examples

[!EXAMPLE]- The symmetric groups example in detail Segal’s “most fundamental” Γ-space \(\mathbf{B}\Sigma\) arises from \(\Sigma\), the category of finite sets and bijections under disjoint union. Choosing a skeleton with one object \(\mathbf{n}\) for each \(n \geq 0\) (the set \(\{1,\ldots,n\}\)), one finds: \[|\hat{\Sigma}(\mathbf{1})| = \bigsqcup_{n \geq 0} B\Sigma_n.\] More explicitly, \(|\hat{\Sigma}(\mathbf{k})| = \bigsqcup_{m_1,\ldots,m_k \geq 0} \prod_{i=1}^k E\Sigma_{m_i} / \prod_{i=1}^k \Sigma_{m_i}\) with \(m = \sum_i m_i\) summed appropriately. The Segal condition holds because disjoint union makes the forgetful functor \(\hat{\Sigma}(\mathbf{n}) \to \hat{\Sigma}(\mathbf{1})^n\) an equivalence.

5. Γ-Spaces and Spectra: The Adjunction 🔑

5.1 The Spectrum Associated to a Γ-Space

A spectrum in Segal’s paper is a sequence of based spaces \(X = \{X_0, X_1, \ldots\}\) with closed embeddings \(X_k \hookrightarrow \Omega X_{k+1}\). The loop spectrum \(\omega X\) has \((\omega X)_k = \bigcup_{i \geq 0} \Omega^i X_{k+i}\).

Given a Γ-space \(A\), the spectrum \(\mathbf{B}A\) is \((B^k A(\mathbf{1}))_{k \geq 0}\) as constructed in §3.

5.2 The Γ-Space Associated to a Spectrum

Given a spectrum \(X\), observe that if \(P\) is a based space, the assignment \(S \mapsto P^S\) (the \(|S|\)-fold power) is naturally a covariant functor \(\Gamma \to \mathbf{Top}\).

Definition 3.1. The Γ-space \(AX\) associated to a spectrum \(X\) is \[(AX)(\mathbf{n}) = \mathrm{Mor}(\mathbf{S}^{\times n}; X),\] where \(\mathbf{S}\) denotes the sphere spectrum and \(\mathrm{Mor}\) means spectrum morphisms. Equivalently, \((AX)(\mathbf{n}) \simeq \mathrm{Mor}(\mathbf{S}; X)^n\) (using the Segal condition check: \(\mathrm{Mor}(\mathbf{S}^{\vee n}; X) \cong \mathrm{Mor}(\mathbf{S};X)^n\)).

Verification: The Segal condition holds because \(\mathbf{S}^{\times n} \simeq \mathbf{S}^{\vee n}\) for spectrum maps, giving \((AX)(\mathbf{n}) \simeq (AX)(\mathbf{1})^n\).

5.3 Adjointness and the Main Equivalence

Proposition 3.3 (Segal). The functors \(B\) and \(A\) form an adjoint pair: \[B: \mathcal{M} \rightleftharpoons \mathcal{S}p : A,\] where \(\mathcal{M}\) is the category of Γ-spaces (with homotopy-classes of weak morphisms) and \(\mathcal{S}p\) is the homotopy category of spectra.

The unit and counit are: - \(A \to A(BA)\): for each \(A\) a map of Γ-spaces; - \(B(AX) \to X\): for each spectrum \(X\) a map of spectra, given by evaluation.

Proposition 3.4 (Segal). (a) \(B\) sends Γ-spaces to connective spectra (\(\pi_p(B^q A) = 0\) for \(p < q\)), and \(AX(\mathbf{1})\) is always grouplike. (b) \(A \to A(BA)\) is an isomorphism in \(\mathcal{M}\) iff \(A(\mathbf{1})\) has a homotopy inverse. (c) \(B(AX) \to X\) is an isomorphism in \(\mathcal{S}p\) iff \(X\) is connective.

This yields the fundamental equivalence: the functors \(A\) and \(B\) restrict to an equivalence of categories between very special Γ-spaces and connective spectra.

6. The Barratt–Priddy–Quillen Theorem 🔑

6.1 The Γ-Space BC

Let \(\Sigma\) denote the category of finite sets and bijections, with symmetric monoidal structure given by disjoint union. The resulting Γ-space \(\mathbf{B}\Sigma\) satisfies: \[\mathbf{B}\Sigma(\mathbf{1}) = \bigsqcup_{n \geq 0} B\Sigma_n,\] where \(\Sigma_n\) is the \(n\)th symmetric group and \(B\Sigma_n\) its classifying space.

6.2 Proof Sketch via Adjointness

Proposition 3.5 (Barratt–Priddy–Quillen, Segal’s proof). The spectrum \(B(\mathbf{B}\Sigma)\) is equivalent to the sphere spectrum \(\mathbb{S}\).

Proof sketch. By the adjunction of Proposition 3.3, it suffices to show \[\mathrm{Hom}_{\mathcal{M}}(\mathbf{B}\Sigma, A) \cong \pi_0(A(\mathbf{1}))\] naturally for any Γ-space \(A\). Since \(\pi_0(A(\mathbf{1})) = \pi_0(AX(\mathbf{1}))\) is the zeroth homotopy group of the \(\Omega\)-spectrum \(AX\), one gets \(\mathrm{Hom}(B(\mathbf{B}\Sigma), X) \cong \pi_0(X)\) for spectra \(X\) — but this is precisely the defining property of the sphere spectrum \(\mathbb{S}\).

To construct the bijection: given a Γ-space \(A\) and a basepoint \(a \in A(\mathbf{1})\), define \(F_n\) as the homotopy-theoretic fibre of \(\varphi_n: A(\mathbf{n}) \to A(\mathbf{1})^n\) over \((a,\ldots,a)\). Then \(n \mapsto F_n\) is a contravariant functor on finite sets and injections. Form the category \(\mathcal{Y}_{F_a}\) of pairs \((n, x \in F_n)\) and construct its associated Γ-space \(\mathbf{B}\Sigma_{F_a}\). The forgetful map \(\mathbf{B}\Sigma_a \to \mathbf{B}\Sigma\) is an isomorphism in \(\mathcal{M}\), giving the desired map \(\mathbf{B}\Sigma \to A\) in \(\mathcal{M}\). Naturality in \(a\) and the component of \(a\) in \(\pi_0 A(\mathbf{1})\) establishes the bijection. \(\square\)

6.3 Stable Cohomotopy as K-Theory

Proposition 3.6 (Segal). More generally, if \(\mathbf{B}\Sigma_X\) is the Γ-space with \(\mathbf{B}\Sigma_X(\mathbf{1}) = \bigsqcup_{n \geq 0} (E\Sigma_n \times_{\Sigma_n} X^n)\), then \(B(\mathbf{B}\Sigma_X) \simeq \Sigma^\infty X_+\), the suspension spectrum of \(X\) with a disjoint basepoint.

The Barratt–Priddy–Quillen theorem is the case \(X = \mathrm{pt}\).

[!TIP]- Solution to Exercise 8 (a) The objects of \((\Sigma, \sqcup)\) up to isomorphism are the finite sets \(\{1,\ldots,n\}\) for \(n \geq 0\), one in each cardinality. Two finite sets are isomorphic iff they have the same cardinality, so \(\pi_0\lvert\hat{\Sigma}(\mathbf{1})\rvert = \{[0],[1],[2],\ldots\} \cong \mathbb{N}\). The monoidal product sends \([m] + [n] = [m+n]\) (disjoint union adds cardinalities), so the monoid structure is exactly \((\mathbb{N}, +, 0)\).

(b) The Grothendieck group of \((\mathbb{N},+)\) is constructed by formally adding inverses: \(\mathbb{N}^{\mathrm{gp}} = (\mathbb{N} \times \mathbb{N})/{\sim}\) where \((a,b) \sim (a',b')\) iff \(a + b' = a' + b\). Write \([a,b]\) for the class of \((a,b)\); then \([a,b]\) represents “\(a - b\)”. The map \(\mathbb{N} \to \mathbb{Z}\), \(n \mapsto n\) is the universal group homomorphism from \((\mathbb{N},+)\), so \(K_0(\Sigma,\sqcup) = \mathbb{N}^{\mathrm{gp}} \cong \mathbb{Z}\), generated by \([1] - [0] = 1 \in \mathbb{Z}\).

(c) The stable \(0\)-stem is \(\pi_0^s = \pi_0(\mathbb{S}) \cong \mathbb{Z}\), generated by the identity map \(\mathrm{id}: S^0 \to S^0\). Since \(B(\mathbf{B}\Sigma) \simeq \mathbb{S}\), we have \[\pi_0(\mathbb{S}) = \pi_0\!\left(B(\mathbf{B}\Sigma)\right) = \pi_0\!\left(\Omega\, B_1(\mathbf{B}\Sigma)\right),\] where \(B_1(\mathbf{B}\Sigma)\) denotes the first delooping in the spectrum. The group completion of \(\pi_0\lvert\hat{\Sigma}(\mathbf{1})\rvert = \mathbb{N}\) is \(\mathbb{Z}\), consistent with \(\pi_0(\mathbb{S}) = \mathbb{Z}\). The BPQ theorem promotes this \(\pi_0\)-level computation to a full equivalence of spectra.

(d) For Proposition 3.6: \(\mathbf{B}\Sigma_X(\mathbf{1}) = \bigsqcup_{n \geq 0} E\Sigma_n \times_{\Sigma_n} X^n\), so \[\pi_0\!\left(\mathbf{B}\Sigma_X(\mathbf{1})\right) = \bigsqcup_{n \geq 0} \pi_0(X^n/\Sigma_n) = \bigsqcup_{n \geq 0} \mathrm{Sym}^n(\pi_0 X),\] the free commutative monoid on \(\pi_0 X\). The group completion is the free abelian group \(\mathbb{Z}[\pi_0 X]\). This is consistent with \(\pi_0(\Sigma^\infty X_+) \cong \mathbb{Z}[\pi_0 X]\), which follows from the stable Hurewicz theorem: the \(0\)th stable homotopy group of \(X_+\) is the free abelian group on the connected components of \(X\).

7. Group Completion and the Grothendieck Construction 📐

7.1 The Group Completion Problem

In practice, one starts with a symmetric monoidal category \((\mathcal{C}, \oplus, 0)\) and forms the Γ-space \(A = |\hat{\mathcal{C}}|\). The zeroth space \(A(\mathbf{1}) = |\mathcal{C}|\) is typically only a monoid up to homotopy (not a group). The spectrum \(\mathbf{B}A\) is connective but the map \(A(\mathbf{1}) \to \Omega B A(\mathbf{1})\) may not be an equivalence.

The K-theory \(k\mathcal{C}\) is the cohomology theory represented by \(\mathbf{B}A\); the zeroth space \(\Omega B_1 = \Omega B A(\mathbf{1})\) is the group completion of the monoid \(\pi_0 A(\mathbf{1}) = \pi_0 |\mathcal{C}|\).

Proposition 4.1 (Segal). If \(|\mathcal{C}|\) has the homotopy type of a CW-complex and \(\pi_0|\mathcal{C}|\) contains a cofinal free abelian monoid, then the natural transformation \[[-; |\mathcal{C}|] \longrightarrow k\mathcal{C}^0(-)\] is universal among transformations from \([-; |\mathcal{C}|]\) to representable abelian-group-valued homotopy functors.

This is Quillen’s group-completion theorem in Segal’s language.

7.2 The Space A’

To construct the group completion explicitly at the Γ-space level, Segal introduces a Γ-space \(A'\) from \(A\) with the following properties: 1. \(\pi_0(A'(\mathbf{1}))\) is the group completion (Grothendieck group) of \(\pi_0(A(\mathbf{1}))\); 2. \(BA \to BA'\) is a weak equivalence of spectra.

The construction uses the simplicial path space \(P: \Delta \to \Delta\), \([k] \mapsto [k+1]\). One defines \[A'(\mathbf{m}) = \left| [k] \mapsto \underbrace{PA([k], \mathbf{m}) \times_{A([k], \mathbf{m})} \cdots}_{} \right|\] as a homotopy-theoretic fibre product, which has the effect of “symmetrizing” the path-space to add inverses.

7.3 Quillen’s Plus-Construction and K-Theory

In the fundamental example \(A(\mathbf{1}) = \bigsqcup_{n \geq 0} B\Sigma_n\), Segal’s construction gives: \[T_{A,\mu} \simeq \mathbb{Z} \times B\Sigma_\infty^+,\] where \(B\Sigma_\infty^+ = B(\varinjlim \Sigma_n)^+\) is Quillen’s plus-construction on the classifying space of the infinite symmetric group. For a ring \(R\) and the category of finitely generated projective \(R\)-modules: \[T_{A,\mu} \simeq K_0(R) \times BGL(R)^+.\]

Thus Segal’s \(A'\) construction recovers Quillen’s algebraic K-theory groups \(K_n(R) = \pi_n(BGL(R)^+)\) for all \(n \geq 1\) as the homotopy groups of a single connective spectrum.

8. Ring Spectra 📐

When a category carries two compatible composition laws (one distributive over the other — analogous to a ring), the associated spectrum inherits a ring structure.

Definition 5.1 (Multiplication on a Γ-Space). A multiplication on a Γ-space \(A\) is a contravariant functor \(\tilde{A}: \Gamma \times \Gamma \to \mathbf{Top}\) together with natural transformations \[i_1: \tilde{A}(S,T) \to A(S), \quad i_2: \tilde{A}(S,T) \to A(T), \quad m: \tilde{A}(S,T) \to A(S \times T),\] such that \((i_1 \times i_2): \tilde{A}(S,T) \to A(S) \times A(T)\) is a homotopy equivalence for all \(S, T\).

A multiplication on \(A\) determines a pairing of spectra \(\mathbf{B}A \wedge \mathbf{B}A \to \mathbf{B}A\), making \(\mathbf{B}A\) a ring spectrum.

For strongly homotopy-associative and commutative ring spectra, one needs a sequence \(A_1, A_2, \ldots\) where \(A_1 = A\), \(A_2\) is a multiplication on \(A_1\), \(A_3\) is a “multiplication on \(A_2\)”, and so on. Segal indicates this leads to \(E_\infty\) ring spectra and promises to return to it elsewhere.

[!TIP]- Solution to Exercise 10 (a) The two symmetric monoidal structures on \(\mathbf{FinSet}\) are \((\sqcup, \emptyset)\) (disjoint union, unit the empty set) and \((\times, \{*\})\) (Cartesian product, unit a one-element set). Distributivity is the standard set-theoretic identity: each element of \(X \times (Y \sqcup Z)\) is a pair \((x, y_{\sqcup z})\) lying in either \(X \times Y\) or \(X \times Z\) (depending on which summand the second component comes from), giving a canonical bijection \(X \times (Y \sqcup Z) \xrightarrow{\sim} (X \times Y) \sqcup (X \times Z)\). The absorbing element: \(X \times \emptyset = \emptyset\) for all \(X\), since there are no pairs \((x, -)\) with second component in \(\emptyset\).

(b) The distributivity \(X \times (Y \sqcup Z) \cong (X \times Y) \sqcup (X \times Z)\) gives a Γ-morphism pairing. On the Γ-space level: the Cartesian product provides a functor \(\tilde{A}: \Gamma \times \Gamma \to \mathbf{Top}\) with \(\tilde{A}(S,T) = \mathbf{B}\Sigma(S \times T)\). The natural transformations \(i_1, i_2, m\) required by Definition 5.1 come from the projections and the product map on finite sets. On the spectrum level, this gives the ring spectrum pairing \(\mathbb{S} \wedge \mathbb{S} \to \mathbb{S}\) — the multiplication on \(\mathbb{S} = B(\mathbf{B}\Sigma)\).

(c) A rig category \(\mathcal{R}\) has two symmetric monoidal structures \((\oplus, 0_\mathcal{R})\) and \((\otimes, 1_\mathcal{R})\) with distributivity \(A \otimes (B \oplus C) \cong (A \otimes B) \oplus (A \otimes C)\) and \(A \otimes 0_\mathcal{R} \cong 0_\mathcal{R}\) (coherently). Define \(F: \mathbf{FinSet} \to \mathcal{R}\) by \(F(\{1,\ldots,n\}) = 1_\mathcal{R}^{\oplus n}\) (the \(n\)-fold \(\oplus\) of the monoidal unit), with \(F(\emptyset) = 0_\mathcal{R}\). This is the unique rig functor, since \(\mathbf{FinSet}\) is freely generated as a rig category by a single object \(\{*\}\): every finite set \(\{1,\ldots,n\}\) is \(\{*\}^{\sqcup n}\), so \(F\) is forced. Any two rig functors from \(\mathbf{FinSet}\) agreeing on \(\{*\}\) must agree everywhere by the free property.

(d) The Γ-category construction is functorial: a rig functor \(F: \mathbf{FinSet} \to \mathcal{R}\) induces a map of Γ-spaces \(\mathbf{B}\Sigma \to A_\mathcal{R}\) and hence a map of ring spectra \(\mathbb{S} = B(\mathbf{B}\Sigma) \to B(A_\mathcal{R})\). By part (c) this map is unique (up to coherent isomorphism). Hence for any ring spectrum \(E\), there is a unique ring map \(\mathbb{S} \to E\) — \(\mathbb{S}\) is the initial ring spectrum. \(\square\)

9. Relationship with Operads (Boardman–Vogt–May) 📐

In Appendix B, Segal relates Γ-spaces to May’s operad actions on spaces, establishing that the two frameworks are equivalent for the purpose of delooping.

Definition (Category of Operators). A category of operators is a topological category \(K\) whose object space is discrete. A \(K\)-diagram is a continuous contravariant functor \(A: K \to \mathbf{Top}\). An operad in the sense of concepts/category-theory/01-categories-functors-natural-transformations|May furnishes an example with object set \(\mathbb{N}\) and a map of categories of operators \(\pi: K \to \Gamma\).

Proposition B.1. If \(\pi: K \to M\) is an equivalence of categories of operators, then \(A \to \pi^* \pi_* A\) and \(\pi_* \pi^* B \to B\) are equivalences for any \(K\)-diagram \(A\) and \(M\)-diagram \(B\).

Proposition B.2. For any category of operators \(K\) there is an equivalence \(\hat{\pi}: \hat{K} \to K\) (the “explosion” of \(K\)) such that any levelwise homotopy equivalence \(A(S) \simeq A'(S)\) can be lifted to an actual isomorphism of \(\hat{K}\)-diagrams. The explosion \(\hat{K}\) is the category of paths in \(K\).

These two propositions establish: (a) any operad action on \(X\) (in May’s sense) gives a Γ-space with \(A(\mathbf{n}) = X^n\); (b) conversely, any Γ-space with the Segal condition gives an operad action after passing through the explosion. The two frameworks — Segal’s Γ-spaces and May’s \(E_\infty\) operads — produce equivalent delooping machines.

10. Realization of Simplicial Spaces 📐

Appendix A addresses a technical issue: the naive geometric realization \(|A| = \int^{[n] \in \Delta} \Delta^n \times A_n\) can behave poorly (not preserve homotopy equivalences levelwise, exit the CW category). Segal introduces two improved realizations:

1. \(\|A\|\) (the thick realization): attach using only injective face maps, then collapse degenerate parts. Gives \(\|A\|_n = |A|^{(n)}/|A|^{(n-1)} \cong \Delta^n \times A_n / \partial \Delta^n \times A_n\) at each filtration level.

2. \(|\tau A|\) (thickened realization): first thicken \(A_n\) to \(\tau_n A = \bigcup_{G \subseteq \{1,\ldots,n\}} [0,1]^G \times A_{n,G}\) (a generalized mapping cylinder of the degeneracy inclusions), then take naive realization.

Proposition A.2. The functor \(A \mapsto |\tau A|\) satisfies: 1. If each \(A_n\) is a CW-complex, so is \(|\tau A|\); 2. If \(A_n \xrightarrow{\sim} A_n'\) for each \(n\), then \(|\tau A| \xrightarrow{\sim} |\tau A'|\); 3. \(|\tau(A \times A')| \simeq |\tau A| \times |\tau A'|\); 4. \(|\tau A| \simeq |A|\) whenever \(A\) is good (each degeneracy \(s_i: A_{n-1} \hookrightarrow A_n\) is a closed cofibration).

A simplicial space is good if all degeneracy inclusions are cofibrations. Realizations \(\|A\|\), \(|\tau A|\), and \(|\mathrm{simp}(A)|\) (the classifying space of the category of simplexes) all give the same homotopy type, with \(|\mathrm{simp}(A)| \simeq |\tau A|\).

References