Solutions to Basic Category Theory Tom Leinster Solutions by positrón0802 https://positron0802.wordpress.com 1 January 2021 Contents Contents 0 Introduction 2 1 Categories, functors and natural transformations 1.1 Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Natural Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 4 4 10 2 Adjoints 2.1 Definition and examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Adjunctions via units and counits . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Adjunctions via initial objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 16 22 27 3 Interlude on sets 3.1 Constructions with sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Small and large categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Historical remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 31 33 36 4 Representables 4.1 Definitions and examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 The Yoneda lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 Consequences of the Yoneda lemma . . . . . . . . . . . . . . . . . . . . . . . . . . 36 36 39 40 5 Limits 5.1 Limits: definition and examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Colimits: definition and example . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 Interactions between functors and limits . . . . . . . . . . . . . . . . . . . . . . . 42 42 46 50 6 Adjoints, representables and limits 6.1 Limits in terms of representables and adjoints . . . . . . . . . . . . . . . . . . . . 6.2 Limits and colimits of presheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3 Interactions between adjoint functors and limits . . . . . . . . . . . . . . . . . . . 53 53 53 60 Appendix: Proof of the general adjoint functor theorem 66 1 Solutions by positrón0802 0. Introduction 0 Introduction Exercise 0.10. Let ๐ : ๐ → ๐ผ (๐) to the identity function of sets. Then the topological space ๐ผ (๐) together with the function ๐ satisfy the universal property illustrated in the following diagram: ๐ ๐ผ (๐) ๐ ∃! continuous ๐ ∀ functions ๐ ∀๐ . Explicitly, given any topological space ๐ and any function ๐ : ๐ → ๐, there exists a unique (continuous) map ๐ : ๐ → ๐ผ (๐) such that ๐ = ๐ โฆ ๐ . Indeed, there is only one way to define ๐ , namely to be given by ๐ (๐ฅ) = ๐ (๐ฅ) for all ๐ฅ ∈ ๐, and this function is continuous since ๐ผ (๐) has the indiscrete topology. Briefly, this property says that any function into an indiscrete space is continuous. Exercise 0.11. Let ๐ : ๐บ → ๐ป denote the trivial homomorphism. The pair (ker ๐, ๐) satisfies the following universal property: ๐โฆ๐ ker ๐ ∃!๐ 0 ๐ ๐ ๐บ ๐ป ∀๐ . ๐โฆ๐ ∀๐พ That is, we have ๐ โฆ๐ = ๐ โฆ๐, and if (๐พ, ๐) is any other pair of a group ๐พ and a morphism ๐ : ๐พ → ๐บ such that ๐ โฆ ๐ = ๐ โฆ ๐, then there exists a unique homomorphism ๐ 0 : ๐พ → ker ๐ such that ๐ โฆ ๐ 0 = ๐ . Indeed, such ๐ 0 must be defined by ๐ 0 (๐) = ๐ (๐) for all ๐ ∈ ๐พ, and this is well-defined since ๐ (๐) ∈ ker ๐ for all ๐ ∈ ๐พ . Exercise 0.12. We are given the following diagram: ๐ ๐ ∩๐ ๐ ๐0 ๐ ∀๐ ๐0 ๐ ๐ ∃!โ ∀๐ . ∀๐ Given a space ๐ and maps ๐ : ๐ → ๐ , ๐ : ๐ → ๐ such that ๐ โฆ ๐ = ๐ โฆ ๐, there exists a unique map โ : ๐ → ๐ such that the diagram commutes. The condition ๐โฆ ๐ = ๐ โฆ๐ is saying that ๐|๐ ∩๐ = ๐๐ ∩๐ , so there is a unique well-defined function โ : ๐ → ๐ making the diagram commute, namely the one given by โ(๐ฅ) = ๐(๐ฅ) if ๐ฅ ∈ ๐ and โ(๐ฅ) = ๐ (๐ฅ) if ๐ฅ ∈ ๐ , and this function โ is continuous by the gluing lemma as ๐ and ๐ are open in ๐ . 2 Solutions by positrón0802 0. Introduction Exercise 0.13. (a) Let ๐ be a ring and ๐ ∈ ๐ . Any ring homomorphism ๐ : Z[๐ฅ] → ๐ such that ๐ (๐ฅ) = ๐ must be given by ! ๐ ๐ ๐ Õ Õ Õ ๐ ๐๐ ๐ฅ ๐ = ๐ (๐๐ )๐ (๐ฅ)๐ = ๐๐ ๐ ๐ , ๐๐ ∈ Z, ๐=0 ๐=0 ๐=0 and it is clear that this is indeed a homomorphism. (b) Since ๐ด is a ring and ๐ ∈ ๐ด, it follows from part (a) that there is a unique ring homomorphism ๐ : Z[๐ฅ] → ๐ด such that ๐ (๐ฅ) = ๐. By assumption, there is also a unique ring homomorphism ๐ : ๐ด → Z[๐ฅ] such that ๐ (๐) = ๐ฅ . Then ๐ โฆ ๐ is a ring homomorphism Z[๐ฅ] → Z[๐ฅ] such that ๐ฅ โฆ→ ๐ฅ, so by uniqueness we have ๐ โฆ ๐ = 1Z[๐ฅ ] . Similarly ๐ โฆ ๐ = 1๐ด, so ๐ is an isomorphism. Exercise 0.14. (a) We seek for (๐, ๐ 1, ๐ 2 ) with the universal property illustrated below: ∀๐1 ∀๐ ∃!๐ ๐1 ๐ ∀๐2 ๐ ๐2 . ๐ Consider ๐ = ๐ ⊕ ๐ with ๐ 1 : ๐ → ๐ and ๐ 2 : ๐ → ๐ the canonical projections. Then, given a cone (๐ , ๐1, ๐2 ), there is only one way to define ๐ : ๐ → ๐ so that ๐ 1 โฆ ๐ = ๐1 and ๐ 2 โฆ ๐ = ๐2, namely to be given by ๐ (๐ฃ) = (๐1 (๐ฃ), ๐2 (๐ฃ)), and this is linear since both ๐1 and ๐2 are linear. (b) Let (๐, ๐ 1, ๐ 2 ) and (๐ 0, ๐ 10 , ๐ 20 ) be cones having the property stated in (a). Then, since 0 (๐ , ๐ 10 , ๐ 20 ) has this property, there is a unique linear map ๐ : ๐ → ๐ 0 such that ๐ 10 โฆ ๐ = ๐ 1 and ๐ 20 โฆ ๐ = ๐ 2 . Similarly there exists a unique linear map ๐ : ๐ 0 → ๐ satisfying ๐ 1 โฆ ๐ = ๐ 10 and ๐ 2 โฆ ๐ = ๐ 20 , and the equations ๐ โฆ ๐ = 1๐ 0 and ๐ โฆ ๐ = 1๐ hold by uniqueness. (c) We seek for (๐, ๐ 1, ๐ 2 ) with the universal property illustrated below: ∀๐1 ∀๐ ∃!๐ ๐ ∀๐2 ๐1 ๐ ๐2 ๐ . Consider ๐ = ๐ ⊕ ๐ with ๐ 1 : ๐ → ๐ and ๐ 2 : ๐ → ๐ the canonical inclusions, ๐ 1 (๐ฅ) = (๐ฅ, 0) for all ๐ฅ and ๐ 2 (๐ฆ) = (0, ๐ฆ) for all ๐ฆ. Then, given a cocone (๐ , ๐1, ๐2 ) there is only one way to define a map ๐ : ๐ → ๐ fitting in the diagram above, namely to be given by ๐ (๐ฅ, ๐ฆ) = ๐1 (๐ฅ) + ๐2 (๐ฆ), and this is linear since both ๐1 and ๐2 are linear. 3 Solutions by positrón0802 1. Categories, functors and natural transformations (d) Given another cocone (๐ 0, ๐ 10 , ๐ 20 ) satisfying the property stated in (c), there is a unique linear map ๐ : ๐ → ๐ 0 such that ๐ โฆ ๐ 1 = ๐ 10 and ๐ โฆ ๐ 2 = ๐ 20 , and similarly a unique linear map ๐ : ๐ 0 → ๐ such that ๐ โฆ ๐ 10 = ๐ 1 and ๐ โฆ ๐ 20 = ๐ 2, which is the inverse of ๐ by uniqueness. 1 1.1 Categories, functors and natural transformations Categories Exercise 1.1.12. The category of modules over a (fixed) ring ๐ ; the category of based topological spaces and based map; the category of smooth manifolds and smooth maps; the category of ringed spaces and morphisms of ringed spaces. Exercise 1.1.13. If ๐, ๐ 0 : ๐ต → ๐ด are both inverses of ๐ : ๐ด → ๐ต, then ๐ = ๐1๐ต = ๐๐ ๐ 0 = 1๐ด๐ 0 = ๐ 0 . Exercise 1.1.14. In the product category ๐×โฌ, identities are given by 1 (๐ด,๐ต) = (1๐ด, 1๐ต ) : (๐ด, ๐ต) → (๐ด, ๐ต), and composition of morphisms by (๐ , ๐) โฆ (๐ 0, ๐ 0) = (๐ ๐ 0, ๐๐ 0). Exercise 1.1.15. To prove that Toph is a category we need the following facts from topology: • Composition is well-defined: if ๐ , ๐ 0 : ๐ → ๐ are homotopic and ๐, ๐ 0 : ๐ → ๐ are homotopic then ๐๐ , ๐ 0 ๐ 0 : ๐ → ๐ are homotopic. • Existence of identities: given a topological space ๐ there exists a map ๐ : ๐ → ๐ such that for all maps ๐ : ๐ → ๐ , ๐ ๐ is homotopic to ๐ , and for all maps ๐ : ๐ → ๐, ๐๐ is homotopic to ๐. We can take ๐ = 1๐ . • Composition is associative: given maps ๐ : ๐ → ๐ , ๐ : ๐ → ๐ and โ : ๐ → ๐ , then โโฆ(๐โฆ๐ ) is homotopic to (โ โฆ ๐) โฆ ๐ . This is clear (they are equal). Two objects ๐, ๐ ∈ Toph are isomorphic if and only if they are homotopy equivalent, i.e. whenever there exist maps ๐ : ๐ → ๐ and ๐ : ๐ → ๐ such that ๐๐ and ๐ ๐ are homotopic to the respective identity maps. 1.2 Functors Exercise 1.2.20. • Generalising the functor ๐1, for all ๐ ≥ 2 there are functors ๐๐ : Top∗ → Ab assigning to a based space its ๐th homotopy group (which happens to be abelian). • The functor Grp → Ab sending a group to its abelianisation. • The functor Ab → Ab sending an abelian group to its torsion subgroup. • For ๐ = 0, 1, 2, . . . , the bifunctors Tor๐๐ (−, −), Ext๐๐ (−, −) : Mod๐ × ๐ Mod → Ab for a fixed ring ๐ . 4 Solutions by positrón0802 1.2 Functors Exercise 1.2.21. Let ๐ : ๐ด → ๐ด 0 be an isomorphism with inverse ๐ : ๐ด 0 → ๐ด. Then 1๐น (๐ด0) = ๐น (1๐ด0 ) = ๐น (๐ ๐) = ๐น (๐ )๐น (๐) and 1๐น (๐ด) = ๐น (1๐ด ) = ๐น (๐๐ ) = ๐น (๐)๐น (๐ ), so ๐น (๐ ) : ๐น (๐ด) → ๐น (๐ด 0) is an isomorphism with inverse ๐น (๐) : ๐น (๐ด 0) → ๐น (๐ด). Exercise 1.2.22. Given an order-preserving map ๐ : ๐ด → ๐ต we obtain a functor ๐น : ๐ → โฌ sending ๐ ∈ ๐ด to ๐น (๐) = ๐ (๐), and an arrow ๐ → ๐ 0 to ๐ (๐) → ๐ (๐ 0). This is well-defined since ๐ ≤ ๐ 0 in ๐ด implies ๐ (๐) ≤ ๐ (๐ 0) in ๐ต. Given arrows ๐ → ๐ 0, ๐ 0 → ๐ 00, their composition has to be the unique arrow ๐ → ๐ 00, and we have unique arrows ๐ (๐) → ๐ (๐ 0), ๐ (๐ 0) → ๐ (๐ 00) whose composition is the unique arrow ๐ (๐) → ๐ (๐ 00), so ๐น ((๐ → ๐ 0) โฆ (๐ 0 → ๐ 00)) = ๐น (๐ → ๐ 00) = ๐ (๐) → ๐ (๐ 00) = (๐ (๐) → ๐ (๐ 0)) โฆ (๐ (๐ 0) → ๐ (๐ 00)). Thus ๐น is a functor. Conversely, if ๐น : ๐ → โฌ is a functor then ๐ : ๐ด → ๐ต given by ๐ (๐) = ๐น (๐) is orderpreserving. Indeed, if ๐ ≤ ๐ 0 then there is an arrow ๐ → ๐ 0, which is sent by ๐น to an arrow ๐ (๐) → ๐ (๐ 0), which means that ๐ (๐) ≤ ๐ (๐ 0). ๐op โ op Exercise 1.2.23. (a) Given ๐, โ ∈ ๐บ and ∗ −−→ ∗ −−→ ∗ in ๐บ op, the composition โ op โฆop ๐op is the opposite arrow of ∗ −−→ ∗ in ๐บ, that is, โ op โฆop ๐op = (๐โ) op . Thus ๐บ op is the group with the same underlying set as ๐บ with multiplication : ๐บ op × ๐บ op → ๐บ op given by โ ๐ = ๐โ. Define ๐ : ๐บ → ๐บ op by ๐ (๐) = ๐−1 . Then ๐ is clearly bijective. Furthermore, ๐ is a group homomorphism: if ๐, โ ∈ ๐บ then ๐ (๐โ) = (๐โ) −1 = โ −1๐−1 = ๐−1 โ −1 = ๐ (๐) ๐ (โ). Thus ๐ is an isomorphism of groups ๐โ (b) Let ๐ be any set with |๐ | ≥ 2 and consider ๐ = End(๐ ). Consider a constant function ๐ ∈ ๐. Then ๐ satisfies ๐ ๐ = ๐ for all ๐ ∈ ๐. Since there is no element ๐ 0 in ๐ such that ๐ ๐ 0 = ๐ 0 for all ๐ ∈ ๐, it follows that ๐ is not isomorphic to ๐ op . Exercise 1.2.24. (This is also Mac Lane’s Exercise I.3.4.) We will show there is no such ๐ . Consider the symmetric groups ๐ 2 Z/2 and ๐ 3 . The inclusion ๐ : ๐ 2 → ๐ 3 and the projection ๐ : ๐ 3 → ๐ 3 /๐ด3 ๐ 2 (where ๐ด3 denotes the alternating group of degree 3) compose to the identity 1๐2 : ๐ 2 → ๐ 2 . As ๐ 3 has trivial centre, such functor ๐ : Grp → Ab would send 1๐ 2 to 1๐ 2 = 1๐ (๐2 ) = ๐ (1๐ 2 ) = ๐ (๐)๐ (๐) = 0, which is a contraction. Exercise 1.2.25. (a) Let ๐ด ∈ ๐. For all ๐ต ∈ โฌ we have ๐น ๐ด (1๐ต ) = ๐น (1๐ด, 1๐ต ) = 1๐น (๐ด,๐ต) , and if ๐ : ๐ต → ๐ต 0 and ๐ : ๐ต 0 → ๐ต 00 are morphisms in โฌ then ๐น ๐ด (๐๐ ) = ๐น (1๐ด, ๐๐ ) = ๐น ((1๐ด, ๐) โฆ (1๐ด, ๐ )) = ๐น (1๐ด, ๐) โฆ ๐น (1๐ด, ๐ ) = ๐น ๐ด (๐) โฆ ๐น ๐ด (๐ ). Thus ๐น ๐ด is a functor. Similarly, given ๐ต ∈ โฌ, ๐น๐ต : ๐ → ๐ defined by ๐น๐ต (๐ด) = ๐น (๐ด, ๐ต) on objects and ๐น๐ต (๐ ) = (๐ , 1๐ต ) on morphisms is a functor. 5 Solutions by positrón0802 1.2 Functors (b) The equation ๐น ๐ด (๐ต) = ๐น (๐ด, ๐ต) = ๐น๐ต (๐ด) is clear. Given ๐ : ๐ด → ๐ด 0 in ๐ and ๐ : ๐ต → ๐ต 0 in โฌ we have 0 ๐น ๐ด (๐) โฆ ๐น๐ต (๐ ) = ๐น (1๐ด0 , ๐) โฆ ๐น (๐ , 1๐ต ) = ๐น (๐ , ๐) = ๐น (๐ , 1๐ต0 ) โฆ ๐น (1๐ด, ๐) = ๐น๐ต0 (๐ ) โฆ ๐น ๐ด (๐) by functoriality of ๐น . (c) We must have ๐น (๐ด, ๐ต) = ๐น ๐ด (๐ต) = ๐น๐ต (๐ด) on objects. If ๐ : ๐ด → ๐ด 0 and ๐ : ๐ต → ๐ต 0 are morphisms in ๐ and โฌ respectively, then ๐น (๐ , ๐) must be given by ๐น (๐ , ๐) = ๐น (๐ , 1๐ต0 ) โฆ ๐น (1๐ด, ๐) = ๐น๐ต0 (๐ ) โฆ ๐น ๐ด (๐), so we define ๐น (๐ , ๐) by this formula. At the same time, ๐น (๐ , ๐) must be given by 0 ๐น (๐ , ๐) = ๐น (1๐ด0 , ๐) โฆ ๐น (๐ , 1๐ต ) = ๐น ๐ด (๐) โฆ ๐น๐ต (๐ ). By the conditions in (b), these two definitions agree. It remains to show that ๐น is indeed a functor when defined in this way. First note that ๐น (1๐ด, 1๐ต ) = ๐น๐ต (1๐ด ) โฆ ๐น ๐ด (1๐ต ) = 1๐น๐ต (๐ด) โฆ 1๐น ๐ด (๐ต) = 1๐น (๐ด,๐ต) for all ๐ด ∈ ๐ and ๐ต ∈ โฌ. If (๐ , ๐) : (๐ด, ๐ต) → (๐ด 0, ๐ต 0) and (๐ 0, ๐ 0) : (๐ด 0, ๐ต 0) → (๐ด 00, ๐ต 00) are morphisms in ๐ × โฌ then ๐น ((๐ 0, ๐ 0) โฆ (๐ , ๐)) = ๐น (๐ 0 ๐ , ๐ 0๐) = ๐น๐ต00 (๐ 0 ๐ ) โฆ ๐น ๐ด (๐ 0๐) = ๐น๐ต00 (๐ 0) โฆ ๐น๐ต00 (๐ ) โฆ ๐น ๐ด (๐ 0) โฆ ๐น ๐ด (๐) 0 = ๐น๐ต00 (๐ 0) โฆ ๐น ๐ด (๐ 0) โฆ ๐น๐ต0 (๐ ) โฆ ๐น ๐ด (๐) = ๐น (๐ 0, ๐ 0) โฆ ๐น (๐ , ๐). Hence ๐น : ๐ × โฌ → ๐ is indeed a functor, the unique one satisfying the conditions in (b). Exercise 1.2.26. (This is also Mac Lane’s Exercise II.3.5.) Given a topological space ๐ let ๐ถ (๐ ) be the ring of continuous functions ๐ → R, and given a map ๐ : ๐ → ๐ of spaces let ๐ถ (๐ ) : ๐ถ (๐ ) → ๐ถ (๐ ) be defined by โ โฆ→ โ โฆ ๐ for all maps ๐ : ๐ → R. This is well-defined since composition of continuous functions is continuous. Moreover, this gives a contravariant functor from Top to Ring: given ๐ : ๐ → ๐ and ๐ : ๐ → ๐, we have ๐ถ (๐ ๐) = ๐ถ (๐)๐ถ (๐ ); furthermore, clearly ๐ถ (1๐ ) = 1๐ถ (๐ ) for all spaces ๐ . Exercise 1.2.27. Let ๐ be a category with precisely three objects ๐, ๐, ๐ and only two non-identity morphisms ๐ : ๐ → ๐, ๐ : ๐ → ๐, and โฌ be a category with exactly two objects ๐, ๐ and only one non-identity morphism โ : ๐ → ๐. Let ๐น : ๐ → โฌ be given by ๐น (๐) = ๐ and ๐น (๐) = ๐น (๐) = ๐ on objects, and by ๐น (๐ ) = ๐น (๐) = โ on morphisms. An easier example would be letting ๐ have two elements and only identity morphisms, โฌ have one element and only the identity, and letting ๐น : ๐ด → ๐ต be the only possible functor. Also, any non-injective function of sets ๐ → โฌ, considered as discrete categories, works. Exercise 1.2.28. (a) We list them here: 6 Solutions by positrón0802 1.2 Functors • Forgetful functors that forget “structure”, as Grp → Set, Ring → Set, Vect๐ → Set, Ring → Ab and Ring → Mon. All of these are faithful; for instance, given two groups ๐บ, ๐ป ∈ Grp, if ๐ : ๐บ → ๐ป is a function between the underlying sets then there is at most one group homomorphism ๐บ → ๐ป whose underlying set function is ๐ , namely ๐ itself if it is a homomorphism. Thus Grp → Set is faithful, and the same argument works for the others. However, none of these is full, as forgetting some structure gives rise to more morphisms. For instance, we can find two groups ๐บ, ๐ป and a function ๐บ → ๐ป between the underlying sets which is not a group homomorphism, so Grp → Set is not full, and the same argument works for the other functors. • The forgetful functor Ab → Grp (which forgets a “property”) is full and faithful. It is faithful by the same reason as in the previous example, but now it is also full since any group homomorphism between abelian groups is an abelian group homomorphism. • Free functors: Set → Grp, Set → CRing, Set → Vect๐ , etc. All of these satisfy the same universal property in their corresponding categories. For instance, if ๐ is a set then the free group ๐น (๐) on ๐ have the following universal property: if ๐ : ๐ → ๐น (๐) is the natural inclusion, ๐ป is a group and ๐ : ๐ → ๐ป a (set) function, there is a unique group homomorphism ๐ : ๐น (๐) → ๐ป such that ๐ ๐ = ๐ : ๐ ๐น (๐) ๐ ∃! homomorphism ๐ ∀ functions ๐ ∀๐ป Thus, if ๐ : Grp → Set denote the forgetful functor, for any group ๐ป we have a bijection Set(๐, ๐ (๐ป )) Grp(๐น (๐), ๐ป ). In particular, for two sets ๐,๐ we have a bijection Set(๐, ๐ ๐น (๐ )) Grp(๐น (๐), ๐น (๐ )). We have an inclusion ๐ ⊂ ๐ ๐น (๐ ) of sets, so if ๐ : ๐ → ๐ is a function inducing a homomorphism ๐น (๐) → ๐น (๐ ), then ๐ is unique for ๐ can be seen as a function ๐ → ๐ ๐น (๐ ). Thus the free functor Set → Grp is faithful, and the same argument works for Set → CRing and Set → Vect๐ . However neither of these is full. For instance, if a homomorphism ๐ : ๐น (๐) → ๐น (๐ ) corresponds via the bijection to a function ๐ → ๐ ๐น (๐ ) with image not in ๐ , then there is no function ๐ → ๐ inducing ๐. • The fundamental group functor ๐1 : Top∗ → Grp. This functor is not faithful since two different based maps ๐ → ๐ can induce the same morphism in ๐1 . For example, any two (based) maps ๐ 2 → ๐ 2 will induce the trivial map on ๐ 1 since ๐ 1 (๐ 2 ) = 1. (Moreover, since ๐2 (๐ 2 ) = Z ≠ 0, then ๐ 1 as a functor Toph∗ → Grp is not faithful either.) To show that ๐1 : Top∗ → Grp is not full consider the spaces R๐ 3 and R๐ 2, whose fundamental group is Z/2. Recall that ๐ป ∗ (R๐ ๐ ; Z/2Z) = Z/2Z[๐ฅ]/(๐ฅ ๐+1 ) for all ๐ ≥ 1. Thus any map ๐ : R๐ 3 → R๐ 2 induces the zero map on ๐ป 1 (−; Z/2Z), for if ๐ฅ ∈ ๐ป 1 (R๐ 2 ; Z/2Z) is a generator then 0 = ๐ ∗ (๐ฅ 3 ) = ๐ ∗ (๐ฅ) 3, so ๐ ∗ (๐ฅ) = 0 ∈ ๐ป 1 (R๐ 3 ; Z/2Z). Since there are 7 Solutions by positrón0802 1.2 Functors natural isomorphisms ๐ป 1 (R๐ ๐ ; Z/2Z) ๐ป 1 (R๐ ๐ ; Z) ๐ป 1 (R๐ ๐ ; Z) ๐ 1 (R๐ ๐ ), then any map R๐ 3 → R๐ 2 induces the trivial map on ๐ 1, so ๐ 1 : Top∗ → Grp is not a full functor. (The above example could also show that ๐1 : Toph∗ → Grp is not full as long as we could show that there is a map R๐ 3 → R๐ 2 which is not nullhomotopic. Is there?) • The ๐th homology group functor ๐ป๐ : Top → Ab. As in the previous examples, any map ๐ ๐+1 → ๐ ๐+1 will induce the trivial map on ๐ป๐ , so none of the functors ๐ป๐ : Toph → Ab is faithful. The same example as before shows that ๐ป 1 : Toph → Ab is not full. Now assume ๐ ≥ 2 and let ๐, ๐ > 0 be such that ๐ + ๐ = ๐. We show that any map ๐ : ๐ ๐ → ๐ ๐ × ๐ ๐ induces the trivial map on ๐ป๐ . Let ๐ผ ∈ ๐ป ๐ (๐ ๐ ) and ๐ฝ ∈ ๐ป ๐ (๐ ๐ ) be generators, and ๐ผ 0 = ๐ผ × 1 ∈ ๐ป ๐ (๐ ๐ × ๐ ๐ ), ๐ฝ 0 = 1 × ๐ฝ ∈ ๐ป ๐ (๐ ๐ × ๐ ๐ ). Then ๐พ = ๐ผ 0 ∪ ๐ฝ 0 ∈ ๐ป ๐ (๐ ๐ × ๐ ๐ ) is a generator by Künneth theorem, so ๐ ∗ : ๐ป ๐ (๐ ๐ ×๐ ๐ ) → ๐ป ๐ (๐ ๐ ) sends ๐พ to ๐ ∗ (๐ผ 0)∪๐ ∗ (๐ฝ 0) = 0. Thus ๐ is trivial on ๐ป ๐ , hence also on ๐ป๐ (by the universal coefficient theorem).This shows that ๐ป๐ : Top → Ab is not full. (And neither is ๐ป ๐ .) (Moreover, consider now ๐ป๐ as a functor Toph → Ab. For ๐ ≥ 3 consider ๐ = ๐ − 1, ๐ = 1. Since ๐๐+1 (๐ ๐ ) ≠ 0 for all ๐ ≥ 2 then ๐๐ (๐ ๐−1 × ๐ 1 ) ≠ 0, so there are maps ๐ ๐ → ๐ ๐−1 × ๐ 1 which are not nullhomotopic, and hence ๐ป๐ : Toph → Ab is not full for ๐ ≥ 3. For ๐ = 2 this does not work since all maps ๐ 2 → ๐ 1 × ๐ 1 are nullhomotopic. In this case we should find another example. Or maybe it is full?) • A monoid (group) homomorphism ๐น : ๐บ → ๐ป, considered as a functor ๐น : ๐ข → โ between the corresponding categories. ๐น is faithful if and only if it is injective as a map ๐บ → ๐ป, and full if and only if it is surjective as a map ๐บ → ๐ป . • Let ๐บ be a monoid, regarded as a category ๐ข with one element. Consider a functor ๐น : ๐ข → Set, i.e. a left ๐บ-set. Let ∗ ∈ ๐ข be the unique object and ๐ = ๐น (∗). ๐น is faithful if and only if the following condition holds: given ๐, ๐ 0 ∈ ๐บ, if ๐ · ๐ = ๐ 0 · ๐ for all ๐ ∈ ๐ then ๐ = ๐ 0 . This is the same as saying that the action of ๐บ in ๐ is faithful (in the sense of the definition in this context). Now, ๐น is full if and only if for all functions ๐ : ๐ → ๐ there exists some ๐ ∈ ๐บ such that ๐ (๐ ) = ๐ · ๐ , but this is only possible for bijections ๐ . Thus ๐น is full if and only if |๐ | ≤ 1. The same conclusions hold for a functor ๐บ : ๐ข → Vect๐ , a ๐-linear representation of ๐บ . • ๐ด, ๐ต (pre)ordered sets, ๐, โฌ the corresponding categories. Let ๐น : ๐ → โฌ be a functor, that is, an order-preserving map ๐น : ๐ด → ๐ต. Given ๐, ๐ 0 ∈ ๐ด the set ๐(๐, ๐ 0) is empty unless ๐ ≤ ๐ 0, and in this case there is a unique arrow ๐ → ๐ 0 . Thus ๐น is faithful since there is at most one arrow between two given elements of ๐. But ๐น is not necessarily full, for we can have an arrow ๐ (๐) → ๐ (๐ 0) in โฌ such that ๐, ๐ 0 ∈ ๐ด are not related, i.e. there is no arrow ๐ → ๐ 0. • The functor ๐ถ : Topop → Ring sending a space to its ring of continuous real-valued functions. Let ๐, ๐ be topological spaces, and assume that ๐ has the indiscrete topology. If 8 Solutions by positrón0802 1.2 Functors ๐ : ๐ → R is a map and ๐ฆ ∈ ๐ , then ๐(๐ฆ) ∈ R is non-empty closed, so ๐ −1 ({๐(๐ฆ)}) = ๐ . It follows that any map ๐ → R is constant. Thus, if ๐ , ๐ : ๐ → ๐ are two arbitrary maps then ๐ ∗ = ๐∗ : ๐ถ (๐ ) → ๐ถ (๐ ). By letting ๐ have at least two elements we see that ๐ถ is not faithful. Moreover, in the same example we see that the induced morphism ๐ถ (๐ ) → ๐ถ (๐ ) of a map ๐ → ๐ have image precisely the constant functions. Hence ๐ถ is not full either. op • The contravariant Hom functor Hom(−,๐ ) : Vect๐ → Vect๐ for a fixed ๐-vector space ๐ . If ๐ = 0 this functor is clearly full but not faithful, so assume ๐ ≠ 0. First we show that Hom(−,๐ ) is faithful. Let ๐ , ๐ ∈ Vect๐ and ๐ , ๐ : ๐ → ๐ be linear maps. Suppose that ๐ ≠ ๐, so that there exists ๐ข ∈ ๐ such that ๐ฃ B ๐ (๐ข) ≠ ๐(๐ข) B ๐ฃ 0 . Extend {๐ฃ } to a basis for ๐ , such that if ๐ฃ and ๐ฃ 0 are linear independent then ๐ฃ 0 belongs to this basis. Let ๐ฟ : ๐ → ๐ be a linear map sending ๐ฃ to some ๐ค ≠ 0 and all other basis elements to 0. If ๐ฃ, ๐ฃ 0 are linear dependent there exists ๐ผ ∈ ๐, ๐ผ ≠ 0, 1 such that ๐ผ๐ฃ = ๐ฃ 0, so that (๐ ∗ ๐ฟ) (๐ข) = ๐ฟ(๐ฃ) = ๐ค and (๐∗ ๐ฟ) (๐ข) = ๐ฟ(๐ฃ 0) = ๐ผ๐ค ≠ ๐ค . If ๐ฃ, ๐ฃ 0 are linear independent then (๐ ∗ ๐ฟ) (๐ข) = ๐ฟ(๐ฃ) = ๐ค ≠ 0 and (๐∗ ๐ฟ) (๐ข) = ๐ฟ(๐ฃ 0) = 0. Thus ๐ ∗ ≠ ๐∗ . It follows that the functor Hom(−,๐ ) is faithful. Now we show that Hom(−,๐ ) is not full. It is enough to consider the case ๐ = ๐. (For then it will hold for all ๐ ≠ 0 as every ๐ contains ๐ as a subspace.) Recall that if ๐ is a vector space with countable infinite dimension then ๐ ∗ has uncountable dimension. Thus not all linear maps ๐ ∗ → ๐ ∗ ๐ are duals of a linear map ๐ → ๐ , so Hom(−, ๐) is not full. • The functors ๐ป ๐ : Topop → Ab assigning a space to its ๐th cohomology group. None of these is faithful, for any map ๐ ๐+1 → ๐ ๐+1 induces the trivial map on ๐ป ๐ . This shows moreover that ๐ป ๐ is not faithful as a functor Tophop → Ab. The same examples as for ๐ป๐ above show that ๐ป ๐ : Topop → Ab is not full for ๐ ≠ 2. (And ๐ป ๐ : Tophop → Ab is not full for ๐ ≥ 3. Is it for ๐ = 2?) • A functor ๐น : ๐ขop → Set, i.e. a right ๐บ-set. This is analogous to example above of functors ๐ข → Set, i.e. left ๐บ-sets. • A presheaf on a category ๐, i.e. a functor ๐ op → Set. One does not expect this to be faithful or full unless we have an specific example. Consider a presheaf on a space ๐ ∈ Top, i.e. a functor ๐น : ๐ช(๐ ) op → Set. Then ๐น is faithful since for any two elements ๐ , ๐ ∈ ๐ช(๐ ), there is at most one morphism ๐ → ๐ . In general one does not expect ๐น to be full, for instance if ๐ ⊄ ๐ then there is no morphism ๐ → ๐ in ๐ช(๐ ) op, but there could be functions ๐น (๐ ) → ๐น (๐ ), e.g. if they are both non-empty. In the particular case in which ๐น is the presheaf of real continuous functions, ๐น is also not full in general for the same reason. (b) We can use one the previous examples: Let ๐บ and ๐ป be monoids; let ๐ข and โ denote the corresponding categories. The a functor ๐ข → โ is full (resp. faithful) if and only if the 1Z corresponding homomorphism ๐บ → ๐ป is surjective (resp. injective). Thus Z −→ Z is full and 9 Solutions by positrón0802 1.3 Natural Transformations 2 (2 0) 1 faithful, Z → − Z is faithful but not full, Z → − Z/2 is full but not faithful, and Z × Z −−−→ Z is neither full nor faithful. Exercise 1.2.29. (a) Let (๐, ≤๐ ) be a poset, i.e. a (partially) ordered set. Then a subcategory of ๐ is a poset (๐, ≤๐ ) such that ๐ ⊂ ๐ and if ๐ฅ ≤๐ ๐ฆ for ๐ฅ, ๐ฆ ∈ ๐, then ๐ฅ ≤๐ ๐ฆ. A subcategory ๐ ⊂ ๐ is full if and only if ๐ is a subposet of ๐, i.e. if ๐ฅ ≤๐ ๐ฆ ⇔ ๐ฅ ≤๐ ๐ฆ for all ๐ฅ, ๐ฆ ∈ ๐. (b) As we do not require closure under inverses, the subcategories of a group are precisely its submonoids (and the empty category). Only one subcategory is full, namely the whole group itself. 1.3 Natural Transformations Exercise 1.3.25. ๐ Mod. • Given a ring ๐ , there is a natural isomorphism 1๐ Mod Hom๐ (๐ , −) : ๐ Mod → • Given a ring ๐ , there is a natural isomorphism 1๐ Mod ๐ ⊗๐ − : ๐ Mod → ๐ Mod. • Let ๐ ≥ 2 and consider the functors ๐๐ , ๐ป๐ : Top∗ → Ab. Then there is a natural transformation โ : ๐๐ ⇒ ๐ป๐ such that for all based spaces (๐, ๐ฅ 0 ) ∈ Top∗, โ (๐ ,๐ฅ 0 ) : ๐๐ (๐, ๐ฅ 0 ) → ๐ป๐ (๐ ) is the Hurewicz homomorphism. Exercise 1.3.26. First assume that ๐ผ is an isomorphism and let ๐ฝ : ๐บ ⇒ ๐น be an inverse of ๐ผ . Then ๐ผ๐ด โฆ ๐ฝ๐ด = 1๐บ (๐ด) and ๐ฝ๐ด โฆ ๐ผ๐ด = 1๐น (๐ด) for all ๐ด ∈ ๐, so ๐ผ๐ด : ๐น (๐ด) → ๐บ (๐ด) is an isomorphism for all ๐ด ∈ ๐. Conversely, assume that ๐ผ๐ด : ๐น (๐ด) → ๐บ (๐ด) is an isomorphism for all ๐ด ∈ ๐. Then for each ๐ด ∈ ๐ there exists an inverse ๐ฝ๐ด : ๐บ (๐ด) → ๐น (๐ด) of ๐ผ๐ด . If we prove that ๐ฝ defines a natural transformation ๐บ ⇒ ๐น then ๐ฝ is an inverse of ๐ผ . For this purpose, we need to show that for each ๐ด, ๐ด 0 ∈ ๐ and ๐ : ๐ด → ๐ด 0, the diagram ๐บ (๐ด) ๐บ (๐ ) ๐ฝ๐ด0 ๐ฝ๐ด ๐น (๐ด) ๐บ (๐ด 0) ๐น (๐ ) ๐น (๐ด 0) commutes. By naturality of ๐ผ we have ๐บ (๐ ) โฆ ๐ผ๐ด = ๐ผ๐ด0 โฆ ๐น (๐ ), so that ๐น (๐ ) โฆ ๐ฝ๐ด = ๐ฝ๐ด0 โฆ ๐ผ๐ด0 โฆ ๐น (๐ ) โฆ ๐ฝ๐ด = ๐ฝ๐ด0 โฆ ๐บ (๐ ) โฆ ๐ผ๐ด โฆ ๐ฝ๐ด = ๐ฝ๐ด0 โฆ ๐บ (๐ ). It follows that ๐ฝ : ๐บ ⇒ ๐น is a natural transformation which is inverse to ๐ผ . Exercise 1.3.27. Define a functor ๐นe: [๐ op, โฌop ] → [๐, โฌ] op as follows. Given a functor ๐ป : ๐ op → โฌop, let ๐นe(๐ป ) : ๐ → โฌ be the functor given on objects ๐ด ∈ ๐ by ๐นe(๐ป ) (๐ด) = ๐ป (๐ด) 10 Solutions by positrón0802 1.3 Natural Transformations and on morphisms ๐ ∈ ๐(๐ด, ๐ด 0) = ๐ op (๐ด 0, ๐ด) by ๐นe(๐ป ) (๐ ) = ๐ป (๐ ) ∈ โฌop (๐ป (๐ด 0), ๐ป (๐ด)) = โฌ(๐ป (๐ด), ๐ป (๐ด 0)). Given a natural transformation ๐ป ๐ op ๐ผ โฌop ๐พ we need ๐นe(๐ผ) ∈ [๐, โฌ] op ( ๐นe(๐ป ), ๐นe(๐พ)) = [๐, โฌ] ( ๐นe(๐พ), ๐นe(๐ป )), i.e. a natural transformation ๐นe(๐ผ) : ๐นe(๐พ) ⇒ ๐นe(๐ป ). For each ๐ด ∈ ๐ we let ๐นe(๐ผ)๐ด = ๐ผ๐ด ∈ โฌop (๐ป (๐ด), ๐พ (๐ด)) = โฌ(๐พ (๐ด), ๐ป (๐ด)). Given ๐ ∈ ๐(๐ด, ๐ด 0) = ๐ op (๐ด 0, ๐ด) the diagram ๐นe(๐พ) (๐ด) = ๐พ (๐ด) ๐นe(๐พ) (๐ )=๐พ (๐ ) ๐นe(๐พ) (๐ด 0) = ๐พ (๐ด 0) ๐นe(๐ผ)๐ด0 =๐ผ๐ด0 ๐นe(๐ผ)๐ด =๐ผ๐ด ๐นe(๐ป ) (๐ด) = ๐ป (๐ด) ๐นe(๐ป ) (๐ )=๐ป (๐ ) ๐นe(๐ป ) (๐ด 0) = ๐ป (๐ด 0) commutes by naturality of ๐ผ, so ๐นe(๐พ) ๐ ๐นe(๐ผ) โฌ ๐นe(๐ป ) is a natural transformation. If ๐ผ : ๐ป ⇒ ๐พ and ๐ฝ : ๐พ → ๐ฟ are natural transformations of functors ๐ป, ๐พ, ๐ฟ : ๐ op ⇒ โฌop then ๐นe(๐ผ โฆ ๐ฝ)๐ด = ๐ผ๐ด โฆ ๐ฝ๐ด = ๐นe(๐ผ)๐ด โฆ ๐นe(๐ฝ)๐ด, so ๐นe(๐ผ โฆ ๐ฝ) = ๐นe(๐ผ) โฆ ๐นe(๐ฝ). It follows that ๐นe is a functor [๐ op, โฌop ] → [๐, โฌ] op . e : [๐, โฌ] op → [๐ op, โฌop ] as follows. Given a functor ๐ป : ๐ → โฌ let Now define a functor ๐บ op op e(๐ป ) : ๐ → โฌ be the functor given on objects ๐ด ∈ ๐ op by ๐บ e(๐ป ) (๐ด) = ๐ป (๐ด) and on morph๐บ op 0 0 e isms ๐ ∈ ๐ (๐ด , ๐ด) = ๐(๐ด, ๐ด ) by ๐บ (๐ป ) (๐ ) = ๐ป (๐ ) ∈ โฌ(๐ป (๐ด), ๐ป (๐ด 0)) = โฌop (๐ป (๐ด 0), ๐ป (๐ด)). e(๐ผ) ∈ [๐ op, โฌop ] (๐บ e(๐พ), ๐บ e(๐ป )) be given by Given ๐ผ ∈ [๐, โฌ] op (๐พ, ๐ป ) = [๐, โฌ] (๐ป, ๐พ) let ๐บ op e ๐บ (๐ผ)๐ด = ๐ผ๐ด ∈ โฌ(๐ป (๐ด), ๐พ (๐ด)) = โฌ (๐พ (๐ด), ๐ป (๐ด)) for all ๐ด ∈ ๐. Thus, if ๐ ∈ ๐ op (๐ด 0, ๐ด) = ๐(๐ด, ๐ด 0) then e(๐ป ) (๐ด) = ๐ป (๐ด) ๐บ e(๐ป ) (๐ )=๐ป (๐ ) ๐บ e(๐ป ) (๐ด 0) = ๐ป (๐ด 0) ๐บ e(๐ผ)๐ด0 =๐ผ๐ด0 ๐บ e(๐ผ)๐ด =๐ผ๐ด ๐บ e(๐พ) (๐ด) = ๐พ (๐ด) ๐บ e(๐พ) (๐ )=๐พ ( ๐ ) ๐บ e(๐พ) (๐ด 0) = ๐พ (๐ด 0) ๐บ commutes by naturality of ๐ผ, so e(๐พ) ๐บ ๐ op e(๐ผ) ๐บ โฌop e(๐ป ) ๐บ 11 Solutions by positrón0802 1.3 Natural Transformations e : [๐, โฌ] op → [๐ op, โฌop ]. is a natural transformation. This defines a functor ๐บ It is clear that ๐น๐บ = 1 [๐,โฌ] op and ๐บ๐น = 1 [๐ op,โฌop ] , so ๐น is an isomorphism [๐ op, โฌop ] [๐, โฌ] op . Exercise 1.3.28. (a) Let ๐ : ๐ด × ๐ต๐ด → ๐ต be given by ๐ (๐, ๐) = ๐(๐) for all ๐ ∈ ๐ด and ๐ ∈ ๐ต๐ด . (b) Let โ : ๐ด → ๐ต (๐ต ๐ด) be given by โ(๐) (๐ ) = ๐ (๐) for all ๐ ∈ ๐ด and ๐ ∈ ๐ต๐ด . Exercise 1.3.29. First assume that the family (๐ผ๐ด,๐ต : ๐น (๐ด, ๐ต) → ๐บ (๐ด, ๐ต))๐ด∈๐,๐ต ∈โฌ is a natural transformation ๐น ⇒ ๐บ . Fix ๐ด ∈ โฌ and consider the family (๐ผ๐ด,๐ต : ๐น ๐ด (๐ต) → ๐บ ๐ด (๐ต))๐ต ∈โฌ . To show this defines a natural transformation ๐น ๐ด ⇒ ๐บ ๐ด we need to prove that for all ๐ต, ๐ต 0 ∈ โฌ and ๐ ∈ โฌ(๐ต, ๐ต 0), the diagram ๐น ๐ด (๐ต) = ๐น (๐ด, ๐ต) ๐น ๐ด (๐ )=๐น (1๐ด ,๐ ) ๐น ๐ด (๐ต 0) = ๐น (๐ด, ๐ต 0) ๐ผ๐ด,๐ต0 ๐ผ๐ด,๐ต ๐บ ๐ด (๐ต) = ๐บ (๐ด, ๐ต) ๐บ ๐ด (๐ )=๐บ (1๐ด ,๐ ) ๐บ ๐ด (๐ต 0) = ๐บ (๐ด, ๐ต 0) commutes, and it indeed does since ๐ผ๐ด,๐ต : ๐น ⇒ ๐บ is natural. Similarly, the family (๐ผ๐ด,๐ต : ๐น๐ต (๐ด) → ๐บ ๐ต (๐ด))๐ด∈๐ is a natural transformation ๐น๐ต ⇒ ๐บ ๐ต . Conversely, assume that the above families define natural transformations ๐น ๐ด ⇒ ๐บ ๐ด and ๐น๐ต ⇒ ๐บ ๐ต for all ๐ด ∈ ๐, ๐ต ∈ โฌ. Let (๐ด, ๐ต), (๐ด 0, ๐ต 0) ∈ ๐ × โฌ and (๐ , ๐) ∈ ๐ × โฌ((๐ด, ๐ต), (๐ด 0, ๐ต 0)). We want to show that the diagram ๐น ( ๐ ,๐) ๐น (๐ด, ๐ต) ๐น (๐ด 0, ๐ต 0) ๐ผ๐ด0,๐ต0 ๐ผ๐ด,๐ต ๐บ (๐ด, ๐ต) ๐บ (๐ ,๐) ๐บ (๐ด 0, ๐ต 0) commutes. Since ๐น (๐ , ๐) = ๐น (๐ , 1๐ต ) โฆ ๐น (1๐ด, ๐), ๐บ (๐ , ๐) = ๐บ (๐ , 1๐ต ) โฆ ๐บ (1๐ด, ๐), and the diagram ๐น (๐ด, ๐ต) ๐น (1๐ด ,๐) ๐น (๐ด, ๐ต 0) ๐ผ๐ด,๐ต0 ๐ผ๐ด,๐ต ๐บ (๐ด, ๐ต) ๐น (๐ ,1๐ต ) ๐บ (1๐ด ,๐) ๐บ (๐ด, ๐ต 0) ๐น (๐ด 0, ๐ต 0) ๐ผ๐ด0,๐ต0 ๐บ (๐ ,1๐ต ) ๐บ (๐ด 0, ๐ต 0) commutes by hypothesis, the result follows. Exercise 1.3.30. Let ∗Z denote the unique object in Z and ∗๐บ the unique object in ๐บ . The oneto-one correspondence described assigns to an element ๐ ∈ ๐บ the functor Z → ๐บ sending a morphism ๐ ∈ Z(∗Z, ∗Z ) to ๐๐ ∈ ๐บ (∗๐บ , ∗๐บ ). Note that every natural transformation between functors Z → ๐บ is an isomorphism by Exercise 1.3.26, since every element in a group is invertible. 12 Solutions by positrón0802 1.3 Natural Transformations If โ, ๐ are functors Z → ๐บ, a natural isomorphism ๐ผ : โ → ๐ amounts to a morphism ๐ผ ∈ ๐บ (∗๐บ , ∗๐บ ) such that the diagram ∗๐บ โ (๐) ∗๐บ ๐ผ ∗๐บ ๐ผ ∗๐บ ๐ (๐) commutes for all ๐ ∈ Z, i.e. ๐ผ ∈ ๐บ is such that ๐ผโ๐ = ๐๐ ๐ผ for all ๐ ∈ Z. This holds if and only if ๐ผโ๐ผ −1 = ๐. Thus, the equivalence relation on ๐บ is that of conjugacy. Exercise 1.3.31. (a) Given ๐, ๐ 0 ∈ โฌ and ๐ ∈ โฌ(๐, ๐ 0), define Sym(๐ ) ∈ Set(Sym(๐ ), Sym(๐ 0)) to be given by โ โฆ→ ๐ โ๐ −1 for โ ∈ Sym(๐ ). Now define Ord(๐ ) ∈ Set(Ord(๐ ), Ord(๐ 0)) as follows. Given a total order ≤ on ๐ let Ord(๐ )(≤) be the order on ๐ 0 given by ๐ฆ ๐ง if and only if ๐ −1 (๐ฆ) ≤ ๐ −1 (๐ง). It is clear that both these definitions give functors โฌ → Set. (b) Suppose that ๐ผ : Sym ⇒ Ord is a natural transformation. Let ๐, ๐ 0 ∈ โฌ be finite groups of the same cardinality and ๐ : ๐ → ๐ 0 be a bijection. Then there is a commutative diagram Sym(๐ ) Sym(๐ ) Sym(๐ 0) ๐ผ๐ 0 ๐ผ๐ Ord(๐ ) Ord(๐ ) Ord(๐ 0) . Consider the elements 1๐ ∈ Sym(๐ ), 1๐ 0 ∈ Sym(๐ 0) and denote ๐ผ๐ (1๐ ) by ≤๐ and ๐ผ๐ 0 (1๐ 0 ) by ≤๐ 0 . By definition, we have Sym(๐ )(1๐ ) = 1๐ 0 . Hence Ord(๐ ) (≤๐ ) = ≤๐ 0 . This means that no matter what ๐ is, Ord(๐ ) sends ≤๐ to ≤๐ 0 . This situation is clearly impossible by considering, e.g., ๐ = ๐ 0 and |๐ | ≥ 2 since there is at least two different bijections giving different total orders in ๐ . (c) It is clear that |Sym(๐ )| = |Ord(๐ )| = ๐!. Since Sym(๐ ) and Ord(๐ ) have the same cardinal, we have Sym(๐ ) Ord(๐ ) for all ๐ ∈ โฌ. But this isomorphism is not natural in ๐ as there is no natural transformation Sym ⇒ Ord. Exercise 1.3.32. (a) By assumption there exists a functor ๐บ : โฌ → ๐ and natural isomorphisms ๐ : ๐บ๐น ⇒ 1๐ and ๐ : ๐น๐บ ⇒ 1โฌ . If ๐ต ∈ โฌ, then ๐บ (๐ต) ∈ ๐ and ๐น๐บ (๐ต) ๐ต via ๐. Hence ๐น is essentially surjective on objects. Now let ๐ด, ๐ด 0 ∈ ๐ and consider the function ๐(๐ด, ๐ด 0) → โฌ(๐น (๐ด), ๐น (๐ด 0)) ๐ โฆ→ ๐น (๐ ). 13 Solutions by positrón0802 1.3 Natural Transformations To show it is injective assume that ๐น (๐ ) = ๐น (๐) for ๐ , ๐ ∈ ๐(๐ด, ๐ด 0). Then ๐บ๐น (๐ ) = ๐บ๐น (๐). By naturality of ๐ the diagram ๐บ๐น (๐ด) ๐บ๐น (๐ )=๐บ๐น (๐) ๐บ๐น (๐ด 0) ๐๐ด0 ๐๐ด ๐ด0 ๐ด ๐ commutes. Since ๐๐ด is an isomorphism we have ๐ = ๐๐ด0 โฆ ๐บ๐น (๐ ) โฆ ๐๐ด−1, and the same can be said about ๐, so ๐ = ๐. Thus ๐น is faithful. Now we show that the above function is surjective for all ๐ด, ๐ด 0 ∈ ๐. Let โ ∈ โฌ(๐น (๐ด), ๐น (๐ด 0)) be arbitrary. There exists a unique ๐ ∈ ๐(๐ด, ๐ด 0) making the diagram ๐บ๐น (๐ด) ๐บ (โ) ๐บ๐น (๐ด 0) ๐๐ด0 ๐๐ด ๐ด0 ๐ด ๐ commute, namely ๐ = ๐๐ด0 โฆ ๐บ (โ) โฆ ๐๐ด−1 . Since ๐บ is faithful (by the previous proof since it is an equivalence), then for proving ๐น (๐ ) = โ it suffices to prove ๐บ๐น (๐ ) = ๐บ (โ). But both ๐บ๐น (๐ ) and ๐บ (โ) fill the dashed arrow in the commutative diagram ๐บ๐น (๐ด 0) ๐บ๐น (๐ด) ๐๐ด0 ๐๐ด ๐ด ๐ ๐ด0 , so both equal ๐๐ด−10 โฆ ๐ โฆ ๐๐ด . Thus ๐น (๐ ) = โ and it follows that ๐น is full. (b) Define ๐บ : โฌ → ๐ as follows. Since ๐น is essentially surjective, given ๐ต ∈ โฌ there exists ๐บ (๐ต) ∈ ๐ and an isomorphism ๐๐ต : ๐น๐บ (๐ต) → ๐ต. We want ๐ to define a natural isomorphism ๐น๐บ ⇒ 1๐ต , i.e. that for all ๐ต, ๐ต 0 ∈ โฌ and ๐ ∈ โฌ(๐ต, ๐ต 0) the diagram ๐น๐บ (๐ต) ๐น๐บ (๐ ) ๐น๐บ (๐ต 0) ๐๐ต0 ๐๐ต ๐ต0 ๐ต ๐ commutes. Since ๐๐ต0 is an isomorphism this amounts to having ๐น๐บ (๐ ) = ๐๐ต−10 โฆ ๐ โฆ ๐๐ต . Thus, given ๐ ∈ โฌ(๐ต, ๐ต 0) we define ๐บ (๐ ) to be the unique morphism ๐บ (๐ต) → ๐บ (๐ต 0) corresponding to ๐๐ต−10 โฆ ๐ โฆ ๐๐ต : ๐น๐บ (๐ต) → ๐น๐บ (๐ต 0) under the function ๐(๐บ (๐ต), ๐บ (๐ต 0)) → โฌ(๐น๐บ (๐ต), ๐น๐บ (๐ต 0)) โ โฆ→ ๐น (โ), 14 Solutions by positrón0802 1.3 Natural Transformations which is a bijection since ๐น is full and faithful. If ๐ ∈ โฌ(๐ต, ๐ต 0) and ๐ ∈ โฌ(๐ต 0, ๐ต 00) then the equation ๐น๐บ (๐) โฆ ๐น๐บ (๐ ) = ๐น๐บ (๐ โฆ ๐ ) guarantees ๐บ (๐) โฆ ๐บ (๐ ) = ๐บ (๐ โฆ ๐ ), so ๐บ is a well-defined functor โฌ → ๐. By construction, ๐ : ๐น๐บ ⇒ 1๐ต is a natural isomorphism. It remains to construct a natural isomorphism ๐ : ๐บ๐น ⇒ 1๐ . Given ๐ด ∈ ๐, let ๐๐ด : ๐บ๐น (๐ด) → ๐ด be the unique morphism corresponding to ๐ ๐น (๐ด) : ๐น๐บ๐น (๐ด) → ๐น (๐ด) under the bijection ๐(๐บ๐น (๐ด), ๐ด) → โฌ(๐น๐บ๐น (๐ด), ๐น (๐ด)) โ โฆ→ ๐น (โ). Since ๐ ๐น (๐ด) is an isomorphism, so is ๐๐ด . We thus have a family of isomorphisms (๐๐ด : ๐ด → ๐บ๐น (๐ด))๐ด∈๐ . To show that ๐ is natural, let ๐ด, ๐ด 0 ∈ ๐, ๐ ∈ ๐(๐ด, ๐ด 0) and consider the diagram ๐บ๐น (๐ด) ๐บ๐น (๐ ) ๐บ๐น (๐ด 0) ๐๐ด0 ๐๐ด ๐ด0 . ๐ด ๐ We want to show that it commutes. But applying ๐น we obtain the commutative diagram ๐น๐บ๐น (๐ ) ๐น๐บ๐น (๐ด) ๐น๐บ๐น (๐ด 0) ๐น (๐๐ด0 )=๐ ๐น (๐ด0 ) ๐น (๐๐ด )=๐ ๐น (๐ด) ๐น (๐ด) ๐น (๐ด 0) , ๐น (๐ ) so the first diagram commutes since the functor ๐น is faithful. Thus ๐ is a natural isomorphism. We conclude that ๐น is an equivalence. Exercise 1.3.33. (This is also Mac Lane’s Exercise I.4.6.) Composition in Mat is given by multiplication of matrices. For each finite-dimensional ๐-vector space ๐ fix an ordered basis {๐๐๐ }1≤๐ ≤dim ๐ . If ๐ = ๐น ๐ for some ๐, we take the (ordered) standard basis. Define a functor ๐ : FDVect → Mat as follows. For ๐ ∈ FDVect let ๐ (๐ ) = dim ๐ . If ๐ฟ ∈ FDVect(๐ ,๐ ) let Mat(dim ๐ , dim๐ ) be the dim๐ ×dim ๐ matrix of ๐ฟ with respect to the bases {๐๐๐ }๐ , {๐๐ ๐ } ๐ . Then ๐ is a functor FDVect → Mat. On the other hand, for each ๐ ∈ Mat let ๐บ (๐) = ๐ ๐ , and given ๐ด ∈ Mat(๐, ๐) let ๐บ (๐ด) : ๐ ๐ → ๐๐ have matrix ๐ด with respect to the standard bases for ๐ ๐ and ๐๐ . Then ๐บ : Mat → FDVect is a functor such that ๐๐บ = 1Mat . Furthermore, for all ๐ ∈ FDVect there is an isomorphism ๐๐ : ๐ → ๐ ๐ , ๐ = dim ๐ , sending the ordered basis {๐๐๐ }1≤๐ ≤๐ onto the (ordered) standard basis for ๐ ๐ . Then, if ๐ : ๐ → ๐ is a linear map, the square ๐๐ ๐๐ ๐ ๐บ๐ (๐ ) ๐ ๐ ๐๐ ๐๐ 15 Solutions by positrón0802 2. Adjoints Í ๐ ๐ commutes, where ๐ = dim ๐ , ๐ = dim๐ . Indeed, if ๐ (๐๐๐ ) = ๐ ๐=1 ๐ด๐ ๐ ๐ for each ๐ = 1, . . . , ๐, Í ๐ ๐๐ then both ๐บ๐ (๐ ) โฆ๐๐ and ๐๐ โฆ ๐ send ๐๐๐ to ๐ ๐=1 ๐ด๐ ๐ ๐ for all ๐. Then ๐ is a natural isomorphism 1FDVect ๐บ๐ and we conclude that FDVect is equivalent to Mat. Our equivalence does not involve a canonical functor FDVect → Mat since we fixed bases for all finite-dimensional vector spaces. However, the functor ๐บ : Mat → FDVect is canonical. Exercise 1.3.34. For any category ๐ we have ๐ ' ๐ via the identity functor 1๐ . It is also clear that this relation is symmetric; it remains to prove it is transitive. So assume ๐, ๐, โฐ are categories such that ๐ ' ๐ and ๐ ' โฐ. Then there exist functors ๐น : ๐ → ๐, ๐บ : ๐ → ๐ and ๐ป : ๐ → โฐ, ๐พ : โฐ → ๐ such that ๐บ๐น 1๐ , ๐น๐บ 1๐, ๐พ๐ป 1๐ and ๐ป๐พ 1โฐ . Consider the functors ๐ป ๐น : ๐ → โฐ and ๐บ๐พ : โฐ → ๐; we show there are isomorphisms ๐บ๐พ๐ป ๐น 1๐ and ๐ป ๐น๐บ๐พ 1โฐ . Let ๐ผ : ๐พ๐ป ⇒ 1๐ be a natural isomorphism. Then ๐ผ ๐ท : ๐พ๐ป (๐ท) → ๐ท is an isomorphism for all ๐ท ∈ ๐ (Exercise 1.3.26). In particular, given ๐ถ ∈ ๐, ๐ผ ๐น (๐ถ) : ๐พ๐ป ๐น (๐ถ) → ๐น (๐ถ) is an isomorphism, so ๐บ (๐ผ ๐น (๐ถ) ) : ๐บ๐พ๐ป ๐น (๐ถ) → ๐บ๐น (๐ถ) is an isomorphism by functoriality of ๐บ . Let ๐ฝ : ๐บ๐น ⇒ 1๐ be a natural isomorphism. Define ๐๐ถ = ๐ฝ๐ถ โฆ ๐บ (๐ผ ๐น (๐ถ) ) : ๐บ๐พ๐ป ๐น (๐ถ) → ๐ถ. Then (๐๐ถ )๐ถ ∈๐ defines a family of isomorphisms. It remains to prove naturality: Let ๐ถ, ๐ถ 0 ∈ ๐ and ๐ ∈ ๐(๐ถ, ๐ถ 0). Consider the diagram ๐บ๐พ๐ป ๐น (๐ถ) ๐บ๐พ๐ป ๐น (๐ ) ๐บ (๐ผ ๐น (๐ถ 0 ) ) ๐บ (๐ผ ๐น (๐ถ ) ) ๐บ๐น (๐ถ) ๐บ๐พ๐ป ๐น (๐ถ 0) ๐บ๐น (๐ ) ๐บ๐น (๐ถ 0) ๐ฝ๐ถ 0 ๐ฝ๐ถ ๐ถ0 ๐ถ ๐ The upper square commutes by naturality of ๐ผ and functoriality of ๐บ, and the lower square commutes by naturality of ๐ฝ. By definition of ๐, this shows that ๐ : ๐บ๐พ๐ป ๐น ⇒ 1๐ is a natural transformation, hence a natural isomorphism (Exercise 1.3.26 again). Similarly there is a natural isomorphism ๐ป ๐น๐บ๐พ ⇒ 1โฐ and therefore ๐ ' โฐ. We conclude that equivalence of categories is an equivalence relation. 2 2.1 Adjoints Definition and examples Exercise 2.1.12. Adjoint functors: • If ๐ , ๐ are rings, and ๐ด is an (๐, ๐ )-bimodule, then the functor ๐ด ⊗๐ − : ๐ Mod → ๐ Mod is left adjoint to the functor ๐ Mod(๐ด, −) : ๐ Mod → ๐ Mod. (Tensor-hom adjunction.) 16 Solutions by positrón0802 2.1 Definition and examples • Let ๐ : ๐ → ๐ be a function of sets. Let ๐ −1 (−) : ๐ซ(๐ ) → ๐ซ(๐ ), ๐ต โฆ→ ๐ −1 (๐ต), be the inverse image functor. Then the image functor ๐ (−) : ๐ซ(๐ ) → ๐ซ(๐ ), ๐ด โฆ→ ๐ (๐ด), is left adjoint to ๐ −1 (−). Moreover, ๐ −1 (−) has a right adjoint given by sending ๐ด ∈ ๐ซ(๐ ) to the largest subset ๐ต of ๐ such that ๐ −1 (๐ต) ⊂ ๐ด. • Let Σ : Toph∗ → Toph∗ denote the (reduced) suspension functor, and Ω : Toph∗ → Toph∗ denote the loop space functor. Then ๐ is left adjoint to Ω. Initial and terminal objects: • The category Top∗ of pointed topological spaces has no initial object, and any one-point space is terminal. • For a vector space ๐, Vect๐ the zero vector space 0 ∈ Vect๐ is both initial and terminal. • It ๐ is a discrete category with more than one object, then ๐ has no initial nor terminal object. • The category Field has no initial nor terminal object, since given ๐น, ๐พ ∈ Field there is no morphism ๐น → ๐พ if char ๐น ≠ char ๐พ . • In Exercise 0.13(a) we showed that (Z[๐ฅ], ๐ฅ) is initial in the category of pointed rings and basepoint-preserving ring homomorphisms. Exercise 2.1.13. Let ๐น : ๐ → โฌ and ๐บ : โฌ → ๐ be adjoint functors between discrete categories, so that for all ๐ด ∈ ๐ and ๐ต ∈ โฌ we have a bijection โฌ(๐น (๐ด), ๐ต) ๐(๐ด, ๐บ (๐ต)). This means that ๐น (๐ด) = ๐ต if and only if ๐บ (๐ต) = ๐ด. So ๐น and ๐บ are mutually inverse and therefore ๐ โฌ. Exercise 2.1.14. First assume that the equation ๐ ๐ ๐น (๐) ๐บ (๐) ๐ ๐ (๐ด 0 → − ๐ด→ − ๐บ (๐ต) −−−−→ ๐บ (๐ต 0)) = (๐น (๐ด 0) −−−→ ๐น (๐ด) → − ๐ต→ − ๐ต 0) holds for all ๐, ๐ and ๐. By considering ๐ = 1๐ด : ๐ด → ๐ด we obtain (the transpose of) equation (2.2) for all ๐ and ๐ (with ๐ = ๐ in that equation), and by considering ๐ = 1๐ต → ๐ต we obtain equation (2.3) for all ๐ and ๐ . Conversely, assume (2.2) and (2.3) hold and let ๐ : ๐ด 0 → ๐ด, ๐ : ๐ด → ๐บ (๐ต) and ๐ : ๐ต → ๐ต 0 . Then ๐ ๐ ๐บ (๐) ๐น (๐) ๐บ (๐)โฆ๐ ๐น (๐) ๐ ๐ (๐ด 0 → − ๐ต→ − ๐ต 0), − ๐ด→ − ๐บ (๐ต) −−−−→ ๐บ (๐ต 0)) = ๐น (๐ด 0) −−−→ ๐น (๐ด) −−−−−−→ ๐ต 0 = (๐น (๐ด 0) −−−→ ๐น (๐ด) → the first equality by (2.3) and the second one by (the transpose of) (2.2). Exercise 2.1.15. Given ๐ต ∈ โฌ there is a bijection โฌ(๐น (๐ผ ), ๐ต) ๐(๐ผ, ๐บ (๐ต)). Since ๐ผ ∈ ๐ is initial there is precisely one map ๐ผ → ๐บ (๐ต), hence precisely one map ๐น (๐ผ ) → ๐ต. Thus ๐น (๐ผ ) ∈ โฌ is initial. Similarly, if ๐ ∈ โฌ is terminal and ๐ด ∈ ๐, the bijection ๐(๐ด, ๐บ (๐ )) โฌ(๐น (๐ด),๐ ) shows that there is precisely one map ๐บ (๐ ) → ๐ด, so ๐บ (๐ ) ∈ ๐ is terminal. 17 Solutions by positrón0802 2.1 Definition and examples Exercise 2.1.16. (a) First consider the forgetful functor ๐ : [๐บ, Set] → Set, whose object function sends a ๐บ-set ๐ ∈ [๐บ, Set] to its underlying set ๐ (∗) ∈ Set. We will show that ๐ has both left and right adjoints. First let ๐น : Set → [๐บ, Set] be given on objects by sending ๐ ∈ Set to ๐บ × ๐ ∈ [๐บ, Set], where the ๐บ-action on ๐บ × ๐ is given by โ · (๐, ๐ฅ) = (โ๐, ๐ฅ) for all ๐, โ ∈ ๐บ and ๐ฅ ∈ ๐ . It is clear that this defines a ๐บ-action on ๐บ × ๐, and a map ๐ : ๐ → ๐ of sets induces a map ๐น (๐ ) : ๐บ × ๐ → ๐บ × ๐ of ๐บ-sets given by (๐, ๐ฅ) โฆ→ (๐, ๐ (๐ฅ)), which clearly preserves the ๐บ-action. We claim that ๐น a ๐ , i.e., for ๐ ∈ Set, ๐ ∈ [๐บ, Set] there is an isomorphism ๐๐ ,๐ : [๐บ, Set] (๐บ × ๐, ๐ ) → Set(๐, ๐ (๐ )) natural in ๐ and ๐ . Let ๐ ∈ ๐บ denote the identity element. For ๐ ∈ [๐บ, Set] (๐บ × ๐, ๐ ) define ๐๐ ,๐ (๐ ) = ๐e ∈ Set(๐, ๐ (๐ )) to be given by ๐e(๐ฅ) = ๐ (๐, ๐ฅ) for all ๐ฅ ∈ ๐ . On the other hand, for ๐e ∈ Set(๐, ๐ (๐ )) define ๐๐ ,๐ ( ๐e) = ๐ ∈ [๐บ, Set] (๐น (๐ ), ๐ ) by ๐ (๐, ๐ฅ) = ๐ · ๐e(๐ฅ). Note that this is well-defined since ๐ (โ · (๐, ๐ฅ)) = ๐ (โ๐, ๐ฅ) = (โ๐) · ๐e(๐ฅ) = โ · (๐ · ๐e(๐ฅ)) = โ · ๐e(๐, ๐ฅ). We claim ๐๐−1,๐ = ๐๐ ,๐ . Indeed, for ๐ ∈ [๐บ, Set] (๐บ × ๐, ๐ ) and ๐ฅ ∈ ๐ we have ๐๐ ,๐ โฆ ๐๐ ,๐ (๐ ) (๐, ๐ฅ) = ๐ · ๐๐ ,๐ (๐ ) (๐ฅ) = ๐ · ๐ (1, ๐ฅ) = ๐ (๐ · (๐, ๐ฅ)) = ๐ (๐, ๐ฅ), and for ๐e ∈ Set(๐, ๐ (๐ )) and (๐, ๐ฅ) ∈ ๐บ × ๐ we have ๐๐ ,๐ โฆ ๐๐ ,๐ ( ๐e) (๐ฅ) = ๐๐ ,๐ ( ๐e) (๐, ๐ฅ) = ๐ · ๐e(๐ฅ) = ๐e(๐ฅ). Hence ๐๐ ,๐ is a bijection with inverse ๐๐ ,๐ . It remains to show naturality. This amounts to showing that for all ๐, ๐ 0 ∈ Set and ๐ , ๐ 0 ∈ [๐บ, Set], and morphisms ๐ : ๐ 0 → ๐, ๐ : ๐ → ๐ 0, the diagram [๐บ, Set] (๐บ × ๐, ๐ ) ๐๐ ,๐ Set(๐ ,๐ (๐)) [๐บ,Set] (๐น (๐ ),๐) [๐บ, Set] (๐บ Set(๐, ๐ (๐ )) × ๐ 0, ๐ 0) ๐๐ 0,๐ 0 Set(๐ 0, ๐ (๐ 0)) commutes (c.f. Exercise 2.1.14). So let ๐ ∈ [๐บ, Set] (๐บ × ๐, ๐ ) and ๐ฅ 0 ∈ ๐ 0 . On the one hand we have Set(๐ , ๐ (๐)) (๐๐ ,๐ (๐)) (๐ฅ 0) = ๐ (๐) โฆ ๐๐ ,๐ (๐) โฆ ๐ (๐ฅ 0) = ๐(๐(๐, ๐ (๐ฅ 0))), and on the other hand ๐๐ 0,๐ 0 ([๐บ, Set] (๐น (๐ ), ๐) (๐)) (๐ฅ 0) = ๐๐ 0,๐ 0 (๐ โฆ โ โฆ ๐น (๐ )) (๐ฅ 0) = ๐ โฆ ๐ โฆ ๐น (๐ ) (๐, ๐ฅ 0) = ๐(๐(๐, ๐ (๐ฅ 0))). Thus the diagram indeed commutes and it follows that the isomorphism ๐๐ ,๐ is natural in ๐ and ๐ . We conclude that ๐น : Set → [๐บ, Set] is left adjoint to ๐ : [๐บ, Set] → Set. Now define ๐ป : Set → [๐บ, Set] as follows. For ๐ ∈ Set set ๐ป (๐ ) = ๐ ๐บ ∈ [๐บ, Set], the set of functions ๐บ → ๐, with ๐บ-action given by (๐ · ๐ ) (โ) = ๐ (โ๐) for all ๐, โ ∈ ๐บ, ๐ ∈ ๐ ๐บ . It 18 Solutions by positrón0802 2.1 Definition and examples is clear that ๐ · ๐ = ๐ for all ๐ ∈ ๐ ๐บ . Moreover, if ๐, ๐ 0, โ ∈ ๐บ then (๐ 0๐ · ๐ ) (โ) = ๐ (โ๐ 0๐) = (๐ · ๐ )(โ๐ 0) = (๐ 0 · (๐ · ๐ )) (โ), so this indeed defines a ๐บ-action. A map ๐ : ๐ → ๐ of sets induces a map ๐ป (๐) : ๐ ๐บ → ๐ ๐บ given by [๐ป (๐) (๐ )] (โ) = ๐ โฆ ๐ (โ). This is a morphism of ๐บ-sets by the equation [๐ป (๐) (๐ · ๐ )] (โ) = ๐ โฆ (๐ · ๐ ) (โ) = ๐ โฆ ๐ (โ๐) = [๐ · ๐ป (๐) (๐ )] (โ), so ๐ป : Set → [๐บ, Set] is a functor. We claim ๐ a ๐ป, so that we seek a natural bijection ๐๐ ,๐ : Set(๐ (๐ ), ๐ ) → [๐บ, Set] (๐ , ๐ ๐บ ). For ๐ ∈ Set(๐ (๐ ), ๐ ) define ๐๐ ,๐ (๐ ) : ๐ → ๐ ๐บ by ๐๐ ,๐ (๐ ) (๐ฆ) (๐) = ๐ (๐ · ๐ฆ) for all ๐ฆ ∈ ๐ , ๐ ∈ ๐บ . If ๐ ∈ Set(๐ (๐ ), ๐ ), ๐ฆ ∈ ๐ and ๐, โ ∈ ๐บ we have [๐๐ ,๐ (๐ ) (๐ · ๐ฆ)] (โ) = ๐ (โ๐ · ๐ฆ) = [๐ · ๐๐ ,๐ (๐ ) (๐ฆ)] (โ), so ๐๐ ,๐ (๐ ) is a morphism of ๐บ-sets. Now define a map ๐๐ ,๐ : [๐บ, Set] (๐ , ๐ ๐บ ) → Set(๐ (๐ ), ๐ ) by ๐๐ ,๐ ( ๐e) (๐ฆ) = ๐e(๐ฆ) (๐) for all ๐e ∈ [๐บ, Set] (๐ , ๐ ๐บ ) and ๐ฆ ∈ ๐ . It is straightforward to show that ๐๐ ,๐ = ๐๐−1,๐ , so ๐๐ ,๐ is a bijection. It remains to show that this bijection is natural. For ๐, ๐ 0 ∈ Set and ๐ , ๐ 0 ∈ [๐บ, Set], and morphisms ๐ : ๐ 0 → ๐ , ๐ : ๐ → ๐ 0, we need to show that the diagram Set(๐ (๐ ), ๐ ) ๐๐ ,๐ [๐บ,Set] ( ๐ ,๐ป (๐)) Set(๐ (๐ ),๐) Set(๐ (๐ 0), ๐ 0) [๐บ, Set] (๐ , ๐ ๐บ ) ๐๐ 0,๐ 0 [๐บ, Set] (๐ 0, (๐ 0)๐บ ) commutes. Given ๐ ∈ Set(๐ (๐ ), ๐ ), ๐ฆ 0 ∈ ๐ 0 and ๐ ∈ ๐บ we compute ๐๐ 0,๐ 0 (๐ โฆ ๐ โฆ ๐ (๐ )) (๐ฆ 0) (๐) = (๐ โฆ ๐ โฆ ๐ (๐ )) (๐ · ๐ฆ 0) = ๐ (๐(๐ (๐ · ๐ฆ 0))) and (๐ป (๐) โฆ๐๐ ,๐ (๐) โฆ ๐ ) (๐ฆ 0) (๐) = ๐ป (๐) [๐๐ ,๐ (๐) (๐ (๐ฆ 0))] (๐) = ๐ (๐๐ ,๐ (๐) (๐ (๐ฆ 0)) (๐)) = ๐ (๐(๐ · ๐ (๐ฆ 0))). Both results agree since ๐ is a morphism of ๐บ-sets, so that ๐ (๐ · ๐ฆ 0) = ๐ · ๐ (๐ฆ 0). It follows that the bijection ๐๐ ,๐ is natural in ๐ and ๐ , and we conclude that ๐ a ๐ป . Consider now the functor Δ : Set → [๐บ, Set] that equips a set with the trivial left ๐บ-action. That is, if ๐ ∈ Set then Δ๐ is the ๐บ-set such that ๐ · ๐ฅ = ๐ฅ for all ๐ฅ ∈ ๐ . We have two interesting functors [๐บ, Set] → Set as follows. We have the functor (−)๐บ : [๐บ, Set] → Set sending a ๐บset ๐ to its ๐บ-fixed point subset ๐ ๐บ = {๐ฅ ∈ ๐ | ๐ · ๐ฅ = ๐ฅ for all ๐ ∈ ๐บ }, and the functor (−)/๐บ : [๐บ, Set] → Set sending a ๐บ-set ๐ to the set ๐ /๐บ = {๐บ · ๐ฅ | ๐ฅ ∈ ๐ } of orbits of ๐ under ๐บ . Then we have adjunctions (−)/๐บ a Δ a (−)๐บ . Similarly as in the previous adjunctions, these can be proved directly. Alternatively, we are going to deduce them on Exercise 6.1.6. (b) Definitions analogous to those in (a) give functors [๐บ, Vect๐ ] → Vect๐ left and right adjoint to the forgetful functor ๐ : Vect๐ → [๐บ, Vect๐ ], with the ๐-vector space structure on 19 Solutions by positrón0802 2.1 Definition and examples ๐บ × ๐ given by ๐ผ (๐, ๐ฅ) = (๐, ๐ผ๐ฅ), and the vector space structure on the set of functions ๐บ → ๐ given by (๐ผ ๐ ) (๐) = ๐ผ ๐ (๐). We also have a functor Δ : Vect๐ → [๐บ, Vect๐ ] endowing a ๐-vector space with the trivial left ๐บ-action. For a ๐บ-vector space ๐ , the ๐บ-fixed point subset ๐ ๐บ is a ๐-vector subspace of ๐ , so we have a functor (−)๐บ : [๐บ, Vect๐ ] → Vect๐ which is right adjoint to Δ. However, we no longer have a functor (−)/๐บ, as there is no natural way of endowing the set of orbits ๐ /๐บ of a ๐บ-vector space ๐ under ๐บ with a ๐-vector space structure. Exercise 2.1.17. More precisely, the functor Δ : Set → [๐ช(๐ ) op, Set] is given as follows. If ๐ด ∈ Set then Δ๐ด(๐ ) = ๐ด for all ๐ ∈ ๐ช(๐ ), and Δ๐ด(๐ ) = 1๐ด for all morphisms ๐ : ๐ → ๐ in ๐ช(๐ ) op . If ๐ : ๐ด → ๐ด 0 is a map of sets, Δ(๐)๐ = ๐ for all ๐ ⊂ ๐ . For ๐ ⊂ ๐ ∈ ๐ช(๐ ) let ๐๐ ,๐ : ๐ → ๐ denote the unique arrow ๐ → ๐ in ๐ช(๐ ) op . First we find a right adjoint Γ to Δ. For ๐น ∈ [๐ช(๐ ) op, Set] let Γ(๐น ) = ๐น (๐ ), and for a natural transformation ๐ : ๐น ⇒ ๐บ of functors ๐ช(๐ ) op → Set let Γ(๐ผ) = ๐๐ : ๐น (๐ ) → ๐บ (๐ ). Then Γ defines a functor [๐ช(๐ ) op, Set] → Set. To show that Δ a Γ we need bijections ๐๐ด,๐น : [๐ช(๐ ) op, Set] (Δ๐ด, ๐น ) → Set(๐ด, ๐น (๐ )) natural in ๐ด ∈ Set and ๐น ∈ [๐ช(๐ ) op, Set]. Given a natural transformation ๐ : Δ๐ด ⇒ ๐น, let ๐๐ด,๐น (๐) = ๐๐ : ๐ด → ๐น (๐ ). Conversely, define ๐๐ด,๐น : Set(๐ด, ๐น (๐ )) → [๐ช(๐ ) op, Set] (Δ๐ด, ๐น ) as follows. If ๐ : ๐ด → ๐น (๐ ) is a map of sets and ๐ ∈ ๐ช(๐ ), let ๐๐ด,๐น (๐ )๐ : ๐ด → ๐น (๐ ) be the composition ๐น (๐ ๐ ,๐ ) โฆ ๐ . That ๐๐ด,๐น (๐ ) defines a natural transformation Δ๐ด ⇒ ๐น follows from the equation ๐น (๐๐ ,๐ ) โฆ๐๐ด,๐น (๐ )๐ = ๐๐ด,๐น (๐ )๐ for all ๐ ⊂ ๐ , which holds by functoriality of ๐น . It is clear that ๐๐ด,๐น and ๐๐ด,๐น are inverses to each other. It remains to show that the bijection ๐๐ด,๐น is natural, that is, that given ๐ด, ๐ด 0 ∈ Set, ๐น, ๐น 0 ∈ [๐ช(๐ ) op, Set], and morphisms ๐ : ๐ด 0 → ๐ด, ๐ผ : ๐น ⇒ ๐น 0, the diagram [๐ช(๐ ) op, Set] (Δ๐ด, ๐น ) ๐๐ด,๐น Set(๐ด, ๐น (๐ )) [๐ช (๐ ) op ,Set] (Δ๐ ,๐ผ) Set(๐ ,Γ (๐ผ)) [๐ช(๐ ) op, Set] (Δ๐ด 0, ๐น 0) ๐๐ด0,๐น 0 Set(๐ด 0, ๐น 0 (๐ )) commutes. This is a straightforward computation: for a natural transformation ๐ : Δ๐ด ⇒ ๐น we have Γ(๐ผ) โฆ ๐๐ด,๐น (๐) โฆ ๐ = ๐ผ๐ โฆ ๐๐ โฆ ๐ = (๐ผ โฆ ๐ โฆ Δ๐ )๐ = ๐๐ด0,๐น 0 (๐ผ โฆ ๐ โฆ Δ๐ ). It follows that the bijections ๐๐ด,๐น define an adjunction Δ a Γ. We now find a right adjoint ∇ to Γ. Given ๐ด ∈ Set let ∇๐ด ∈ [๐ช(๐ ) op, Set] be defined as follows. On objects define ∇๐ด by ∇๐ด(๐ ) = ๐ด and ∇๐ด(๐ ) = {∗} for all ๐ ≠ ๐ . On morphisms, let ∇๐ด(1๐ ) = 1๐ด, and if ๐ ≠ ๐ ⊂ ๐ and ๐๐ ,๐ is the unique morphism ๐ → ๐ on ๐ช(๐ ) op, let ∇๐ด(๐๐ ,๐ ) : ∇๐ด(๐ ) → {∗} be the unique possible map. Then ∇๐ด is a functor ๐ช(๐ ) op → Set. For a map of sets ๐ : ๐ด → ๐ด 0, let ∇๐ : ∇๐ด ⇒ ∇๐ด 0 be the natural transformation given by (∇๐ )๐ = ๐ , 20 Solutions by positrón0802 2.1 Definition and examples and (∇๐ )๐ = 1 {∗} if ๐ ≠ ๐ . Then ∇ is a functor Set → [๐ช(๐ ) op, Set]. To show it is right adjoint to Γ we seek bijections ๐ ๐น,๐ด : Set(๐น (๐ ), ๐ด) → [๐ช(๐ ) op, Set] (๐น, ∇๐ด) natural in ๐น ∈ [๐ช(๐ ) op, Set] and ๐ด ∈ Set. For ๐ : ๐น (๐ ) → ๐ด let ๐ ๐น,๐ด (๐ ) : ๐น ⇒ ∇๐ด be given by ๐ ๐น,๐ด (๐ )๐ = ๐ and ๐ ๐น,๐ด (๐ )๐ : ๐น (๐ ) → {∗} the unique map if ๐ ≠ ๐ . Then ๐ ๐น,๐ด (๐ ) is clearly natural. Conversely, if we are given a natural transformation ๐ : ๐น ⇒ ∇๐ด let ๐ ๐น,๐ด (๐) = ๐๐ : ๐น (๐ ) → ๐ด, so that we have a function ๐ ๐น,๐ด : [๐ช(๐ ) op, Set] (๐น, ∇๐ด) → Set(๐น (๐ ), ๐ด). Then it is clear that −1 , so ๐ ๐ ๐น,๐ด = ๐ ๐น,๐ด ๐น,๐ด is a bijection. It remains to show naturality. Given ๐ด, ๐ด 0 ∈ Set, ๐น, ๐น 0 ∈ [๐ช(๐ ) op, Set], and morphisms ๐ : ๐ด → ๐ด 0, ๐ผ : ๐น 0 ⇒ ๐น, we need to show commutativity of diagram Set(๐น (๐ ), ๐ด) ๐ ๐น ,๐ด [๐ช(๐ ) op, Set] (๐น, ∇๐ด) [๐ช (๐ ) op ,Set] (๐ผ,∇๐ ) Set(Γ (๐ผ),๐ ) Set(๐น 0 (๐ ), ๐ด 0) ๐ ๐น 0,๐ด0 [๐ช(๐ ) op, Set] (๐น 0, ∇๐ด 0) . For ๐ : ๐น (๐ ) → ๐ด, we want to show that (∇๐ โฆ ๐ ๐น,๐ด (๐) โฆ ๐ผ)๐ = [๐ ๐น 0,๐ด0 (๐ โฆ ๐ โฆ ๐ผ๐ )]๐ for all ๐ ⊂ ๐ . For ๐ ≠ ๐ this is clear as there is only one map ๐น 0 (๐ ) → {∗}, and for ๐ = ๐ we have (∇๐ โฆ ๐ ๐น,๐ด (๐) โฆ ๐ผ)๐ = (∇๐ )๐ โฆ ๐ ๐น,๐ด (๐)๐ โฆ ๐ผ๐ = ๐ โฆ ๐ โฆ ๐ผ๐ = [๐ ๐น 0,๐ด0 (๐ โฆ ๐ โฆ ๐ผ๐ )]๐ . It follows that ๐ ๐น,๐ด is a bijection natural in ๐ด and ๐น . We conclude that Γ a ∇. Now we construct a left adjoint Π to Δ. Given ๐น ∈ [๐ช(๐ ) op, Set] set Π(๐น ) = ๐น (∅), and for a natural transformation ๐ : ๐น ⇒ ๐บ let Π(๐ผ) = ๐ผ ∅ : ๐น (∅) → ๐บ (∅). This defines a functor Π : [๐ช(๐ ) op, Set] → Set. To show that Π a Δ, define, for all ๐ด ∈ Set and ๐น ∈ [๐ช(๐ ) op, Set], ๐ ๐น,๐ด : Set(๐น (∅), ๐ด) → [๐ช(๐ ) op, Set] (๐น, Δ๐ด) by ๐ ๐น,๐ด (๐ )๐ = ๐ โฆ ๐น (๐๐ ,∅ ) : ๐น (๐ ) → ๐ด for all ๐ ∈ Set(๐น (∅), ๐ด) and ๐ ⊂ ๐ . Then ๐ ๐น,๐ด (๐ ) is a natural transformation ๐น ⇒ Δ๐ด since given ๐ ⊂ ๐ we have ๐ ๐น,๐ด (๐ )๐ = ๐ ๐น,๐ด (๐ )๐ โฆ ๐น (๐๐ ,๐ ) by functoriality of ๐น . On the other direction define ๐ ๐น,๐ด : [๐ช(๐ ) op, Set] (๐น, Δ๐ด) → Set(๐น (∅), ๐ด) by −1 , so ๐ ๐น,๐ด (๐ผ) = ๐ผ ∅ : ๐น (∅) → ๐ด for all ๐ผ : ๐น ⇒ Δ๐ด. It is straightforward to check that ๐ ๐น,๐ด = ๐ ๐น,๐ด ๐ ๐น,๐ด is a bijection. To prove naturality of this bijection let ๐ด, ๐ด 0 ∈ Set, ๐น, ๐น 0 ∈ [๐ช(๐ ) op, Set], and morphisms ๐ : ๐ด → ๐ด 0, ๐ผ : ๐น 0 ⇒ ๐น, and consider the diagram Set(๐น (∅), ๐ด) ๐ ๐น ,๐ด [๐ช(๐ ) op, Set] (๐น, Δ๐ด) [๐ช (๐ ) op ,Set] (๐ผ,Δ๐ ) Set(Π (๐ผ),๐ ) Set(๐น 0 (∅), ๐ด 0) ๐ ๐น 0,๐ด0 [๐ช(๐ ) op, Set] (๐น 0, Δ๐ด 0) . 21 Solutions by positrón0802 2.2 Adjunctions via units and counits To prove it is commutative let ๐ : ๐น (∅) → ๐ด and ๐ ⊂ ๐ . Then ๐ ๐น 0,๐ด0 (๐ โฆ ๐ โฆ โฆ๐ผ ∅ )๐ = ๐ โฆ ๐ โฆ ๐ผ ∅ โฆ ๐น 0 (๐๐ ,∅ ) = ๐ โฆ ๐ โฆ ๐น (๐๐ ,∅ ) โฆ ๐ผ๐ = (Δ๐ โฆ ๐ ๐น,๐ด (๐) โฆ ๐ผ)๐ , the second equality by naturality of ๐ผ . It follows that ๐ ๐น,๐ด is a bijection natural in ๐ด and ๐น and hence Π a Δ. At last, we find a left adjoint Λ to Π. For ๐ด ∈ Set let Λ(๐ด) be the functor ๐ช(๐ ) op → Set given as follows. On objects, Λ(๐ด) (∅) = ๐ด and Λ(๐ด) (๐ ) = ∅ if ๐ ≠ ∅; on morphisms, let Λ(๐ด)(1 ∅ ) = 1๐ด and let Λ(๐ด) (๐๐ ,๐ ) be the empty function if ๐ ≠ ∅. If ๐ : ๐ด → ๐ด 0 is a map of sets let Λ(๐ ) : Λ(๐ด) ⇒ Λ(๐ด 0) have ∅-component Λ(๐ )∅ = ๐ , and ๐ -component Λ(๐ )๐ the empty function if ๐ ≠ ∅. The naturality of Λ(๐ ) is automatic. Then Λ is a functor Set → [๐ช(๐ ) op, Set]. We now find natural bijections ๐๐ด,๐น : [๐ช(๐ ) op, Set] (Λ(๐ด), ๐น ) → Set(๐ด, ๐น (∅)). Define ๐๐ด,๐น by ๐๐ด,๐น (๐ผ) = ๐ผ ∅ for all ๐ผ : Λ(๐ด) ⇒ ๐น . On the other hand, let ๐ ๐น,๐ด : Set(๐ด, ๐น (∅)) → [๐ช(๐ ) op, Set] (Λ(๐ด), ๐น ) send ๐ : ๐ด → ๐น (∅) to the transformation ๐ ๐น,๐ด (๐ ) : Λ(๐ด) ⇒ ๐น with ∅component ๐ ๐น,๐ด (๐ )∅ = ๐ and ๐ -component ๐ ๐น,๐ด (๐ )๐ the empty function if ๐ ≠ ∅. Then ๐ ๐น,๐ด (๐ ) −1 . is of course natural, so ๐ ๐น,๐ด is well-defined, and moreover ๐ ๐น,๐ด = ๐ ๐น,๐ด 0 0 op 0 Given ๐ด, ๐ด ∈ Set, ๐น, ๐น ∈ [๐ช(๐ ) , Set], and morphisms ๐ : ๐ด → ๐ด, ๐ผ : ๐น ⇒ ๐น 0, consider the diagram [๐ช(๐ ) op, Set] (Λ(๐ด), ๐น ) ๐๐ด,๐น [๐ช (๐ ) op ,Set] (Λ(๐ ),๐ผ) Set(๐ด, ๐น (∅)) Set(๐ ,Π (๐ผ)) [๐ช(๐ ) op, Set] (Λ(๐ด 0), ๐น 0) ๐๐ด0,๐น 0 Set(๐ด 0, ๐น 0 (∅)) . If ๐ : Δ๐ด ⇒ ๐น is a natural transformation then Π(๐ผ) โฆ ๐๐ด,๐น (๐) โฆ ๐ = ๐ผ ∅ โฆ ๐ ∅ โฆ Λ(๐ )∅ = (๐ผ โฆ ๐ โฆ Λ(๐ ))∅ = ๐๐ด0,๐น 0 (๐ผ โฆ ๐ โฆ Λ(๐ )), so the diagram commutes. It follows that ๐ is a natural isomorphism, so that Λ a Π. This completes the chain of adjoint functors Λ a Π a Δ a Γ a ∇. 2.2 Adjunctions via units and counits Exercise 2.2.10. First assume (a) holds. Given ๐ ∈ ๐ด, then ๐ (๐) ∈ ๐ต and ๐ (๐) ≤ ๐ (๐), so ๐ ≤ ๐๐ (๐) by assumption. Similarly, given ๐ ∈ ๐ต then ๐(๐) ≤ ๐(๐) implies ๐ ๐(๐) ≤ ๐. Now assume (b) holds. Let ๐ ∈ ๐ด and ๐ ∈ ๐ต. First suppose that ๐ (๐) ≤ ๐. We have ๐๐ (๐) ≤ ๐(๐) as ๐ is order-preserving, so ๐ ≤ ๐๐ (๐) ≤ ๐(๐) by assumption. Similarly, suppose that ๐ ≤ ๐(๐). We have ๐ (๐) ≤ ๐ ๐(๐) since ๐ is order preserving, so ๐ (๐) ≤ ๐ ๐(๐) ≤ ๐ by assumption. 22 Solutions by positrón0802 2.2 Adjunctions via units and counits Exercise 2.2.11. (a) Let ๐ด ∈ Fix(๐บ๐น ). Then ๐๐ด is an isomorphism, so ๐น (๐๐ด ) is an isomorphism. Since ๐ ๐น (๐ด) โฆ ๐น (๐๐ด ) = 1๐น (๐ด) it follows that ๐ ๐น (๐ด) is also an isomorphism. Thus ๐น restricts to a well-defined functor ๐น 0 : Fix(๐บ๐น ) → Fix(๐น๐บ). Similarly, ๐บ restricts to a well-defined functor ๐บ 0 : Fix(๐น๐บ) → Fix(๐บ๐น ). Furthermore, ๐ 0 : 1Fix(๐บ๐น ) ⇒ ๐บ 0๐น 0 given by ๐๐ด0 = ๐๐ด and ๐ 0 : ๐น 0๐บ 0 ⇒ 1Fix(๐น๐บ) given by ๐๐ต0 = ๐๐ต are well-defined. Since by definition ๐ 0 and ๐ 0 are isomorphisms, it follows that (๐น 0, ๐บ 0, ๐ 0, ๐ 0) is an equivalence. (b) Let ๐ : Top → Set be the forgetful functor. Recall (from Example 2.1.5) that ๐ has a left adjoint ๐ท : Set → Top which equips a set ๐ with the discrete topology. The unit ๐ of the adjunction ๐ท a ๐ is given at a set ๐ by the identity function ๐๐ = 1๐ : ๐ → ๐ ๐ท (๐), so that Fix(๐ ๐ท) = Set. On the other hand, the counit ๐ at a topological space ๐ is the map ๐๐ : ๐ท๐ (๐ ) → ๐ whose underlying set function is the identity function. Thus ๐๐ is a homeomorphism if and only if ๐ is discrete, and therefore Fix(๐ท๐ ) is the full subcategory Topd of Top spanned by the discrete topological spaces. It follows that ๐ and ๐ท induce an equivalence Topd ' Set which associates to a discrete space its underlying set, and equips a set with the discrete topology. Similarly, ๐ : Top → Set has a right adjoint ๐ผ : Set → Top which equips a set with the indiscrete topology. Its counit is given at a set ๐ by the identity function 1๐ = ๐ ๐ผ (๐) → ๐, which is always a bijection. Its unit is given at a space ๐ by the map ๐ → ๐ผ๐ (๐ ) whose underlying set function is the identity function; it is a homeomorphism if and only if ๐ is indiscrete. It follows that ๐ and ๐ผ induce an equivalence Topi ' Set, where Topi is the full subcategory of Top spanned by the indiscrete topological spaces. Consider now the forgetful functor ๐ : Ab → Grp, whose left adjoint ๐น : Grp → Ab sends a group to its the abelianisation (see Example 2.1.3 (c)). Since ๐น is the identity on Ab ⊂ Grp (or more precisely, can be taken in this way), the counit of ๐น a ๐ at an abelian group ๐ด is the identity morphism 1๐ด = ๐น๐ (๐ด) → ๐ด, so it is always an isomorphism. On the other hand, the counit of ๐น a ๐ at a group ๐บ is the canonical quotient ๐บ → ๐ ๐น (๐บ) onto its abelianisation, so it is an isomorphism if and only if ๐บ is abelian. It follows that the equivalence induced by the adjunction ๐น a ๐ is just the identity Ab ' Ab. Exercise 2.2.12. (a) Let ๐น a ๐บ, ๐น : ๐ → โฌ, be an adjunction with counit ๐. Let ๐๐ด,๐ต : ๐(๐ด, ๐บ (๐ต)) → โฌ(๐น (๐ด), ๐ต) be the structural isomorphisms of the adjunction. Then ๐๐ต = ๐๐บ (๐ต),๐ต (1๐บ (๐ต) ) for all ๐ต ∈ โฌ, by definition. By naturality of ๐ we have ๐ โฆ ๐๐ต = ๐ โฆ ๐๐บ (๐ต),๐ต (1๐บ (๐ต) ) = ๐๐บ (๐ต),๐ต0 โฆ ๐บ (๐)∗ (1๐บ (๐ต) ) = ๐๐บ (๐ต),๐ต0 โฆ ๐บ (๐) for all ๐ : ๐ต → ๐ต 0, that is, the composite map ๐๐บ (๐ต),๐ต0 ๐บ (−) โฌ(๐ต, ๐ต 0) −−−−→ ๐(๐บ (๐ต), ๐บ (๐ต 0)) −−−−−−→ โฌ(๐น๐บ (๐ต), ๐ต 0) is given by ๐ โฆ→ ๐ โฆ ๐๐ต . 23 Solutions by positrón0802 2.2 Adjunctions via units and counits First suppose that ๐บ is full and faithful. Then the above composition is bijective for all ๐ต, ๐ต 0 ∈ ๐ต. In particular, given ๐ต ∈ โฌ then for ๐ต 0 = ๐น๐บ (๐ต) there exists a map โ : ๐ต → ๐น๐บ (๐ต) such that โโฆ๐๐ต = 1๐น๐บ (๐ต) . Moreover, since ๐บ is full and faithful then ๐๐ต has a right inverse if and only if ๐บ (๐๐ต ) has a right inverse, and the latter holds by the triangle identity ๐บ (๐๐ต ) โฆ๐๐บ (๐ต) = 1๐บ (๐ต) . Thus ๐๐ต has both a left and a right inverse, which then must coincide. It follows that ๐๐ต is an isomorphism. Conversely, suppose that ๐ is a natural isomorphism. Let ๐ต, ๐ต 0 ∈ โฌ. Then the composition written above is bijective with inverse given by ๐ โฆ→ ๐ โฆ ๐๐ต−1 for all ๐ : ๐บ๐น (๐ต) → ๐ต 0 . Since ๐๐บ (๐ต),๐ต0 is a bijection, it follows that ๐บ (−) is a bijection. Thus ๐บ is full and faithful. (b) We want to see which of the right adjoints given in the examples are full and faithful. • Forgetful functors that forget “structure”, as Vect๐ → Set, Grp → Set, Grp → Mon, Top → Set, etc. They are all right adjoint to “free” functors. These forgetful functors are not full, so these adjunctions are not reflections. • The forgetful functor ๐ : Ab → Grp, which forgets a “property”. Its right adjoint ๐น : Grp → Ab is given by the abelianisation. The functor ๐ is full and faithful, so the adjunction ๐น a ๐ is a reflection. • The forgetful functor ๐ : Top → Set has a right adjoint ๐ผ : Set → Top endowing a set with the indiscrete topology. The functor ๐ผ is full and faithful, so ๐ a ๐ผ is a reflection. • For ๐ต a set, the functor (−) ๐ต : Set → Set is right adjoint to − × ๐ต : Set → Set. If ๐ต = ∅, the functor (−) ๐ต sends every set to the empty set, so is clearly not faithful. If ๐ต has precisely one element, (−) ๐ต is the identity functor, hence full and faithful, so that (− × ๐ต) a (−) ๐ต is a reflection. Finally, if ๐ต has more than one element, the functor (−) ๐ต is not full. For instance, if ๐ต has precisely two elements, then (−) ๐ต is the functor ๐ถ โฆ→ ๐ถ × ๐ถ, and for sets ๐ถ and ๐ท, in general not all maps ๐ถ × ๐ถ → ๐ท × ๐ท have the form ๐ × ๐ for some ๐ : ๐ถ → ๐ท. • Let ๐ be a category. If ๐ has an initial object, then the unique functor ๐ → 1 has a left adjoint (given by the initial object). The right adjoint ๐ → 1 is full and faithful if and only if ๐ has precisely one arrow between every two objects, so this adjunction is a reflection if and only if this is the case. If ๐ has a terminal object, a functor 1 → ๐ given by a terminal object is right adjoint to ๐ → 1. Such a functor 1 → ๐ is full and faithful, so this adjunction is a reflection. Exercise 2.2.13. (a) Let โ : ๐ซ(๐พ) → ๐ซ(๐ฟ), ๐ โฆ→ ๐ (๐ ), be the image functor. (It is a functor being order-preserving.) The equations ๐ ⊂ ๐ −1 (๐ (๐ )) for all ๐ ∈ ๐ซ(๐พ) and ๐ (๐ −1 (๐)) ⊂ ๐ for all ๐ ∈ ๐ซ(๐ฟ) (which hold for any set function) translate into ๐ ≤ ๐ ∗โ(๐ ) for all ๐ ∈ ๐ซ(๐พ), โ๐ ∗ (๐) ≤ ๐ for all ๐ ∈ ๐ซ(๐ฟ). This means precisely that โ a ๐ ∗ (see Example 2.2.7, c.f. Exercise 2.2.10). 24 Solutions by positrón0802 2.2 Adjunctions via units and counits Now let us find a right adjoint ๐ : ๐ซ(๐พ) → ๐ซ(๐ฟ) to ๐ ∗ . For ๐ ⊂ ๐พ, let ๐(๐ ) be the largest subset ๐ of ๐ฟ such that ๐ −1 (๐) ⊂ ๐ . More explicitly, ๐ is given as follows. Let ๐ = ๐ฟ \ ๐ (๐พ). Given ๐ ⊂ ๐พ let ๐ 0 ⊂ ๐ be the subset consisting of points ๐ก ∈ ๐ such that there exists some ๐ข ∉ ๐ with ๐ (๐ข) = ๐ (๐ก). Then ๐(๐ ) = ๐ (๐ \ ๐ 0) ∪ ๐. Suppose ๐1 ⊂ ๐2 ⊂ ๐พ . If ๐ก ∈ ๐1 \ ๐10, then there is no ๐ข ∉ ๐1 such that ๐ (๐ข) = ๐ (๐ก). In particular there is no ๐ข ∉ ๐2 such that ๐ (๐ข) = ๐ (๐ก), and hence ๐ก ∈ ๐2 \ ๐20 . Thus ๐ is order-preserving, i.e. a functor ๐ซ(๐พ) → ๐ซ(๐ฟ). Now, proving ๐ ∗ a ๐ amounts to showing the equations ๐ ⊂ ๐(๐ −1 (๐)) for all ๐ ∈ ๐ซ(๐ฟ), ๐ −1 (๐(๐ )) ⊂ ๐ for all ๐ ∈ ๐ซ(๐พ). Given ๐ ∈ ๐ซ(๐ฟ), note that (๐ −1 (๐)) 0 = ∅, for if ๐ก ∈ ๐ −1 (๐) and ๐ข ∈ ๐พ are such that ๐ (๐ข) = ๐ (๐ก), then ๐ข ∈ ๐ −1 (๐). Thus ๐(๐ −1 (๐)) = ๐ (๐ −1 (๐)) ∪ ๐, and this set contains ๐. Furthermore, if ๐ ⊂ ๐พ then ๐ −1 (๐(๐ )) = ๐ −1 (๐ (๐ \ ๐ 0) ∩ ๐) = ๐ −1 (๐ (๐ \ ๐ 0)) ⊂ ๐ \ ๐ 0 ⊂ ๐ . Hence the above equations hold, and we conclude that ๐ ∗ a ๐. (b) Let us write our definitions of ๐ and โ above for the particular case ๐ ∗ : ๐ซ(๐ ) → ๐ซ(๐ ×๐ ). Given ๐ ⊂ ๐ × ๐ we have โ(๐ ) = ๐ (๐ ). This is the set of points ๐ฅ ∈ ๐ such that (๐ฅ, ๐ฆ) ∈ ๐ for some ๐ฆ ∈ ๐ , i.e., the set of points ๐ฅ ∈ ๐ satisfying ๐ (๐ฅ) = ∃๐ฆ ∈ ๐ such that (๐ฅ, ๐ฆ) ∈ ๐ . The unit of the adjunction โ a ๐ ∗ at a subset ๐ ⊂ ๐ is the inclusion ๐ ⊂ ๐ −1 (๐ (๐ )). This can be written as a logical implication as (∀๐ ⊂ ๐ × ๐ ) (∀(๐ฅ, ๐ฆ) ∈ ๐ × ๐ ) ((๐ฅ, ๐ฆ) ∈ ๐ =⇒ (∃๐ฆ 0 ∈ ๐ ) ((๐ฅ, ๐ฆ 0) ∈ ๐ )). Similarly, the counit at a subset ๐ ⊂ ๐ is the inclusion ๐ (๐ −1 (๐)) ⊂ ๐. This can be written as (∀๐ ⊂ ๐ ) (∀๐ฅ ∈ ๐ ) ((∃๐ฆ ∈ ๐ ) ((๐ฅ, ๐ฆ) ∈ ๐ −1 (๐)) =⇒ ๐ฅ ∈ ๐). On the other hand ๐(๐ ) = ๐ (๐ \ ๐ 0) ∪ ๐ with ๐ 0 defined as in (a) and ๐ = ๐ \ ๐ (๐ × ๐ ) = ∅. By definition, the set ๐ \ ๐ 0 consists of those pairs (๐ฅ, ๐ฆ) ∈ ๐ such that (๐ฅ, ๐ฆ 0) ∈ ๐ for all ๐ฆ 0 ∈ ๐ . Thus ๐(๐ ) is the set of points satisfying ๐ 0 (๐ฅ) = ∀๐ฆ ∈ ๐ , (๐ฅ, ๐ฆ) ∈ ๐ . The unit of the adjunction ๐ ∗ a ๐ can be written as a logical implication as (∀๐ ⊂ ๐ ) (∀๐ฅ ∈ ๐ ) (๐ฅ ∈ ๐ =⇒ (∀๐ฆ ∈ ๐ ) ((๐ฅ, ๐ฆ) ∈ ๐ −1 (๐))), and the counit as (∀๐ ⊂ ๐ × ๐ ) (∀(๐ฅ, ๐ฆ) ∈ ๐ × ๐ ) ((∀๐ฆ 0 ∈ ๐ ) ((๐ฅ, ๐ฆ 0) ∈ ๐ ) =⇒ (๐ฅ, ๐ฆ) ∈ ๐ ). 25 Solutions by positrón0802 2.2 Adjunctions via units and counits Exercise 2.2.14. Let ๐ : 1๐ ⇒ ๐บ๐น and ๐ : ๐น๐บ ⇒ 1โฌ be the unit and counit of the adjunction, respectively. We will show that they induce natural transformations ๐ ∗ : 1 [๐,๐ฎ] ⇒ ๐น ∗๐บ ∗ and ๐ ∗ : ๐บ ∗ ๐น ∗ ⇒ 1 [โฌ,๐ฎ] which will be the unit and counit, respectively, of the adjunction ๐บ ∗ a ๐น ∗ . For all ๐ป ∈ [๐, ๐ฎ] let ๐๐ป∗ : ๐ป ⇒ ๐ป๐บ๐น be the transformation with components (๐๐ป∗ )๐ด = ๐ป (๐๐ด ) : ๐ป (๐ด) → ๐ป๐บ๐น (๐ด) for all ๐ด ∈ ๐. Since ๐ is natural, so is ๐๐ป∗ . Moreover, given ๐ป, ๐ป 0 ∈ [๐, ๐ฎ] and ๐ผ : ๐ป ⇒ ๐ป 0 consider the diagram ∗ ๐๐ป ๐ป ๐ป๐บ๐น ๐ผ ๐ผ๐บ๐น ๐ป0 ∗ ๐๐ป 0 ๐ป 0๐บ๐น . For all ๐ด ∈ ๐ we have (๐ผ๐บ๐น โฆ ๐๐ป∗ )๐ด = ๐ผ๐บ๐น (๐ด) โฆ ๐ป (๐๐ด ) = ๐ป 0 (๐๐ด ) โฆ ๐ผ๐ด = (๐๐ป∗ 0 โฆ ๐ผ)๐ด by naturality of ๐ผ, so the diagram commutes. Thus ๐ ∗ : 1 [๐,๐ฎ] ⇒ ๐น ∗๐บ ∗ is indeed natural. Similarly, for all ๐พ ∈ [โฌ, ๐ฎ] let ๐๐พ∗ = : ๐พ๐น๐บ ⇒ ๐พ have components (๐๐พ∗ )๐ต = ๐พ (๐๐ต ) : ๐พ๐น๐บ (๐ต) → ๐พ (๐ต) for all ๐ต ∈ โฌ. Then ๐ ∗ is a well-defined natural transformation ๐บ ∗ ๐น ∗ ⇒ 1 [โฌ,๐ฎ] . To complete the proof it suffices to show that ๐ ∗ and ๐ ∗ satisfy the triangle identities ๐บ∗ ๐บ ∗๐ ∗ 1๐บ ∗ ๐บ ∗ ๐น ∗๐บ ∗ ๐∗๐น ∗ ๐น∗ ๐ ∗๐บ ∗ 1๐น ∗ ๐บ∗ , ๐น ∗๐บ ∗ ๐น ∗ ๐น ∗๐ ∗ ๐น∗ . First consider the left-hand side triangle. Given ๐ป ∈ [๐, ๐ฎ], we want to show that the triangle ∗ ๐บ ๐๐ป ๐ป๐บ ๐ป๐บ๐น๐บ ∗ ๐๐ป๐บ 1๐ป๐บ ๐ป๐บ commutes, and this follows from the computation ∗ (๐๐ป๐บ โฆ ๐๐ป∗ ๐บ)๐ต = ๐ป๐บ (๐๐ต ) โฆ ๐ป (๐๐บ (๐ต) ) = ๐ป (๐บ (๐๐ต ) โฆ ๐๐บ (๐ต) ) = ๐ป (1๐บ (๐ต) ) = 1๐ป๐บ (๐ต) for ๐ต ∈ โฌ, where we use one triangle identity for ๐ and ๐ in the third equality. Similarly, given ๐พ ∈ [โฌ, ๐ฎ] and ๐ด ∈ ๐ we have ∗ (๐๐พ∗ ๐น โฆ ๐๐พ๐น )๐ด = ๐พ (๐ ๐น (๐ด) ) โฆ ๐พ๐น (๐๐ด ) = ๐พ (๐ ๐น (๐ด) โฆ ๐น (๐๐ด )) = ๐พ (1๐น (๐ด) ) = 1๐พ๐น (๐ด) , so that ๐ ∗ and ๐ ∗ satisfy the right-hand side triangle identity above. We conclude that ๐บ ∗ a ๐น ∗ with unit ๐ ∗ and counit ๐ ∗ . 26 Solutions by positrón0802 2.3 2.3 Adjunctions via initial objects Adjunctions via initial objects Exercise 2.3.8. Let ๐ป, ๐พ be groups, ๐น : ๐ป → ๐พ, ๐บ : ๐พ → ๐ป be functors (i.e. group homomorphisms), and suppose that we have an adjunction ๐น a ๐บ . The unit of this adjunction amounts to an element โ ∈ ๐ป such that ๐บ๐น (๐ฅ) = โ๐ฅโ −1 for all ๐ฅ ∈ ๐ป (c.f. Exercise 1.3.30). Similarly, the counit is an element ๐ ∈ ๐พ such that ๐ฆ = ๐๐น๐บ (๐ฆ)๐ −1 for all ๐ฆ ∈ ๐พ . Moreover, the triangle identities say that ๐๐น (โ) = 1 and ๐บ (๐)โ = 1, so that ๐น (โ) = ๐ −1 and ๐บ (๐) = โ −1 . It follows that an adjunction ๐น a ๐บ amounts to group homomorphisms ๐น : ๐ป → ๐พ, ๐บ : ๐ป → ๐พ such that there exist โ ∈ ๐ป, ๐ ∈ ๐พ, with ๐น (โ) = ๐ −1, ๐บ (๐) = โ −1, such that ๐บ๐น is conjugation by โ and ๐น๐บ is conjugation by ๐ −1 . In particular, both ๐บ๐น and ๐น๐บ are isomorphisms, and therefore so are ๐น and ๐บ . Exercise 2.3.9. Dual of Corollary 2.3.7. Let ๐น : ๐ → โฌ be a functor. Then ๐น has a right adjoint if and only if for each ๐ต ∈ โฌ, the category (๐น ⇒ ๐ต) has an initial object. The proof goes through the dual statements of Lemma 2.3.5 and Theorem 2.3.6. Dual of Lemma 2.3.5. Take an adjunction ๐น a ๐บ, ๐น : ๐ → ๐ต, and an object ๐ต ∈ โฌ. Then the counit map ๐๐ต : ๐น๐บ (๐ต) → ๐ต is a terminal object of (๐น ⇒ ๐ต). ๐ Proof of Dual of Lemma 2.3.5. Let (๐ด, โ) ∈ (๐น ⇒ ๐ต). We want a unique arrow (๐ด, โ) → − (๐บ (๐ต), ๐๐ต ), i.e. a unique ๐ : ๐ด → ๐บ (๐ต) such that โ = ๐๐ต โฆ ๐น (๐ ). By Lemma 2.2.4, ๐ = โ is the unique such arrow. Dual of Theorem 2.3.6. Take categories and functors ๐น : ๐ → โฌ, ๐บ : โฌ → ๐. There is a one-toone correspondence between: (a) adjunctions ๐น a ๐บ; (b) natural transformations ๐ : ๐น๐บ ⇒ 1โฌ such that ๐๐ต : ๐น๐บ (๐ต) → ๐ต is terminal in (๐น ⇒ ๐ต) for all ๐ต ∈ โฌ. Proof of Dual of Theorem 2.3.6. It remains to show that if ๐ : ๐น๐บ ⇒ 1โฌ satisfies (b) then there exists a unique natural transformation ๐ : 1๐ ⇒ ๐บ๐น such that (๐, ๐) satisfies the triangle identities. To prove uniqueness assume that ๐ and ๐ 0 are natural transformations 1๐ ⇒ ๐บ๐น such that both (๐, ๐) and (๐ 0, ๐) satisfy the triangle identities. For ๐ด ∈ ๐, the diagram ๐น (๐ด) ๐น (๐๐ด ) ๐น๐บ๐น (๐ด) ๐ ๐น (๐ด) 1๐น (๐ด) ๐น (๐ด) shows that ๐๐ด is a map (๐ด, 1๐น (๐ด) ) → (๐บ๐น (๐ด), ๐ ๐น (๐ด) ) in (๐น ⇒ ๐น (๐ด)). The same can be said about ๐๐ด0 . Since ๐ ๐น (๐ด) is terminal in (๐น ⇒ ๐น (๐ด 0)) it follows that ๐๐ด0 = ๐๐ด . Hence ๐ = ๐ 0 . 27 Solutions by positrón0802 2.3 Adjunctions via initial objects To show existence define ๐ : 1๐ ⇒ ๐บ๐น as follows. Given ๐ด ∈ ๐ let ๐๐ด : ๐ด → ๐บ๐น (๐ด) be the unique map (1๐น (๐ด) , ๐ด) → (๐บ๐น (๐ด), ๐ ๐น (๐ด) ) in (๐น ⇒ ๐น (๐ด)). We need to show that ๐ is indeed natural. Given ๐ด, ๐ด 0 ∈ ๐ and ๐ : ๐ด → ๐ด 0, the diagrams ๐น (๐ด) ๐น (๐๐ด ) ๐น๐บ๐น (๐ด) ๐น๐บ๐น (๐ ) ๐น๐บ๐น (๐ด 0) ๐น (๐ด) ๐น (๐ ) ๐น (๐ด 0) ๐ ๐น (๐ด0 ) ๐น ( ๐ )โฆ๐ ๐น (๐ด) ๐น (๐๐ด0 ) 1๐น (๐ด0 ) ๐น (๐ด 0) ๐น๐บ๐น (๐ด 0) ๐ ๐น (๐ด0 ) ๐น (๐ด 0) ๐น (๐ ) ๐น (๐ ) commute. Thus ๐บ๐น (๐ ) โฆ ๐๐ด and ๐๐ด0 โฆ ๐ are both maps (๐ด, ๐น (๐ )) → (๐บ๐น (๐ด 0), ๐ ๐น (๐ด0) ) in (๐น ⇒ ๐น (๐ด 0)); since ๐ ๐น (๐ด0) is terminal we have ๐บ๐น (๐ ) โฆ๐๐ด = ๐๐ด0 โฆ ๐ , and hence ๐ is natural. By definition (๐, ๐) satisfies the triangle identity given in the triangle of the right-hand side diagram above, it remains to show the other one, i.e. we want to show that the diagram ๐บ (๐ต) ๐๐บ (๐ต) ๐บ๐น๐บ (๐ต) ๐บ (๐๐ต ) 1๐บ (๐ต) ๐บ (๐ต) commutes for all ๐ต ∈ โฌ. This follows by considering the commutative diagrams ๐น๐บ (๐ต) ๐น (๐๐บ (๐ต) ) ๐น๐บ๐น๐บ (๐ต) ๐น๐บ (๐๐ต ) ๐น๐บ (๐ต) ๐น๐บ (๐ต) ๐น (1๐บ (๐ต) ) ๐๐ต ๐๐ต โฆ๐น๐บ (๐๐ต ) ๐น๐บ (๐ต) ๐๐ต ๐๐ต ๐ต, ๐๐ต ๐ต. Indeed, we see that both ๐น๐บ (๐๐ต ) โฆ ๐๐บ (๐ต) and 1๐บ (๐ต) are maps (๐บ (๐ต), ๐๐ต ) → (๐บ (๐ต), ๐๐ต ) in (๐น ⇒ ๐ต), so they are equal. This completes the proof. Proof of Dual of Corollary 2.3.7. By the dual of Lemma 2.3.5 it remains to show that if for each ๐ต ∈ โฌ, (๐น ⇒ ๐ต) has an initial object, then ๐น has a right adjoint. Given ๐ต ∈ โฌ let (๐บ (๐ต), ๐๐ต : ๐น๐บ (๐ต) → ๐ต) be an initial object of (๐น ⇒ ๐ต). For ๐ : ๐ต → ๐ต 0 let ๐บ (๐) be unique map (๐บ (๐ต), ๐ โฆ ๐๐ต ) → (๐บ (๐ต 0), ๐๐ต0 ) in (๐น ⇒ ๐ต 0). Then ๐บ : โฌ → ๐ is a functor, and the definition ensures that ๐ : ๐น๐บ ⇒ 1โฌ is a natural transformation. By the dual of Theorem 2.3.6 it follows that ๐น a ๐บ . Exercise 2.3.10. In view of Remark 2.2.8 the tuple (๐น, ๐บ, ๐, ๐) is not necessarily an adjunction since (๐, ๐) may not satisfy the triangle identities. However we may show that ๐น a ๐บ with unit ๐ (and some counit ๐ 0) using Theorem 2.3.6. More precisely, we will show that ๐ : 1๐ ⇒ ๐บ๐น is such that ๐๐ด : ๐ด → ๐บ๐น (๐ด) is initial in (๐ด ⇒ ๐บ) for all ๐ด ∈ ๐. 28 Solutions by positrón0802 2.3 Adjunctions via initial objects First we define a natural transformation ๐ 0 : ๐น๐บ ⇒ 1โฌ so that the triangle identity ๐บ (๐๐ต0 ) โฆ ๐๐บ (๐ต) = 1๐บ (๐ต) holds. Define ๐๐ต0 : ๐น๐บ (๐ต) → ๐ต by requiring the following diagram to be commutative: ๐๐ต0 ๐น๐บ (๐ต) ๐ต ๐น๐บ (๐๐ต−1 ) ๐๐ต ๐น๐บ๐น๐บ (๐ต) −1 ) ๐น (๐๐บ (๐ต) ๐น๐บ (๐ต) . Then ๐ 0 : ๐น๐บ ⇒ 1โฌ is indeed natural. (Even so, naturality of ๐ 0 will not be used in this proof.) To show the desired triangle identity we want to show that the composite map ๐๐บ (๐ต) −1 ) ๐บ๐น (๐๐บ (๐ต) ๐บ๐น๐บ (๐๐ต−1 ) ๐บ (๐๐ต ) ๐บ (๐ต) −−−−→ ๐บ๐น๐บ (๐ต) −−−−−−−→ ๐บ๐น๐บ๐น๐บ (๐ต) −−−−−−−−→ ๐บ๐น๐บ (๐ต) −−−−→ ๐บ (๐ต) is the identity 1๐บ (๐ต) . First note that by naturality of ๐ we have ๐บ๐น๐บ (๐๐ต−1 ) โฆ ๐๐บ (๐ต) = ๐๐บ๐น๐บ (๐ต) โฆ ๐บ (๐๐ต−1 ), and ๐บ๐น (๐๐บ (๐ต) ) โฆ ๐๐บ (๐ต) = ๐๐บ๐น๐บ (๐ต) โฆ ๐๐บ (๐ต) . Since ๐๐บ (๐ต) is an isomorphism, it follows from the second equation that ๐บ๐น (๐๐บ (๐ต) ) = ๐๐บ๐น๐บ (๐ต) . Thus the composite map above is equal to ๐บ (๐๐ต ) โฆ ๐บ๐น (๐๐บ−1(๐ต) ) โฆ ๐บ๐น๐บ (๐๐ต−1 ) โฆ ๐๐บ (๐ต) = ๐บ (๐๐ต ) โฆ ๐บ๐น (๐๐บ (๐ต) ) −1 โฆ ๐๐บ๐น๐บ (๐ต) โฆ ๐บ (๐๐ต−1 ) = ๐บ (๐๐ต ) โฆ ๐บ๐น (๐๐บ (๐ต) ) −1 โฆ ๐บ๐น (๐๐บ (๐ต) ) โฆ ๐บ (๐๐ต ) −1 = 1๐บ (๐ต) , as desired. Note that since ๐๐บ (๐ต) is an isomorphism, it follows that ๐บ (๐๐ต0 ) = ๐๐บ−1(๐ต) . Now we are ready to show that ๐ : 1๐ ⇒ ๐บ๐น is such that ๐๐ด : ๐ด → ๐บ๐น (๐ด) is initial in (๐ด ⇒ ๐บ) for all ๐ด ∈ ๐. Let ๐ด ∈ ๐ be arbitrary. Given (๐ต, โ) ∈ (๐ด ⇒ ๐บ) we want to show that there is a unique arrow ๐ : (๐น (๐ด), ๐๐ด ) → (๐ต, โ) in (๐ด ⇒ ๐บ). That is, we want a unique ๐ : ๐น (๐ด) → ๐ต such that ๐บ (๐) โฆ ๐๐ด = โ. To show uniqueness assume that ๐ satisfies the given equation. Then ๐บ (๐) = โ โฆ ๐๐ด−1 = ๐๐บ−1(๐ต) โฆ ๐บ๐น (โ) = ๐บ (๐๐ต0 ) โฆ ๐บ๐น (โ) = ๐บ (๐๐ต0 โฆ ๐น (โ)). Since ๐บ is full and faithful (Exercise 1.3.32) we have ๐ = ๐๐ต0 โฆ ๐น (โ), which shows that ๐ is unique. It remains to see that defining ๐ = ๐๐ต0 โฆ ๐น (โ) gives indeed an arrow (๐น (๐ด), ๐๐ด ) → (๐ต, โ), i.e. that the diagram ๐๐ด ๐ด โ ๐บ๐น (๐ด) ๐บ (๐)=๐บ (๐๐ต0 )โฆ๐บ๐น (โ) ๐บ (๐ต) commutes. This holds since ๐บ (๐๐ต0 ) = ๐๐บ−1(๐ต) and ๐ is natural. We conclude, by Theorem 2.3.6, that ๐น is left adjoint to ๐บ . 29 Solutions by positrón0802 2.3 Adjunctions via initial objects Exercise 2.3.11. For the sake of contradiction suppose that there exists a set ๐ such that ๐๐ : ๐ → ๐ ๐น (๐) is not injective. Then there exist ๐ฅ, ๐ฆ ∈ ๐ such that ๐ฅ ≠ ๐ฆ and ๐๐ (๐ฅ) = ๐๐ (๐ฆ). We will prove that this implies that for all ๐ ∈ Set, the map ๐๐ : ๐ → ๐ ๐น (๐ ) is constant. So let ๐ ∈ Set. If |๐ | ≤ 1 there is nothing to prove, so assume |๐ | ≥ 2. Let ๐ก 1 ≠ ๐ก 2 ∈ ๐ . Let ๐ : ๐ → ๐ be any function sending ๐ฅ to ๐ก 1 and ๐ฆ to ๐ก 2 . By naturality of ๐, the diagram ๐๐ ๐ ๐น (๐) ๐ ๐ ๐น (๐ ) ๐ ๐ ๐ ๐น (๐ ) ๐๐ commutes. Since ๐๐ (๐ฅ) = ๐๐ (๐ฆ), we have ๐๐ (๐ก 1 ) = ๐๐ (๐ (๐ฅ)) = ๐๐ (๐ (๐ฆ)) = ๐๐ (๐ก 2 ). It follows that ๐๐ : ๐ → ๐ ๐น (๐ ) is constant, as asserted. Now let ๐ด ∈ ๐ be such that ๐ (๐ด) has at least two elements, say ๐ 1 ≠ ๐ 2 ∈ ๐ (๐ด). Let ๐ : ๐ (๐ด) → ๐ (๐ด) be any map such that ๐ (๐ 1 ) = ๐ 2 and ๐ (๐ 2 ) = ๐ 1 . By Lemma 2.3.5 there exists a unique ๐ : ๐น๐ (๐ด) → ๐ด such that the diagram ๐ (๐ด) ๐๐ (๐ด) ๐ ๐น๐ (๐ด) ๐ (๐) ๐ ๐ (๐ด) commutes. But ๐๐ (๐ด) is constant, say with value ๐ง ∈ ๐ ๐น๐ (๐ด). Then ๐ (๐) (๐ง) = ๐ (๐ 1 ) = ๐ 2 and ๐ (๐)(๐ง) = ๐ (๐ 2 ) = ๐ 1, a contradiction. It follows that ๐๐ : ๐ → ๐ ๐น (๐) is injective for all ๐ ∈ Set. In the case of the usual adjunction ๐น a ๐ between Grp and Set, where ๐ : Grp → Set is the forgetful functor and ๐น : Set → Grp is the free functor, this means that for each set ๐, the underlying set of the free group ๐น (๐) contains ๐. Exercise 2.3.12. Define a functor ๐บ : Par → Set∗ as follows. For ๐ด ∈ Par let ๐บ (๐ด) = ๐ด q {∗} with basepoint ∗, i.e. we take the (disjoint) union of ๐ด and the one-point set {∗} and take as basepoint the point the unique element of the latter. If (๐, ๐ ) : ๐ด → ๐ต is a map in Par let ๐บ (๐, ๐ ) : ๐ด q {∗} → ๐ต q {∗} ( ๐ (๐), ๐ ∈ ๐, ๐ โฆ→ ∗, ๐ ∉ ๐. If (๐, ๐ ) : ๐ด → ๐ต and (๐ , ๐ ) : ๐ต → ๐ถ are maps in Par then their composition is (๐ , ๐) โฆ (๐, ๐ ) = (๐ ∩ ๐ −1 (๐ ), ๐ โฆ ๐ |๐∩๐ −1 (๐ ) ). Thus ๐บ ((๐ , ๐) โฆ (๐, ๐ )) and ๐บ (๐ , ๐) โฆ ๐บ (๐, ๐ ) both send an element ๐ ∈ ๐ ∩ ๐ −1 (๐ ) to ๐๐ (๐) ∈ ๐ถ, and an element ๐ ∉ ๐ ∩ ๐ −1 (๐ ) to ∗ ∈ ๐ถ. Thus ๐บ : Par → Set∗ is a functor. We will show that ๐บ is an equivalence. First we show that it is full and faithful. Given ๐ด, ๐ต ∈ Par we want to prove that the map ๐บ (−) Par(๐ด, ๐ต) −−−−→ Set∗ (๐ด q ∗, ๐ต q ∗) 30 Solutions by positrón0802 3. Interlude on sets is a bijection. Given any pointed map ๐ : ๐ด q ∗ → ๐ต q ∗, it restricts to a map ๐ 0 : ๐ด → ๐ต, and ๐บ (๐ด, ๐ 0) = ๐ . Thus ๐บ (−) is surjective. Now assume that ๐บ (๐, ๐ ) = ๐บ (๐ , ๐) for (๐, ๐ ), (๐ , ๐) : ๐ด → ๐ต. Then both ๐ and ๐ are the complement of ๐บ (๐, ๐ ) −1 (∗) = ๐บ (๐ , ๐) −1 (∗) in ๐ด q {∗}, so ๐ = ๐ . Moreover, ๐ = ๐ on ๐ = ๐ by definition of ๐บ on morphisms. Thus ๐บ (−) is injective, hence a bijection. It remains to prove that ๐บ is essentially surjective on objects: if (๐ด, ๐) is any pointed set then ๐บ (๐ด \ ๐) = ((๐ด \ ๐) q {∗}, ∗) (๐ด, ๐) as pointed sets via the bijection ๐ : ๐บ (๐ด \ ๐) → (๐ด, ๐) whose restriction to ๐ด \ ๐ is the identity and such that ๐ (∗) = ๐. We conclude that ๐บ is an equivalence. Now we describe Set∗ as a coslice category. Let {∗} ∈ Set be the one-point set and consider the coslice category {∗}/Set = ({∗} ⇒ Set). It has objects pairs (๐ด, โ) where โ : ∗ → ๐ด, and a morphism (๐ด, โ) → (๐ต, ๐) is a map ๐ : ๐ด → ๐ต such that the diagram โ {∗} ๐ด ๐ ๐ ๐ต commutes. A map โ : {∗} → ๐ด is just a choice of basepoint โ(∗) ∈ ๐ด, and a morphism (๐ด, โ) → (๐ต, ๐) is precisely a basepoint-preserving map ๐ : (๐ด, โ(∗)) → (๐ต, ๐ (∗)). Thus {∗}/Set Set∗ . 3 3.1 Interlude on sets Constructions with sets Exercise 3.1.1. First we find a left adjoint ๐น to Δ. We want ๐น : Set × Set → Set such that for all ๐, ๐, ๐ ∈ Set, a map ๐น (๐, ๐ ) → ๐ is the same thing as a pair of maps ๐ → ๐, ๐ → ๐ . On objects define ๐น (๐, ๐ ) = ๐ q๐ . On morphisms, send a pair of maps ๐ : ๐ → ๐ 0, ๐ : ๐ → ๐ 0 to ๐น (๐ , ๐) = ๐ q ๐; that is, ๐น (๐ , ๐) is the unique map ๐ q ๐ → ๐ 0 q ๐ 0 whose restriction to ๐ is ๐ (followed by the inclusion ๐ 0 → ๐ 0 q ๐ 0) and whose restriction to ๐ is ๐ (followed by the inclusion ๐ 0 → ๐ 0 q ๐ 0). For any ๐, ๐ , ๐ ∈ Set we then have a canonical map ๐๐ ,๐ ,๐ : Set(๐ q ๐ , ๐ ) → Set × Set((๐, ๐ ), (๐, ๐ )) given by ๐ โฆ→ (๐ |๐ , ๐ |๐ ), which is a bijection with inverse given by sending a pair of maps ๐ : ๐ → ๐, ๐ : ๐ → ๐ to the unique map ๐ : ๐ q ๐ → ๐ whose restriction to ๐ is ๐๐ = ๐ and whose restriction to ๐ is ๐๐ = ๐. Naturality amounts to proving that given maps of sets ๐ : ๐ 0 → ๐, ๐ : ๐ 0 → ๐ and โ : ๐ → ๐ 0, the diagram Set(๐ q ๐ , ๐ ) ๐๐ ,๐ ,๐ Set × Set((๐, ๐ ), (๐, ๐ )) (๐ q๐) ∗ โฆโ ∗ (๐ ,๐) ∗ โฆ(โ,โ)∗ Set(๐ 0 q ๐ 0, ๐ 0) ๐๐ 0,๐ 0,๐ 0 Set × Set((๐ 0, ๐ 0), (๐ 0, ๐ 0)) 31 Solutions by positrón0802 3.1 Constructions with sets commutes. By definition, if ๐ : ๐ q ๐ → ๐ is a map, then both paths in the above diagram send ๐ to the pair of maps (โ โฆ ๐ |๐ โฆ ๐ , โ โฆ ๐ |๐ โฆ ๐). It follows that ๐๐ ,๐ ,๐ is a natural isomorphism and therefore ๐น a Δ. We now find a right adjoint to Δ. This is a functor ๐บ : Set × Set → Set such that for all ๐, ๐, ๐ ∈ Set, a map ๐ → ๐บ (๐, ๐ ) is the same thing as a pair of maps ๐ → ๐, ๐ → ๐ . Define ๐บ (๐, ๐ ) = ๐ × ๐ on objects and ๐บ (๐ , ๐) = ๐ × ๐ on morphisms. Given ๐, ๐ , ๐ ∈ Set there is a canonical map ๐๐ ,๐ ,๐ : Set × Set((๐, ๐ ), (๐, ๐ )) → Set(๐, ๐ × ๐ ) sending ๐ : ๐ → ๐, ๐ : ๐ → ๐ to the map ๐ : ๐ → ๐ ×๐ given by ๐ (๐ง) = (๐ (๐ง), ๐(๐ง)), and this is a bijection with inverse sending ๐ : ๐ → ๐ ×๐ to the pair of maps (๐๐ โฆ๐ : ๐ → ๐, ๐๐ โฆ๐ : ๐ → ๐ ), where ๐๐ and ๐๐ are the projections. If ๐ : ๐ → ๐ 0, ๐ : ๐ → ๐ 0 and โ : ๐ 0 → ๐ are maps of sets then the diagram Set × Set((๐, ๐ ), (๐, ๐ )) ๐๐ ,๐ ,๐ (โ×โ) ∗ โฆ( ๐ ×๐)∗ Set(๐, ๐ × ๐ ) โ ∗ โฆ( ๐ ×๐)∗ Set × Set((๐ 0, ๐ 0), (๐ 0, ๐ 0)) ๐๐ 0,๐ 0,๐ 0 Set(๐ 0, ๐ 0 × ๐ 0) commutes, for if (๐, ๐) is a pair of maps ๐ : ๐ → ๐, ๐ : ๐ → ๐ , then both paths in the above diagram send (๐, ๐) to the map ๐ : ๐ 0 → ๐ 0 × ๐ 0 given by ๐ (๐ง) = (๐ โฆ ๐ โฆ โ(๐ง), ๐ โฆ ๐ โฆ โ(๐ง)). Thus ๐๐ ,๐ ,๐ is a bijection natural in ๐, ๐ , ๐, and we conclude that Δ a ๐บ . Exercise 3.1.2. (This is also Mac Lane’s Exercise I.5.8.) Let ๐ be the category with objects triples (๐, ๐, ๐ก) where ๐ is a non-empty set, ๐ ∈ ๐ and ๐ก : ๐ → ๐ is a map. If (๐, ๐, ๐ก), (๐ 0, ๐ 0, ๐ก 0) are two objects, a map (๐, ๐, ๐ก) → (๐ 0, ๐ 0, ๐ก 0) is defined to be function ๐ : ๐ → ๐ 0 such that ๐ (๐) = ๐ 0 and ๐ โฆ๐ก = ๐ก 0 โฆ ๐ . Then the triple (N, 0, ๐ ) is initial in ๐. Indeed, given a non-empty set ๐ , an element ๐ฆ0 ∈ ๐ and a function ๐ก : ๐ → ๐ , we shall prove there exists a unique function ๐ : N → ๐ such that ๐ (0) = ๐ฆ0 and the diagram ๐ N ๐ ๐ N ๐ก ๐ ๐ commutes. This is constructed by induction, starting with ๐ (0) = ๐ฆ0 . If ๐ (๐) has been defined for some ๐ ∈ N, define ๐ (๐ + 1) = ๐ (๐ (๐)) = ๐ก (๐ (๐)). This shows that ๐ indeed exists and is unique. 32 Solutions by positrón0802 3.2 3.2 Small and large categories Small and large categories Exercise 3.2.12. (a) We will prove ๐ (๐) = ๐. On the one hand, © Ø ª Ø ­ ® ๐ (๐) = ๐ ­ ๐ ® = ๐ (๐ ) ⊂ ­ ® ๐ ∈๐ซ (๐ด) ๐ ∈๐ซ (๐ด) « ๐ ⊂๐ (๐ ) ¬ ๐ ⊂๐ (๐ ) Ø ๐ = ๐. ๐ ∈๐ซ (๐ด) ๐ ⊂๐ (๐ ) On the other hand, for each ๐ ∈ ๐ซ(๐ด) such that ๐ ⊂ ๐ (๐ ) we have ๐ (๐ ) ⊂ ๐ 2 (๐ ) since ๐ is order-preserving, so that ๐ 0 = ๐ (๐ ) in turn satisfies ๐ 0 ⊂ ๐ (๐ 0). It follows that ๐ ⊂ ๐ (๐) and therefore ๐ (๐) = ๐. (b) Let ๐ : ๐ซ(๐ด) → ๐ซ(๐ด) be given by ๐ (๐) = ๐ด \ ๐(๐ต \ ๐ (๐)) for all ๐ ⊂ ๐ด. We claim that ๐ is order-preserving. Let ๐ ⊂ ๐ 0 ⊂ ๐ด. Then ๐ (๐) ⊂ ๐ (๐ 0), so that ๐ต \ ๐ (๐ 0) ⊂ ๐ต \ ๐ (๐). Thus ๐(๐ต \ ๐ (๐ 0)) ⊂ ๐(๐ต \ ๐ (๐)) and therefore ๐ (๐) ⊂ ๐ (๐ 0). It follows from part (a) that there exists ๐ ∈ ๐ซ(๐ด) such that ๐ (๐) = ๐, that is, ๐(๐ต \ ๐ (๐)) = ๐ด \ ๐. (c) Let ๐ด and ๐ต be sets and assume that |๐ด| ≤ |๐ต| ≤ |๐ด|. We want to find a bijection โ : ๐ด → ๐ต. By assumption, there exists injective maps ๐ : ๐ด → ๐ต and ๐ : ๐ต → ๐ด. Let ๐ be a subset of ๐ด such that ๐(๐ต \ ๐ (๐)) = ๐ด \ ๐, which exists by part (a). Define ( ๐ (๐) ๐ ∈ ๐, โ : ๐ด → ๐ต, ๐ โฆ→ −1 ๐ (๐) ๐ ∈ ๐ด \ ๐. Note that this is well-defined since ๐ด \๐ is contained in the image of ๐. It remains to show that โ is a bijection. Let ๐, ๐ 0 ∈ ๐ด and suppose that โ(๐) = โ(๐ 0). If ๐ and ๐ 0 both belong to ๐, then ๐ = ๐ 0 by injectivity of ๐ . Similarly ๐ = ๐ 0 if both ๐ and ๐ 0 belong to ๐ด\๐. If ๐ ∈ ๐ and ๐ 0 ∈ ๐ด\๐ = ๐(๐ต \ ๐ (๐)), then ๐ (๐) = ๐−1 (๐ 0) = ๐ for some ๐ ∈ ๐ต \ ๐ (๐), a contradiction. It follows that โ is injective. Now let ๐ ∈ ๐ต be arbitrary; we want to prove that ๐ = โ(๐) for some ๐ ∈ ๐ด. If ๐ ∈ ๐ (๐ด) this is clear. If ๐ ∉ ๐ (๐ด), then ๐(๐) ∈ ๐(๐ต \ ๐ (๐)) = ๐ด \๐, so that โ(๐(๐)) = ๐. It follows that โ is bijective. We conclude that ๐ด ๐ต. Exercise 3.2.13. (a) Suppose that ๐ is surjective. Let ๐ต = {๐ ∈ ๐ด | ๐ ∉ ๐ (๐)} ∈ ๐ซ(๐ด). By assumption, there exists ๐ ∈ ๐ด such that ๐ (๐) = ๐ต. Then either ๐ ∈ ๐ (๐) = ๐ต or ๐ ∉ ๐ (๐) = ๐ต, but either case leads to a contradiction. It follows that ๐ is not surjective. (b) Since the function ๐ด → ๐ซ(๐ด) given by ๐ โฆ→ {๐} is injective, we have |๐ด| ≤ |๐ซ(๐ด)|. As there is no surjective function ๐ด → ๐ซ(๐ด), it follows that |๐ด| < |๐ซ(๐ด)|. Exercise 3.2.14. (a) Let ๐ผ be a set and (๐ด๐ )๐ ∈๐ผ a family of objects of ๐. Then the set ! Ø ๐ (๐ด๐ ) ๐=๐ซ ๐ ∈๐ผ 33 Solutions by positrón0802 3.2 Small and large categories satisfies |๐ (๐ด๐ )| < |๐ | for all ๐ ∈ ๐ผ . By Exercise 2.3.11, the unit map ๐ → ๐ ๐น (๐) is injective, so that |๐ (๐ด๐ )| < |๐ | ≤ |๐ ๐น (๐)|. It follows that ๐น (๐) is an object of ๐ which is not isomorphic to ๐ด๐ for any ๐ ∈ ๐ผ, for an isomorphism ๐น (๐) ๐ด๐ would induce an isomorphism ๐ ๐น (๐) ๐ (๐ด๐ ). (b) It follows from part (a) that the class of isomorphism classes of objects of ๐ is large, so ๐ is not essentially small. (c) For all of the categories Set, Vect๐ , Grp, Ab, Ring, and Top we have a forgetful functor ๐ into the category Set (whose left adjoint is a free functor) which satisfies the assumption of (a). It follows from (b) that none of these categories is essentially small. Exercise 3.2.15. (a) The category Mon is not even essentially small by Exercise 3.2.14(b), as the forgetful functor ๐ : Mon → Set satisfies the assumption of Exercise 3.2.14(a). It is locally small. (b) The group Z viewed as a one-object category is small as it has only one object with Z as set of morphisms. (c) The ordered set of integers Z has a set of objects and at most one arrow between any two objects, hence it is small. (d) The category Cat of small categories is not even essentially small by Exercise 3.2.14(b), by considering the functor ๐ : Cat → Set sending a small category to its set of objects and applying Exercise 3.2.16. Nevertheless, Cat is locally small: if ๐ and ๐ are small categories with sets of objects ob(๐) and ob(๐) and sets of morphisms mor(๐) and mor(๐), respectively, then the class of functors ๐ → ๐ is a subset of the set Set(ob(๐), ob(๐)) × Set(mor(๐), mor(๐)), so it is a set. (d) The multiplicative monoid of cardinals has one object whose class of morphisms is the monoid of cardinals. This class is not a set, for given a set of cardinals {๐ ๐ }๐ ∈๐ผ there exists a cardinal ๐ such that ๐ > ๐ ๐ for all ๐ ∈ ๐ผ . Therefore this category is not locally small. Exercise 3.2.16. First we find a right adjoint ๐ผ to ๐. Given a set ๐, let ๐ผ (๐) be the indiscrete category whose set of objects is ๐. That is, for any two ๐ , ๐ 0 ∈ ๐, there is precisely one arrow ๐ → ๐ 0 . If โ : ๐ → ๐ 0 is a function of sets, let ๐ผ (โ) : ๐ผ (๐) → ๐ผ (๐ 0) have object function โ and send the unique morphism ๐ → ๐ 0 in ๐ผ (๐) to the unique morphism โ(๐ ) → โ(๐ 0) in ๐ผ (๐ 0) for all ๐ , ๐ 0 ∈ ๐. Then ๐ผ : Set → Cat is a functor. We claim ๐ a ๐ผ . We need a bijection ๐ ๐,๐ : Set(๐ (๐), ๐) → Cat(๐, ๐ผ (๐)) natural in ๐ ∈ Cat and ๐ ∈ Set. Let ๐ ๐,๐ send a function โ : ๐ (๐) → ๐ to the functor ๐ → ๐ผ (๐) whose object function is โ and sending a morphism ๐ → ๐ 0 in ๐ to the unique morphism โ(๐) → โ(๐ 0) in ๐ผ (๐). Then ๐ ๐,๐ is bijective with inverse given by sending a functor ๐ → ๐ผ (๐) to its object function. It remains to prove that ๐ ๐,๐ is natural in ๐ and ๐. So let ๐น : ๐ 0 → ๐ be a functor between small categories, โ : ๐ → ๐ 0 be a function between sets and consider the diagram Set(๐ (๐), ๐) ๐ ๐,๐ Set(๐ (๐น ),โ) Set(๐ (๐ 0), ๐ 0) Cat(๐, ๐ผ (๐)) Cat(๐น,๐ผ (โ)) ๐ ๐0,๐ 0 34 Cat(๐ 0, ๐ผ (๐ 0)) . Solutions by positrón0802 3.2 Small and large categories If ๐ : ๐ (๐) → ๐ is a function, then both paths in the above diagram send ๐ to the functor ๐ 0 → ๐ผ (๐ 0) whose object function is โ โฆ ๐ โฆ ๐ (๐น ) and whose morphism function sends an arrow ๐ → ๐ 0 in ๐ 0 to the unique arrow โ โฆ ๐ โฆ ๐น (๐) → โ โฆ ๐ โฆ ๐น (๐ 0) in ๐ผ (๐ 0). It follows that the bijection ๐ ๐,๐ is natural in ๐ ∈ Cat and ๐ ∈ Set, and therefore ๐ a ๐ผ . We now find a left adjoint ๐ท to ๐. Given a set ๐, let ๐ท (๐) be the discrete category whose set of objects is ๐ (that is, the only morphisms are the identity morphisms). This gives a functor ๐ท : Set → Cat. To show that ๐ท a ๐, consider, for ๐ ∈ Set and ๐ ∈ Cat, the function ๐๐,๐ : Cat(๐ท (๐), ๐) → Set(๐, ๐ (๐)) given by sending a functor ๐ท (๐) → ๐ to its object function. Then ๐๐,๐ is clearly bijective. Moreover, if ๐น : ๐ → ๐ 0 is a functor between small categories and โ : ๐ 0 → ๐ a set function, then the diagram Cat(๐ท (๐), ๐) ๐๐,๐ Cat(๐ท (โ),๐น ) Cat(๐ท (๐ 0), ๐ 0) Set(๐, ๐ (๐)) Set(โ,๐ (๐น )) ๐๐ 0,๐0 Set(๐ 0, ๐ (๐ 0)) commutes. Indeed, given a functor ๐บ : ๐ท (๐) → ๐, both paths in the above diagrams send ๐บ to the function ๐ (๐น ) โฆ ๐ (๐บ) โฆ โ : ๐ 0 → ๐ (๐ 0). It follows that ๐๐,๐ is a bijection natural in ๐ ∈ Set and ๐ ∈ Cat, so that ๐ท a ๐. Finally, we find a left adjoint ๐ถ to ๐ท. Given a small category ๐, let ๐ถ (๐) be the quotient of the set of objects of ๐ by the equivalence relation generated by ๐ ∼ ๐ 0 if there exists an arrow ๐ → ๐ 0 in ๐. For ๐ ∈ ๐, write [๐] ∈ ๐ถ (๐) for its equivalence class. If ๐น : ๐ → ๐ 0 is a functor between small categories and ๐ → ๐ 0 is a morphism in ๐, then ๐น (๐) → ๐น (๐ 0) is a morphism in ๐ 0, hence the object function of ๐น descends to a function ๐ถ (๐น ) : ๐ถ (๐) → ๐ถ (๐ 0). Thus ๐ถ : Cat → Set is a functor. For ๐ถ ∈ Cat and ๐ ∈ Set, consider the function ๐ ๐,๐ : Set(๐ถ (๐), ๐) → Cat(๐, ๐ท (๐)) sending a map โ : ๐ถ (๐) → ๐ to the functor ๐ → ๐ท (๐) whose object function sends an element ๐ ∈ ๐ to โ[๐] ∈ ๐, and whose morphism function sends an arrow ๐ → ๐ 0 in ๐ to the identity arrow of โ[๐] = โ[๐ 0] ∈ ๐. If ๐น : ๐ → ๐ท (๐) is a functor, then for every arrow ๐ → ๐ 0 in ๐ the arrow ๐น (๐ → ๐ 0) is an identity arrow, so in particular ๐น (๐) = ๐น (๐ 0) ∈ ๐. Thus the object function of ๐น descends to a map ๐ถ (๐) → ๐. This assignment gives an inverse for ๐ ๐,๐ , which is therefore bijective. It remains to prove naturality on ๐ ∈ Cat and ๐ ∈ Set, which amounts to commutativity of the diagram Set(๐ถ (๐), ๐) ๐ ๐,๐ Set(๐ถ (๐น ),โ) Set(๐ถ (๐ 0), ๐ 0) Cat(๐, ๐ท (๐)) Cat(๐น,๐ท (โ)) ๐ ๐0,๐ 0 35 Cat(๐ 0, ๐ท (๐ 0)) Solutions by positrón0802 3.3 Historical remarks for all ๐น : ๐ 0 → ๐ and โ : ๐ → ๐ 0 . If ๐ : ๐ถ (๐) → ๐ is a function, both compositions above send ๐ to the functor ๐ 0 → ๐ท (๐ 0) whose object function is the composite map ๐ (๐น ) ๐ ๐ โ ๐ (๐ 0) −−−−→ ๐ (๐) → − ๐ถ (๐) → − ๐→ − ๐ 0, where ๐ is the quotient map, and whose morphism function sends an arrow ๐ → ๐ 0 in ๐ 0 to the identity arrow of โ โฆ ๐ [๐น (๐)] = โ โฆ ๐ [๐น (๐ 0)] ∈ ๐ 0 . It follows that ๐ ๐,๐ is natural in ๐ and ๐, so that ๐ถ a ๐ท. This completes the chain of adjoint functors ๐ถ a ๐ท a ๐ a ๐ผ. 3.3 Historical remarks No Exercises. 4 4.1 Representables Definitions and examples Exercise 4.1.26. • Let ๐ be the category of CW complexes, where for ๐, ๐ ∈ ๐, the set ๐(๐, ๐ ) is that of homotopy classes of maps from ๐ to ๐ . Then, given a positive integer ๐ and a group ๐บ, there exists a space ๐พ (๐บ, ๐) representing the functor ๐ โฆ๐ป ๐ (−; ๐บ) : Tophop → Set sending a CW complex to (the underlying set of) its ๐th cohomology group with values in ๐บ . The space ๐พ (๐บ, ๐) is usually called an Eilenberg-MacLane space. • Let ๐ denote the category of connected pointed CW complexes, where for ๐, ๐ ∈ ๐, the set ๐(๐, ๐ ) is that pointed homotopy classes of pointed maps from ๐ to ๐ . A functor ๐ธe∗ : ๐ op → AbZ from ๐ into the category AbZ of Z-graded abelian groups is called a cohomology theory. Given a positive integer ๐, the composition of ๐ธe∗ with the canonical projection AbZ → Ab, ๐ด โฆ→ ๐ด๐ , gives a functor ๐ธf๐ : ๐ op → Ab. Brown’s representability theorem asserts that ๐ธe∗ and any positive integer ๐, there exists ๐พ๐ ∈ ๐ such that ๐ธf๐ is represented by ๐พ๐ . • Generalising Example 4.1.8 of the fundamental group ๐ 1, for all ๐ ≥ 2 the composition of ๐th homotopy group functor ๐ ๐ : Toph∗ → Ab with the forgetful functor ๐ : Grp → Set is represented by (๐ ๐ , ∗) ∈ Toph∗ . • Let ๐น : Mat → Set be the functor sending ๐ โฆ→ R๐ , and ๐ด ∈ Mat(๐, ๐) to the map R๐ → R๐ given by left multiplication by the ๐ × ๐ matrix ๐ด. Then R๐ = ๐น (๐) Mat(1, ๐) naturally in ๐, i.e. ๐น is represented by 1 ∈ Mat. 36 Solutions by positrón0802 4.1 Definitions and examples Exercise 4.1.27. By assumption, there exists a bijection ๐ ๐ต : ๐(๐ต, ๐ด) → ๐(๐ต, ๐ด 0) natural in ๐ต ∈ ๐. By considering ๐ต = ๐ด we have a map ๐ = ๐๐ด (1๐ด ) : ๐ด → ๐ด 0, and by considering ๐ต = ๐ด 0 we have a map ๐ = ๐๐ด−10 (1๐ด0 ) : ๐ด 0 → ๐ด. By naturality, there following diagrams are commutative: ๐(๐ด 0, ๐ด) ๐๐ด0 ๐∗ ๐(๐ด, ๐ด) ๐๐ด ๐(๐ด 0, ๐ด 0) ๐(๐ด, ๐ด) ๐∗ ๐๐ด ๐∗ ๐(๐ด, ๐ด 0) , ๐(๐ด, ๐ด 0) ๐∗ ๐(๐ด 0, ๐ด) ๐๐ด0 ๐(๐ด 0, ๐ด 0) . Thus ๐๐ด (๐โฆ ๐ ) = ๐๐ด โฆ ๐ ∗ (๐) = ๐ ∗ โฆ๐๐ด0 (๐) = ๐ ∗ (1๐ด0 ) = ๐ = ๐๐ด (1๐ด ). Since ๐๐ด is injective, ๐โฆ ๐ = 1๐ด . Furthermore, ๐ โฆ ๐ = ๐∗ โฆ ๐๐ด (1๐ด ) = ๐๐ด0 โฆ ๐∗ (1๐ด ) = ๐๐ด0 (๐) = 1๐ด0 . It follows that ๐ : ๐ด → ๐ด 0 is an isomorphism. Exercise 4.1.28. Recall that ๐๐ is given on objects by ๐๐ (๐บ) = {elements of ๐บ of order 1 or ๐}. If ๐ : ๐บ → ๐บ 0 is a group homomorphism then there is an induced homomorphism ๐๐ (๐) : ๐๐ (๐บ) → ๐๐ (๐บ 0). Indeed, if ๐ฅ ∈ ๐๐ (๐บ) is not the identity then ๐ฅ has order ๐. Thus ๐ (๐ฅ) ๐ = ๐ (๐ฅ ๐ ) = ๐ (1) = 1 and since ๐ is prime, either ๐ (๐ฅ) is the identity or has order ๐. For all ๐บ ∈ Grp define ๐๐บ : ๐๐ (๐บ) → Grp(Z/๐Z, ๐บ) ๐ฅ โฆ→ (๐๐ฅ : 1 โฆ→ ๐ฅ). Note that ๐๐บ is well-defined by definition of ๐๐ (๐บ). It is of course injective. Moreover, it is surjective: any group homomorphism Z/๐Z → ๐บ is determined by its value at 1, and this value must have order 1 or ๐, i.e. must be an element of ๐๐ (๐บ). To show that ๐ : ๐๐ ⇒ Grp(Z/๐Z, −) is natural take a group homomorphism ๐ : ๐บ → ๐บ 0 and consider the diagram ๐๐ (๐บ) ๐๐บ Grp(Z/๐Z, ๐บ) ๐๐ (๐) ๐๐ (๐บ 0) ๐∗ ๐๐บ 0 Grp(Z/๐Z, ๐บ 0) . If ๐ฅ ∈ ๐๐ (๐บ), then ๐ ∗ โฆ๐๐บ (๐ฅ) and ๐๐บ 0 โฆ๐๐ (๐) (๐ฅ) both send 1 to ๐ (๐ฅ). Thus the diagram commutes. It follows that ๐ is a natural isomorphism ๐๐ Grp(Z/๐Z, −), and hence ๐๐ is represented by Z/๐Z. Exercise 4.1.29. Let ๐ denote the forgetful functor CRing → Set. By Exercise 0.13(a), for each ๐ ∈ CRing there is a bijective map ๐๐ : ๐ (๐ ) → CRing(Z[๐ฅ], ๐ ) given by sending an element ๐ ∈ ๐ to the unique morphism ๐๐ : Z[๐ฅ] → ๐ such that ๐๐ (๐ฅ) = ๐ . If ๐ : ๐ → ๐ 0 is a homomorphism of commutative rings and ๐ ∈ ๐ , then ๐ ∗ โฆ ๐๐ (๐ ) and ๐๐ 0 โฆ ๐ (๐), which are maps Z[๐ฅ] → ๐ 0, both send ๐ฅ to ๐ (๐ ), so by the uniqueness they are equal. Thus ๐ : ๐ ⇒ CRing(Z[๐ฅ], −) is a natural isomorphism. 37 Solutions by positrón0802 4.1 Definitions and examples Exercise 4.1.30. Write ๐ = {๐, ๐} for the Sierpiลski space, where {๐} is open. Let ๐ by a topological space. Given a subset ๐ ⊂ ๐ there is a function ๐๐ : ๐ → ๐ given by ๐๐ (๐ฅ) = ๐ if ๐ฅ ∈ ๐ , and ๐๐ (๐ฅ) = ๐ if ๐ฅ ∉ ๐ . The function ๐๐ is continuous since ๐ is open. Furthermore, it is clear that the map ๐๐ : ๐ช(๐ ) → Topop (๐, ๐ ) = Top(๐, ๐) ๐ โฆ→ ๐๐ is bijective with inverse given by sending a continuous map ๐ : ๐ → ๐ to the open subset ๐ −1 (๐) of ๐ . Now let ๐, ๐ ∈ Top and ๐ ∈ Topop (๐, ๐ ) = Top(๐ , ๐ ) and consider the diagram ๐ช(๐ ) ๐๐ ๐∗ ๐ช (๐ ) ๐ช(๐ ) Top(๐, ๐) ๐๐ Top(๐ , ๐) . If ๐ ∈ ๐ช(๐ ), then both ๐ ∗ โฆ ๐๐ (๐ ) and ๐๐ โฆ ๐ช(๐ ) (๐ ) send all of ๐ −1 (๐ ) to ๐ and all of ๐ \ ๐ −1 (๐ ) to ๐. Thus the diagram commutes. It follows that ๐ : ๐ช → Top(−, ๐) is a natural isomorphism, so that ๐ช is represented by ๐. Exercise 4.1.31. Let ๐ผ be the category with only two objects ๐, ๐ 0 ∈ ๐ and precisely one nonidentity morphism, ๐ : ๐ → ๐ 0 . For any ๐ ∈ Cat, a functor ๐น : ๐ผ → ๐ consists of a pair of objects ๐น (๐) = ๐ด, ๐น (๐ 0) = ๐ด 0 together with a morphism ๐น (๐ ) : ๐ด → ๐ด 0 . Thus, for ๐ ∈ Cat define ๐ ๐ : ๐ (๐) → Cat(๐ผ, ๐) to be the function sending an element โ ∈ ๐(๐ด, ๐ด 0) ⊂ ๐ (๐) to the functor ๐น : ๐ผ → ๐ given by ๐น (๐) = ๐ด, ๐น (๐ 0) = ๐ด 0 and ๐น (๐ ) = โ. Then ๐ ๐ is bijective. Moreover, consider a functor ๐บ : ๐ → โฌ between small categories and the diagram ๐ (๐) ๐๐ ๐ (๐บ) ๐ (โฌ) Cat(๐ผ, ๐) ๐บ∗ ๐โฌ Cat(๐ผ, โฌ) . Given โ ∈ ๐(๐ด, ๐ด 0) ⊂ ๐ (๐), both ๐บ ∗ โฆ ๐ ๐ (โ) and ๐ โฌ โฆ ๐ (๐บ) (โ) are the functor ๐ผ → โฌ sending ๐ โฆ→ ๐บ (๐ด), ๐ 0 โฆ→ ๐บ (๐ด 0) and ๐ โฆ→ ๐บ (โ). Thus ๐ : ๐ ⇒ Cat(๐ผ, −) is a natural isomorphism. We conclude that ๐ is represented by ๐ผ . Exercise 4.1.32. ๐น is left adjoint to ๐บ if and only if there exists bijections ๐๐ด,๐ต : โฌ(๐น (๐ด), ๐ต) ๐(๐ด, ๐บ (๐ต)) natural in ๐ด ∈ ๐ and ๐ต ∈ โฌ, which by Exercise 2.1.14 means that for all ๐ : ๐ด 0 → ๐ด, ๐ : ๐ด → ๐บ (๐ต) and ๐ : ๐ต → ๐ต 0 we have ๐บ (๐) โฆ ๐๐ด,๐ต (๐ ) โฆ ๐ = ๐๐ด0,๐ต0 (๐ โฆ ๐ โฆ ๐น (๐)). 38 Solutions by positrón0802 4.2 The Yoneda lemma This is precisely the statement that the diagram โฌ(๐น (๐ด), ๐ต) ๐๐ด,๐ต ๐(๐ด, ๐บ (๐ต)) โฌ(๐น (๐),๐) ๐ (๐,๐บ (๐)) โฌ(๐น (๐ด 0), ๐ต 0) ๐๐ด0,๐ต0 ๐(๐ด 0, ๐บ (๐ต 0)) commutes, i.e. that ๐ is a natural isomorphism โฌ(๐น (−), −) ๐(−, ๐บ (−)). 4.2 The Yoneda lemma Exercise 4.2.2. (Recall that ๐ป ๐ด = ๐(๐ด, −).) Dual of the Yoneda lemma. Let ๐ be a locally small category. Then [๐, Set] (๐ป ๐ด, ๐ ) ๐ (๐ด) naturally in ๐ด ∈ ๐ and ๐ ∈ [๐, Set]. Exercise 4.2.3. (a) Let ∗ denote the unique object of ๐. Recall that a functor ๐น : ๐ op → Set corresponds to the right ๐-set ๐น (∗) where the map ๐น (∗) × ๐ → ๐น (∗) is given by (๐ฅ, ๐) โฆ→ ๐น (๐)(∗). Since ๐ op has only one object ∗, there is only one representable functor ๐ op → Set, namely ๐ (−, ∗). This functor sends ∗ โฆ→ ๐ (∗, ∗) = ๐ and ๐ ∈ ๐ โฆ→ ๐ (๐, ∗) = ๐๐ : ๐ → ๐ where ๐๐ (๐ฅ) = ๐ฅ๐. This is precisely ๐. (b) Assume ๐ผ : ๐ → ๐ exists. Then, given ๐ ∈ ๐ we have ๐ผ (๐) = ๐ผ (1 ·๐) = ๐ผ (1) ·๐ = ๐ฅ ·๐. Hence ๐ผ is unique. Moreover, ๐ผ : ๐ → ๐ given by ๐ผ (๐) = ๐ฅ · ๐ preserves the right ๐-action. It follows that ๐ผ → ๐ผ (1) gives a bijection {maps ๐ → ๐ of right ๐-sets} → ๐ . (c) Let โณ denote the full subcategory of Cat consisting of the one-object categories. Recall from Example 3.2.11 that Mon ' โณ. Furthermore, from Example 1.3.4, a natural transformation between functors ๐ op → Set, where ๐ ∈ Mon, is the same thing as a map of right ๐-sets under the identification of functors ๐ op → Set with right ๐-sets. In (a) and (b) we have shown that given a one-object category ๐ ∈ โณ and a functor ๐ : ๐ op → Set, there is a bijection ๐๐ : [๐ op, Set] (๐, ๐ ) ๐ (∗), where ๐ = ๐ป ∗, with ∗ ∈ ๐ the unique object of ๐. The proof of the Yoneda lemma for โณ is complete once we prove that ๐๐ is natural in ๐, i.e. that if ๐, ๐ 0 ∈ [๐ op, Set] and ๐ : ๐ ⇒ ๐ 0 then the diagram [๐ op, Set] (๐, ๐ ) ๐๐ ๐ (∗) ๐ (∗) ๐∗ [๐ op, Set] (๐, ๐ 0) ๐๐ 0 ๐ 0 (∗) commutes. If ๐ผ : ๐ → ๐ is a map of ๐-sets, then ๐๐ (๐ผ) = ๐ผ (1) and hence ๐ (∗) (๐ผ (1)) = ๐ โฆ ๐ผ (1). On the other hand, ๐ ∗ (๐ผ) = ๐ โฆ ๐ผ, so ๐๐ 0 (๐ โฆ ๐ผ) = ๐ โฆ ๐ผ (1). Thus the diagram indeed commutes and the bijection ๐ผ๐ is natural in ๐ . We conclude that the Yoneda holds on โณ. 39 Solutions by positrón0802 4.3 4.3 Consequences of the Yoneda lemma Consequences of the Yoneda lemma Exercise 4.3.15. (a) Let ๐ : ๐ด → ๐ด 0 be a map in ๐. If ๐ : ๐ด → ๐ด 0 is an isomorphism in ๐, then ๐ฝ (๐ ) : ๐ฝ (๐ด) → ๐ฝ (๐ด 0) is an isomorphism in โฌ by functoriality of ๐ฝ . Conversely, assume that ๐ฝ (๐ ) is an isomorphism. Let ๐ : ๐ด 0 → ๐ด be the unique map ๐ด 0 → ๐ด such that ๐ฝ (๐) = ๐ฝ (๐ ) −1, which exists (and is unique) since ๐ฝ is full and faithful. Then ๐ฝ (๐ โฆ ๐ ) = 1 ๐ฝ (๐ด) and ๐ฝ (๐ โฆ ๐) = 1 ๐ฝ (๐ด0) , and since ๐ฝ is full and faithful, it follows ๐ โฆ ๐ = 1๐ด and ๐ โฆ ๐ = 1๐ด0 . Thus ๐ is an isomorphism. (a). (b) There is exactly one map ๐ : ๐ด → ๐ด 0 such that ๐ฝ (๐ ) = ๐, which is an isomorphism by part (c) This follows immediately from parts (a) and (b). Exercise 4.3.16. (a) Let ๐ด, ๐ด 0 ∈ ๐. Assume that ๐ , ๐ ∈ ๐(๐ด 0, ๐ด) are such that ๐ป ๐ = ๐ป๐ : ๐ป๐ด0 ⇒ ๐ป๐ด . Then for all ๐ต ∈ ๐, (๐ป ๐ )๐ต = (๐ป๐ )๐ต : ๐(๐ต, ๐ด 0) → ๐(๐ต, ๐ด). In particular, by considering ๐ต = ๐ด 0 we obtain ๐ = (๐ป ๐ )๐ด0 (1๐ด0 ) = (๐ป๐ )๐ด0 (1๐ด0 ) = ๐. It follows that ๐ป • is faithful. (b) Let ๐ด, ๐ด 0 ∈ ๐ and ๐ : ๐ป๐ด0 ⇒ ๐ป๐ด . We want to find ๐ : ๐ด 0 → ๐ด such that ๐ป ๐ = ๐. Consider ๐ = ๐๐ด0 (1๐ด0 ). Given any ๐ต ∈ ๐ and ๐ ∈ ๐(๐ต, ๐ด 0), by naturality of ๐ the diagram ๐(๐ด 0, ๐ด 0) ๐๐ด0 ๐∗ ๐(๐ต, ๐ด 0) ๐(๐ด 0, ๐ด) ๐∗ ๐๐ต ๐(๐ต, ๐ด) commutes. Thus ๐๐ต (๐) = ๐๐ต โฆ ๐∗ (1๐ด0 ) = ๐∗ โฆ ๐๐ด0 (1๐ด0 ) = ๐ โฆ ๐ = (๐ป ๐ )๐ต (๐) and hence ๐ป ๐ = ๐. It follows that that ๐ป • is full. (c) We are given an object ๐ด ∈ ๐, a functor ๐ : ๐ op → Set and an element ๐ข ∈ ๐ (๐ด) such that for each ๐ต ∈ ๐ and ๐ฅ ∈ ๐ (๐ต), there is a unique map ๐ฅ : ๐ต → ๐ด such that ๐ (๐ฅ) (๐ข) = ๐ฅ . Thus, for each ๐ต ∈ ๐ the map ๐ ๐ต : ๐ (๐ต) → ๐ป๐ด (๐ต) = ๐ป (๐ต, ๐ด) given by ๐ฅ โฆ→ ๐ฅ is bijective with inverse ๐ป (๐ต, ๐ด) → ๐ (๐ต) given by โ โฆ→ ๐ (โ) (๐ข). It remains to prove that the bijection ๐ ๐ต is natural in ๐ต. So let ๐ : ๐ต 0 → ๐ต be a morphism and consider the diagram ๐ (๐ต) ๐๐ต ๐∗ ๐ (๐ ) ๐ (๐ต 0) ๐ป (๐ต, ๐ด) ๐ ๐ต0 ๐ป (๐ต 0, ๐ด) . Let ๐ฅ ∈ ๐ (๐ต). On the one hand we have ๐ ∗ โฆ ๐ ๐ต (๐ฅ) = ๐ฅ โฆ ๐ , and on the other hand we have ๐ ๐ต0 โฆ ๐ ๐ (๐ฅ) = (๐ ๐ ) (๐ฅ). By definition, (๐ ๐ ) (๐ฅ) is the unique map ๐ต 0 → ๐ด such that ๐ ((๐ ๐ ) (๐ฅ)) (๐ข) = (๐ ๐ )(๐ฅ). Since ๐ (๐ฅ โฆ ๐ ) (๐ข) = (๐ ๐ ) (๐๐ฅ) (๐ข) = ๐ ๐ (๐ฅ), we have (๐ ๐ ) (๐ฅ) = ๐ฅ โฆ ๐ . Thus the diagram commutes. It follows that ๐ : ๐ ⇒ ๐ป๐ด is a natural isomorphism. Exercise 4.3.17. Let ๐ be a discrete category. Then ๐ = ๐ op . A presheaf ๐ : ๐ op → Set on ๐ is simply an assignment of a set ๐ (๐ด) ∈ Set for each ๐ด ∈ ๐. If ๐ด ∈ ๐, the presheaf represented 40 Solutions by positrón0802 4.3 Consequences of the Yoneda lemma by ๐ด if given by ( ๐ป ๐ด (๐ต) = ๐ป๐ด (๐ต) = {1๐ด }, ๐ต = ๐ด, ∅, ๐ต ≠ ๐ด. Now we analyse the Yoneda lemma. Let ๐ be a presheaf on ๐ and ๐ : ๐ป๐ด → ๐ a natural transformation, for some ๐ด ∈ ๐. Then ๐๐ต , for ๐ต ≠ ๐ด, is the empty function, so we must have ๐ (๐ต) = ∅ for all such ๐ต. (In particular, representable presheaves are precisely those ๐ such that for some ๐ด ∈ ๐, ๐ (๐ต) = for all ๐ต ≠ ๐ด and ๐ (๐ด) is a singleton.) So a natural transformation ๐ : ๐ป๐ด → ๐ is determined by ๐๐ด : {1๐ด } → ๐ (๐ด), an element of ๐ (๐ด). So the Yoneda lemma is clear in this case. Finally, consider Corollary 4.3.2. It says that a representation of ๐ consists of an object ๐ด ∈ ๐ and an element ๐ข ∈ ๐ (๐ด) such that ๐ (๐ต) = ∅ for all ๐ต ≠ ๐ด, and for all ๐ฅ ∈ ๐ (๐ด), there is a unique map ๐ฅ : ๐ด → ๐ด with ๐ (๐ฅ) (๐ข) = ๐ฅ . As there is only one map ๐ด → ๐ด, namely 1๐ด, the last condition says that ๐ฅ is the unique element of ๐ (๐ด), so ๐ (๐ด) is a singleton (as we deduced in the paragraph above). Similarly for Corollary 4.3.3. Exercise 4.3.18. (a) Let ๐น, ๐น 0 ∈ [โฌ, ๐]. We want to show that the map ๐ฝ [โฌ, ๐] (๐น, ๐น 0) → − [โฌ, ๐] (๐ฝ ๐น, ๐ฝ ๐น 0) is bijective. First suppose that ๐ผ, ๐ผ 0 : ๐น ⇒ ๐น 0 are natural transformations such that ๐ฝ ๐ผ = ๐ฝ ๐ผ 0 Then, given ๐ต ∈ โฌ, ๐ฝ (๐ผ ๐ต ) = ๐ฝ (๐ผ ๐ต0 ) : ๐ฝ ๐น (๐ต) → ๐ฝ 0๐น (๐ต). Since ๐ฝ is faithful there is at most one arrow ๐ : ๐น (๐ต) → ๐น 0 (๐ต) such that ๐ฝ (๐ ) = ๐ฝ (๐ผ ๐ต ) = ๐ฝ (๐ผ ๐ต0 ). Since both ๐ผ ๐ต and ๐ผ ๐ต0 are such arrows, we have ๐ผ ๐ต = ๐ผ ๐ต0 . It follows that ๐ผ = ๐ผ 0, so the above map is injective. Now we prove that it is also surjective. Let ๐ฝ : ๐ฝ ๐น ⇒ ๐ฝ ๐น 0 be natural. Since ๐ฝ is full and faithful, for each ๐ต ∈ โฌ there is precisely one map ๐ผ ๐ต : ๐น (๐ต) → ๐น 0 (๐ต) such that ๐ฝ (๐ผ ๐ต ) = ๐ฝ๐ต : ๐ฝ ๐น (๐ต) → ๐ฝ ๐น 0 (๐ต). It remains to prove that ๐ผ = (๐ผ ๐ต )๐ต ∈โฌ is natural, for then ๐ฝ (๐ผ) = ๐ฝ. Let ๐ : ๐ต → ๐ต 0 be a map in โฌ and consider the diagram ๐น (๐ต) ๐ผ๐ต ๐น 0 (๐ ) ๐น (๐ ) ๐น (๐ต 0) ๐น 0 (๐ต) ๐ผ ๐ต0 ๐น 0 (๐ต 0) . We want to show it is commutative. Now ๐น 0 (๐ ) โฆ ๐ผ ๐ต and ๐ผ ๐ต0 โฆ ๐น (๐ ) are such that ๐ฝ (๐น 0 (๐ ) โฆ ๐ผ ๐ต ) = ๐ฝ ๐น 0 (๐ ) โฆ ๐ฝ๐ต = ๐ฝ๐ต0 โฆ ๐น ๐ฝ (๐ ) = ๐ฝ (๐ผ ๐ต0 โฆ ๐น (๐ )) by naturality of ๐ฝ. Since ๐ฝ is faithful it follows that ๐น 0 (๐ ) โฆ ๐ผ ๐ต = ๐ผ ๐ต0 โฆ ๐น (๐ ). Thus ๐ผ is natural. We conclude that ๐ฝ โฆ − is full and faithful. (b) Since ๐บ and ๐บ 0 are both maps such that ๐ฝ โฆ ๐บ ๐ฝ โฆ ๐บ 0 and ๐ฝ โฆ − is full and faithful, then ๐บ ๐บ 0 by Lemma 4.3.8(a) (which was proved in Exercise 4.3.15.) (c) We have bijections ๐(๐ด, ๐บ (๐ต)) โฌ(๐น (๐ด), ๐ต) ๐(๐ด, ๐บ 0 (๐ต)) natural in ๐ด ∈ ๐ and ๐ต ∈ โฌ. Consider the Yoneda embedding ๐ป • : ๐ → [๐ op, Set], which is full and faithful by 41 Solutions by positrón0802 5. Limits Corollary 4.3.7. It follows from part (a) that the functor ๐ป • โฆ − : [โฌ, ๐] → [โฌ, [๐ op, Set]] is full and faithful. Note that ๐ป • โฆ ๐บ : โฌ → [๐, Set] is the functor ๐ต โฆ→ ๐ป๐บ (๐ต) = ๐(−, ๐บ (๐ต)), and similarly for ๐ป • โฆ ๐บ 0, so that ๐ป • โฆ ๐บ ๐ป • โฆ ๐บ 0 . Then ๐บ ๐บ 0 follows from (b). 5 5.1 Limits Limits: definition and examples Exercise 5.1.33. This was done in Exercise 0.14(a). Exercise 5.1.34. We will show that if ๐ธ is an equaliser then it is not necessarily a pullback. If the above square is a pullback then it has the following universal property: ∀๐ ∀๐ธ 0 ∃!โ ๐ ๐ธ ∀๐ ๐ ๐ ๐ ๐ ๐ ๐. If the maps ๐ and ๐ are equal, then โ indeed exists and is unique since ๐ธ is a coequaliser. However, we do not expect this to be true if ๐ ≠ ๐ in general. Indeed, consider for instance the category Set, ๐ = {1, 2}, ๐ = {1} and ๐ , ๐ the unique possible maps. Then the pullback is (๐ × ๐, pr1, pr2 ), and the equaliser is (๐, 1๐ ). Since ๐ × ๐ is not isomorphic to ๐, these are not equal. On the other hand, the converse does hold: if the given square is a pullback then (๐ธ, ๐) is the equaliser of ๐ and ๐. To show this, consider the diagram above which illustrates the universal property of the pullback. By taking ๐ and ๐ to be equal, we see that for any ๐ธ 0 and a map ๐ 0 : ๐ธ 0 → ๐ such that ๐ ๐ 0 = ๐๐ 0, there is a unique map โ : ๐ธ 0 → ๐ธ such that ๐โ = ๐ 0 . Hence (๐ธ, ๐) is the equaliser of ๐ and ๐. Exercise 5.1.35. (This is also Mac Lane’s Exercise III.4.8.) Assume we have a commutative diagram ๐ ๐ด ๐ ๐ต ๐ ๐ถ ๐ ๐ท ๐ ๐ธ ๐น ๐ โ in some category ๐ such that the right-hand square is a pullback. First assume that the left-hand square is also a pullback. Let ๐ป be an object together with maps ๐ก 1 : ๐ป → ๐ป, ๐ก 2 : ๐ป → ๐ถ such that ๐โ๐ก 1 = ๐๐ก 2 . We want to find a unique ๐ : ๐ป → ๐ด fitting in a 42 Solutions by positrón0802 5.1 Limits: definition and examples commutative diagram as below: ๐ก2 ๐ป ๐ ๐ ๐ด ๐ ๐ต ๐ถ ๐ก1 ๐ ๐ ๐ท ๐ ๐ธ ๐ โ ๐น. Since the right-hand square is a pullback there is a unique ๐ 0 : ๐ป → ๐ต such that ๐๐ 0 = โ๐ก 1 and ๐๐ 0 = ๐ก 2 . Then, since the left-hand square is a pullback there is a unique ๐ : ๐ป → ๐ด such that ๐๐ = ๐ก 1 and ๐ 0 = ๐ ๐. Then ๐ satisfies ๐๐ = ๐ก 1 and ๐๐ ๐ = ๐ก 2, and is unique as such. We deduce that the outer rectangle is a pullback. Now assume that the outer rectangle is a pullback. Let ๐ป be an object of ๐ and ๐ก 1 : ๐ป → ๐ท, ๐ก 2 : ๐ป → ๐ต be such that ๐๐ก 2 = โ๐ก 1 . We want to find a unique ๐ : ๐ป → ๐ด filling the diagram ๐ป ๐ก2 ๐ ๐ ๐ด ๐ ๐ต ๐ถ ๐ก1 ๐ ๐ ๐ท ๐ ๐ธ โ ๐ ๐น. Since ๐๐๐ก 2 = ๐๐๐ก 2 = ๐โ๐ก 1 and the right-hand square is a pullback, there exists a unique ๐ 0 : ๐ป → ๐ต such that ๐๐ 0 = ๐๐ก 2 and ๐๐ 0 = โ๐ก 1 . As ๐ก 2 satisfies these equations, we have that ๐ 0 = ๐ก 2 . Now, since the outer rectangle is a pullback there exists a unique ๐ : ๐ป → ๐ด such that ๐๐ ๐ = ๐๐ก 2 and ๐๐ = ๐ก 1 . Then ๐๐ ๐ = ๐๐ก 2 and ๐ ๐ ๐ = โ ๐๐ = โ๐ก 1, i.e. ๐ ๐ satisfies the equations for ๐ 0, and hence ๐ ๐ = ๐ก 2 by uniqueness. It follows that the left-hand square is a pullback. ๐ ๐ผ โฆโ ๐ข Exercise 5.1.36. (a) Note that (๐ด −−−→ ๐ท (๐ผ ))๐ผ ∈I is a cone on ๐ท. Indeed, if ๐ผ → − ๐ฝ in I then ๐ท๐ข โฆ ๐ ๐ผ โฆ โ = ๐ ๐ฝ โฆ โ since ๐ฟ is a cone. By the universal property of the limit, there is precisely one map โe: ๐ด → ๐ฟ such that ๐ ๐ผ โฆ โe = ๐ ๐ผ โฆ โ for all ๐ผ ∈ I. Since both โ and โ 0 satisfy this condition, it follows that โ = โ 0 . (b) A diagram ๐ท : I → Set is a pair of sets ๐, ๐ . Its limit is the product ๐ × ๐ together with the projections pr๐ , pr๐ . A map 1 → ๐ × ๐ is an element (๐ฅ, ๐ฆ) ∈ ๐ × ๐ . Thus (a) in this case means that given (๐ฅ, ๐ฆ), (๐ฅ 0, ๐ฆ 0) ∈ ๐ × ๐ , if ๐ฅ = ๐ฅ 0 and ๐ฆ = ๐ฆ 0 then (๐ฅ, ๐ฆ) = (๐ฅ 0, ๐ฆ 0). Exercise 5.1.37. Let ๐ฟ denote the given set ๐ข {(๐ฅ ๐ผ )๐ผ ∈I | ๐ฅ ๐ผ ∈ ๐ท (๐ผ ) for all ๐ผ ∈ I and (๐ท๐ข) (๐ฅ ๐ผ ) = ๐ฅ ๐ฝ for all ๐ผ → − ๐ฝ in I}, 43 Solutions by positrón0802 5.1 Limits: definition and examples ๐๐ผ and ๐ ๐ฝ : ๐ฟ → ๐ท (๐ฝ ) be given by ๐ ๐ฝ ((๐ฅ ๐ผ )๐ผ ∈I ) = ๐ฅ ๐ฝ for all ๐ฝ ∈ I. First note that (๐ฟ −→ ๐ท (๐ผ ))๐ผ ∈I is a ๐๐ผ cone on ๐ท by definition. Now let (๐ด − → ๐ท (๐ผ ))๐ผ ∈I be any cone on ๐ท. We need to show that there is a unique map โ : ๐ด → ๐ฟ such that ๐ ๐ฝ = ๐ ๐ฝ โฆ โ for all ๐ฝ ∈ I. Suppose such โ exists. Given ๐ ∈ ๐ด write โ(๐) = (โ๐ผ (๐))๐ผ ∈I . Then, for all ๐ฝ ∈ I, ๐ ๐ฝ (๐) = ๐ ๐ฝ โฆ โ(๐) = โ ๐ฝ (๐). This shows that if โ exists then it is unique. Moreover, define โ : ๐ด → ๐ฟ by โ(๐) = (๐๐ผ (๐))๐ผ ∈I . Note that ๐๐ผ (๐) ∈ ๐ท (๐ผ ) for ๐ข each ๐ผ, and if ๐ผ → − ๐ฝ is an arrow in I we have (๐ท๐ข) (๐๐ผ (๐)) = ๐ ๐ฝ (๐) since โ is a cone on ๐ท. Thus โ ๐๐ผ well-defined, and moreover satisfies ๐ ๐ฝ = ๐ ๐ฝ โฆ โ for all ๐ฝ ∈ I. It follows that (๐ฟ −→ ๐ท (๐ผ ))๐ผ ∈I is a limit cone. ๐๐ผ ๐ข Exercise 5.1.38. (a) First note that (๐ฟ −→ ๐ท (๐ผ ))๐ผ ∈I is a cone on ๐ท. Indeed, if ๐ผ → − ๐ฝ is an arrow in I we have ๐ ๐ฝ = pr ๐ฝ โฆ๐ = ๐ท๐ข โฆ pr๐ผ โฆ๐ = ๐ท๐ข โฆ ๐ ๐ผ as (๐ฟ, ๐) is the equaliser of ๐ and ๐ก . Now suppose ๐๐ผ that (๐ด − → ๐ท (๐ผ ))๐ผ ∈I is an arbitrary cone on ๐ท. We shall find a unique map โ : ๐ด → ๐ฟ such that Î ๐๐ผ = ๐ ๐ผ โฆ โ for all ๐ผ ∈ I. The family (๐๐ผ )๐ผ ∈I induces a unique map ๐ : ๐ด → ๐ผ ∈I ๐ท (๐ผ ) such that ๐ข pr๐ผ โฆ๐ = ๐๐ผ for all ๐ผ ∈ I. Given ๐ผ → − ๐ฝ an arrow in I then pr ๐ฝ โฆ๐ = ๐ ๐ฝ = ๐ท๐ข โฆ ๐๐ผ = ๐ท๐ข โฆ pr๐ผ โฆ๐ , so ๐ is a map such that ๐ โฆ ๐ = ๐ก โฆ ๐ . Since (๐ฟ, ๐) is an equaliser there is a unique map โ : ๐ด → ๐ฟ such that ๐ โฆ โ = ๐ , and this last equation is equivalent to satisfying ๐ ๐ผ โฆ โ = ๐๐ผ for all ๐ผ ∈ I. We ๐๐ผ conclude that (๐ฟ −→ ๐ท (๐ผ ))๐ผ ∈I is a limit cone on ๐ท. (b) Existence of a terminal object is equivalent to existence of the empty product. Assuming binary products, we have all finite products by iteration. Hence the same proof above as in (a) applies, assuming the category I to be finite. Exercise 5.1.39. (This is also Mac Lane’s Exercise III.4.10.) Denote by ∗ the terminal object of the category ๐. Given two objects ๐, ๐ of ๐, consider the pullback square ๐๐ ๐ธ ๐ ๐ ∗. ๐๐ We claim that (๐ธ, ๐๐ , ๐๐ ) is the product of ๐ and ๐ . Indeed, ๐ด ∈ ๐ and a pair of maps ๐๐ : ๐ด → ๐ and ๐๐ : ๐ด → ๐ , then the respective compositions with ๐ → ∗ and ๐ → ∗ are equal since ∗ is terminal, so there is a unique map ๐ : ๐ด → ๐ธ such that ๐๐ โฆ ๐ = ๐๐ and ๐๐ โฆ ๐ = ๐๐ . That is, (๐ธ, ๐๐ , ๐๐ ) satisfies the universal property of the product. Now, given parallel arrows ๐ , ๐ : ๐ → ๐ , consider the following pullback diagram ๐ ๐ธ ๐ ๐0 ๐ (1๐ ,๐ ) (1๐ ,๐) ๐ ×๐ . Then ๐ = ๐ 0 and ๐ : ๐ธ → ๐ is the equaliser of ๐ and ๐ (see Exercise 5.1.34.) 44 Solutions by positrón0802 5.1 Limits: definition and examples Thus our category has all binary products, equalisers and a terminal object, so it has all finite limits by Proposition 5.1.26(b) (proved in Exercise 5.1.38(b)). Exercise 5.1.40. (a) Monics in Set are precisely the injective maps (Example 5.1.30). Now ๐ : ๐ → ๐ด and ๐ 0 : ๐ 0 → ๐ด are isomorphic in Monic(๐ด) if and only if there exists maps ๐ : ๐ → ๐ 0 and ๐ : ๐ 0 → ๐ such that ๐ ๐ = 1๐ , ๐๐ = 1๐ and ๐ 0 ๐ = ๐, ๐๐ = ๐ 0 . These conditions are equivalent to existence of a bijection ๐ : ๐ → ๐ 0 such that ๐ 0 ๐ = ๐. If such ๐ exists then ๐ and ๐ 0 have the same image since ๐ is a bijection. Conversely, if ๐ and ๐ 0 have the same image, then they are bijections onto the subset ๐(๐ ) = ๐(๐ 0) of ๐ด. Then we can define ๐ : ๐ → ๐ 0 by ๐ (๐ฅ) = ๐ −1 (๐(๐ฅ)), and this is a bijection such that ๐ 0 ๐ = ๐. It follows that an isomorphism class of objects in Monic(๐ด) corresponds precisely to a subset of ๐ด, namely the image of any representative of the class. (b) Monics in each of these categories are precisely the injective morphisms, and the same proof as before applies, since the map ๐ (๐ฅ) = ๐ −1 (๐(๐ฅ)) defined at the end is indeed a morphism in the respective category. Therefore subobjects of Grp, Ring and Vect๐ are subgroups, subrings and subspaces respectively. (c) As in the category of sets, a map ๐ in Top is monic if and only if it is injective. Now, for ๐ด ∈ Top and monics ๐ : ๐ → ๐ด, ๐ 0 : ๐ 0 → ๐ด representing objects of Monic(๐ด), a map from ๐ to ๐ 0 is a homeomorphism ๐ : ๐ → ๐ 0 such that ๐ 0 ๐ = ๐. In this case ๐ : ๐ → ๐ 0 given by ๐ (๐ฅ) = ๐ −1 (๐(๐ฅ)) may not be continuous. Hence subobjects of ๐ด are not its subspaces. They are subsets ๐ ⊂ ๐ด equipped with a topology finer than the subspace topology inherited from ๐ด. Exercise 5.1.41. (This is also (the dual of) Mac Lane’s Exercise III.4.4.) First assume that ๐ is monic and consider the diagram ๐ 1๐ 1๐ ๐ ๐ ๐ ๐ ๐. Suppose we are given arrows ๐, ๐ 0 : ๐ → ๐, such that ๐ โฆ ๐ = ๐ โฆ ๐ 0 . We must show there is a unique โ : ๐ → ๐ such that 1๐ โฆ โ = ๐ and 1๐ โฆ โ = ๐ 0 . But ๐ is monic, so โ = ๐ = ๐ 0 works. Conversely, assume the diagram above is a pullback and let ๐, ๐ 0 : ๐ → ๐ be such that ๐ โฆ ๐ = ๐ โฆ ๐ 0 . Then there is a unique โ : ๐ → ๐ such that 1๐ โฆ โ = ๐ and 1๐ โฆ โ = ๐ 0, so ๐ = ๐ 0 . Exercise 5.1.42. (This is also Mac Lane’s Exercise III.4.5.) Let ๐, ๐ 0 : ๐ → ๐ 0 be arrows such that ๐ 0 โฆ๐ = ๐ 0 โฆ๐ 0 . Then ๐ โฆ ๐ 0 โฆ๐ = ๐ โฆ๐ 0 โฆ๐ = ๐ โฆ๐ 0 โฆ๐ 0 = ๐ โฆ ๐ 0 โฆ๐ 0, so ๐ 0 โฆ๐ = ๐ 0 โฆ๐ 0 as ๐ is monic. Since the square is a pullback and ๐ โฆ ๐ 0 โฆ๐ = ๐ โฆ๐ 0 โฆ๐, there is a unique โ : ๐ → ๐ 0 such that ๐ 0 โฆ โ = ๐ 0 โฆ ๐ and ๐ 0 โฆ โ = ๐ 0 โฆ ๐. Both ๐ and ๐ 0 satisfy the equations for โ, so ๐ = ๐ 0 . It follows that ๐ 0 is monic. 45 Solutions by positrón0802 5.2 5.2 Colimits: definition and example Colimits: definition and example 1๐ Exercise 5.2.21. First assume that ๐ = ๐ก . Consider ๐ −−→ ๐ . Then ๐ 1๐ = ๐ก1๐ . Moreover, any map ๐ ๐→ − ๐ satisfies the condition ๐ ๐ = ๐ก๐, and given any such map then ๐ is the unique map ๐ → ๐ such that 1๐ ๐ = ๐. Thus (๐, 1๐ ) is the equaliser of ๐ and ๐ก . Similarly, (๐ , 1๐ ) is the coequaliser of ๐ and ๐ก . Now suppose that the equaliser (๐, ๐) of ๐ and ๐ก is an isomorphism. Then ๐ ๐ = ๐ก๐ and ๐ = −1 ๐ ๐๐ = ๐ก๐๐ −1 = ๐ก . Similarly, if the coequaliser of ๐ and ๐ก is an isomorphism then ๐ = ๐ก . Exercise 5.2.22. (a) By Example 5.2.9, the coequaliser of ๐ and 1๐ is the quotient ๐ /∼ where ∼ is the equivalence relation generated by ๐ (๐ฅ) ∼ ๐ฅ for all ๐ฅ ∈ ๐, together with the quotient map ๐ → ๐ /∼ . This is ๐ /∼ where ๐ฅ ∼ ๐ฆ if there exists ๐ ≥ 1 such that ๐ ๐ (๐ฅ) = ๐ฆ or ๐ ๐ (๐ฆ) = ๐ฅ . (b) In Top, the coequaliser is again the quotient ๐ /∼ as before, equipped with the quotient topology. Let ๐ ∈ [0, 2๐] be irrational and consider the map ๐ : ๐ 1 → ๐ 1 given by ๐ ๐๐ก โฆ→ ๐ ๐ (๐ก +๐ ) . Let ๐ : ๐ → ๐ be the coequaliser of ๐ and 1๐ , where ๐ = ๐ /∼ is the quotient of ๐ by the relation generated by ๐ (๐ฅ) ∼ ๐ฅ for all ๐ฅ ∈ ๐ . Then ๐ is uncountable, for each equivalence class [๐ฅ] is countable. Let ๐ ⊂ ๐ be a non-empty open set, and [๐ฅ] ∈ ๐ . Since ๐ −1 (๐ ) is open in ๐, there exists ๐ > 0 such that ๐ต(๐ฅ, ๐) ∩ ๐ 1 ⊂ ๐ −1 (๐ ). Then ๐ต(๐ง, ๐) ∩ ๐ 1 ⊂ ๐ −1 (๐ ) for all ๐ง ∈ ๐ such that [๐ง] = [๐ฅ]. Now, let ๐ง ∈ ๐ be arbitrary. As ๐ is irrational, the set ๐ด = {๐ ๐ (๐ฅ 0) | ๐ ∈ Z} is dense in ๐, so there exists ๐ค ∈ ๐ด ∩ ๐ต(๐ฅ, ๐) ∩ ๐ 1 . Then [๐ค] = [๐ง] and ๐ค ∈ ๐ −1 (๐ ), so ๐ (๐ง) = ๐ (๐ค) ∈ ๐ . Thus ๐ −1 (๐ ) = ๐, so ๐ = ๐ . It follows that ๐ has the indiscrete topology. Exercise 5.2.23. (a) Let ๐ denote the inclusion (N, +, 0) โฉ− → (Z, +, 0). Let (๐, +๐ , 0๐ ) be a monoid and ๐ , ๐ : (Z, +, 0) → (๐, +๐ , 0๐ ) morphisms of monoids such that ๐ ๐ = ๐๐. Then ๐ (๐) = ๐(๐) for all ๐ ≥ 0. Let ๐ < 0. Then 0๐ = ๐ (0) = ๐ (๐ − ๐) = ๐ (๐) +๐ ๐ (−๐), and similarly 0๐ = ๐ (−๐) +๐ ๐ (๐), so ๐ (๐) is the inverse of ๐ (−๐) in ๐. Similarly ๐(๐) is the inverse of ๐(−๐) in ๐. Since inverses are unique and ๐ (−๐) = ๐(−๐), we have ๐ (๐) = ๐(๐). Thus ๐ (๐) = ๐(๐) for all ๐ < 0 and therefore ๐ = ๐. (b) (This is also Mac Lane’s Exercise I.5.4.) Let ๐ denote the inclusion Z โฉ− → Q. Let ๐ be a ring and ๐,๐ : Q → ๐ be morphisms such that ๐๐ = ๐๐. Then ๐ = ๐ on Z. If ๐ ∈ Z \ {0} then ๐ ๐ 1 1 1 −1 −1 1 = ๐ (1) = ๐ ๐ = ๐ (๐)๐ ๐ , so ๐ ๐ = ๐ (๐) = ๐ (๐) = ๐ ๐ . Thus ๐ ๐ = ๐ (๐)๐ (๐) −1 = ๐ ๐ ๐ (๐)๐ (๐) −1 = ๐ ๐ for all ๐ ∈ Q, so ๐ = ๐ . It follows that ๐ is epic (and not surjective!). Exercise 5.2.24. (a) Epics in Set are precisely the surjective maps (Example 5.2.18). In general, a surjective function ๐ : ๐ → ๐ induces an equivalence relation on ๐ given by ๐ฅ ∼ ๐ฅ 0 if ๐ (๐ฅ) = ๐ (๐ฅ 0). Given epics ๐ : ๐ด → ๐, ๐ 0 : ๐ด → ๐ 0, an isomorphism from ๐ to ๐ 0 is a map ๐ : ๐ → ๐ 0 such that ๐ ๐ = ๐ 0 and there is a map ๐ : ๐ 0 → ๐ with ๐๐ 0 = ๐ and ๐๐ = 1, ๐ ๐ = 1. This amounts to a bijection ๐ : ๐ → ๐ 0 such that ๐ ๐ = ๐ 0 . So assume there is such bijection. Let ๐, ๐ 0 ∈ ๐ด. Then ๐ (๐) = ๐ (๐ 0) =⇒ ๐ ๐ (๐) = ๐ ๐ (๐ 0) =⇒ ๐ 0 (๐) = ๐ 0 (๐ 0), and similarly (using ๐ −1 ) ๐ 0 (๐) = ๐ 0 (๐ 0) =⇒ ๐ (๐) = ๐ (๐ 0). Thus the equivalence relations induced on ๐ด by ๐ and ๐ 0 coincide. Conversely, assume that ๐ (๐) = ๐ (๐ 0) ⇐⇒ ๐ 0 (๐) = ๐ 0 (๐ 0) for all ๐, ๐ 0 ∈ ๐ด. Define ๐ : ๐ → ๐ 0 by 46 Solutions by positrón0802 5.2 Colimits: definition and example ๐ (๐ฅ) = ๐ 0 (๐) where ๐ is any preimage of ๐ฅ under ๐. This is well-defined by assumption, and is a bijection such that ๐ ๐ = ๐ 0 . Thus ๐ and ๐ 0 are isomorphic in Epic(๐ด). We deduce that the quotient objects of ๐ด are in canonical one-to-one correspondence with the equivalence relations of ๐ด, namely a quotient object represented by some ๐ : ๐ด → ๐ corresponds to the equivalence relation on ๐ด such that ๐ ∼ ๐ 0 if and only if ๐ (๐) = ๐ (๐ 0). (b) (The proof that epic =⇒ surjective on Grp can be found in Mac Lane’s Exercise I.5.5.) ๐ ๐ An epimorphism ๐บ − → ๐ป induces an isomorphism ๐ e: ๐บ/ker ๐ → ๐ป . Two epimorphisms ๐บ − → ๐ป, ๐ ๐บ− → ๐ป 0 are isomorphic in Epic(๐บ) if and only if there is an isomorphism ๐ : ๐ป → ๐ป 0 such that ๐๐ = ๐ . If there is such ๐ then ker ๐ = ker๐ . Conversely, if ker ๐ = ker๐, let ๐ = (๐e) −1๐ e: ๐ป → ๐ป 0 . Then ๐ is an isomorphism, and is given by ๐ (๐) โฆ→ ๐ (๐) for ๐ ∈ ๐บ, so ๐๐ = ๐ . Thus ๐,๐ ∈ Epic(๐บ) are isomorphic if and only if ker ๐ = ker๐ . Hence the correspondence ๐ ↔ ker ๐ is one-to-one between quotient objects of ๐บ and normal subgroups of ๐บ . (Any kernel is normal, and any normal subgroup ๐ of ๐บ is the kernel of ๐บ → ๐บ/๐ .) Exercise 5.2.25. (a) Let ๐ : ๐ด → ๐ต be a morphism. First suppose that ๐ is split monic and let ๐ : ๐ต → ๐ด be such that ๐๐ = 1๐ด . Consider the maps ๐๐, 1๐ต : ๐ต → ๐ต. Then ๐๐๐ = ๐1๐ด = 1๐ต๐. Moreover, if โ : ๐ถ → ๐ต is a map such that ๐๐โ = 1๐ต โ = โ, then ๐โ : ๐ถ → ๐ด satisfies ๐(๐โ) = โ, and if โ 0 : ๐ถ → ๐ด satisfies ๐โ 0 = โ then โ 0 = ๐๐โ 0 = ๐โ. Thus ๐ is the equaliser of the maps ๐๐ : ๐ต → ๐ต and 1๐ต : ๐ต → ๐ต, so it is regular monic. Now assume that ๐ is regular monic, and let ๐ถ be an object and ๐ , ๐ : ๐ต → ๐ถ maps of which ๐ is an equaliser. Suppose that โ, โ 0 : ๐ → ๐ด are maps such that ๐โ = ๐โ 0 . Since ๐ is an equaliser, given any map ๐ : ๐ → ๐ต such that ๐ ๐ = ๐๐ then there exists a unique โe: ๐ → ๐ด such that e so โ = โ 0 by ๐โe = ๐. Taking ๐ = ๐โ = ๐โ 0 then both โ and โ 0 satisfy the condition for โ, uniqueness. Thus ๐ is monic. (b) Let ๐ : ๐ด → ๐ต be monic in Ab, that is, ๐ is a monomorphism. Consider ๐ถ = ๐ต/im ๐, and the maps ๐ : ๐ต → ๐ถ, 0 : ๐ต → ๐ถ, where ๐ is the projection. Then ๐๐ = 0๐ = 0, and is universal as such. Indeed, if ๐ : ๐ → ๐ต is a homomorphism such that ๐๐ = 0, then ๐ (๐ ) ⊂ ๐ (๐ด), and we can define a map ๐ : ๐ → ๐ด by ๐ (๐ฅ) = ๐ if ๐ (๐ฅ) = ๐ (๐). This is well-defined since ๐ is injective. Moreover, it is the unique map with the property ๐๐ = ๐ . Thus ๐ : ๐ด → ๐ต is the equaliser of ๐ : ๐ต → im ๐ and 0 : ๐ต → ๐ถ. It follows that all monics in Ab are regular monics. To show that not all monics in Ab are split monics consider ๐ : Z → Z given by ๐ (1) = 2. Then ๐ is a monomorphism, but there is no ๐ : Z → Z such that ๐๐ = 1Z . Indeed, any such ๐ would send 2 to 1, which is not possible. (c) We will show that the regular monics in Top are precisely the embeddings, i.e. the injective maps which are homeomorphisms onto their image. Let โ : ๐ → ๐ be regular monic in Top. In particular it is monic, i.e. it is injective. Let โ : ๐ → โ(๐ ) be obtained from โ by restricting its codomain. Let ๐ , ๐ : ๐ → ๐ be maps of which โ is an equaliser. Consider the inclusion map ๐ : โ(๐ ) โฉ− → ๐ . Then ๐ ๐ = ๐๐, so there is a unique ๐ : โ(๐ ) → ๐ such that ๐ = โ๐ = ๐โ๐. As ๐ is monic, it follows that โ๐ = 1โ (๐ ) . Thus, the inverse function ๐ : โ(๐ ) → ๐ of โ is continuous, so 47 Solutions by positrón0802 5.2 Colimits: definition and example โ is an embedding. Conversely, assume that โ : ๐ → ๐ is an embedding. Consider the pushout โ ๐ ๐ ๐ โ ๐ ๐ ๐. By the construction of pushouts in Top, ๐ = ๐ q ๐ /∼ where an element โ(๐ฅ) ∈ ๐ in the first summand is identified with the same element โ(๐ฅ) ∈ ๐ in the second summand, and ๐ , ๐ are the inclusions. The equaliser of ๐ , ๐ is then the inclusion ๐ 0 → ๐ of the subspace ๐ 0 of ๐ where both ๐ and ๐ are equal, i.e. โ(๐ ) → − ๐ . Since โ is an embedding this is the same as โ : ๐ → ๐ . Exercise 5.2.26. A regular epic is thus a map which is a coequaliser, and a split epic is a map with a right inverse. As in Exercise 5.2.25(a), split epic =⇒ regular epic =⇒ epic. (a) Let ๐ : ๐ด → ๐ต be map in a category. If ๐ is an isomorphism, then the equations ๐ −1 ๐ = 1 and ๐ ๐ −1 = 1 imply that ๐ is split epic and split monic, in particular monic and regular epic. Conversely, assume that ๐ is both monic and regular epic. Let ๐, โ : ๐ถ → ๐ด be maps of which ๐ is a coequaliser. Then ๐ ๐ = ๐ โ, so ๐ = โ as ๐ is monic. By Exercise 5.2.21, ๐ is an isomorphism (b) It suffices to prove that epic =⇒ split epic in Set. Assume ๐ : ๐ด → ๐ต is epic in Set, i.e. ๐ is surjective. By the axiom of choice it follows that ๐ has a section, that is, there exists ๐ : ๐ต → ๐ด such that ๐ ๐ = 1๐ต . Thus ๐ is split epic. (c) In Top the epimorphisms are precisely the surjective (continuous) maps. Let ๐ ∈ Top and suppose that ๐ 0 ∈ Top is another space with the same underlying set as ๐, but whose topology is strictly finer. Then the identity ๐ 0 → ๐ is a map in Top, and is epic, but is is not split since the identity ๐ → ๐ 0 is not continuous. It follows from (b) that Top does not satisfy the axiom of choice. In Grp the epics are precisely the surjections (c.f. Exercise 5.2.24 and Mac Lane’s Exercise I.5.5.) Consider the quotient map ๐ : Z ∗ Z → Z ⊕ Z of the free group Z ∗ Z on two generators onto its abelianisation Z ⊕ Z. Then ๐ is epic, but is not split. Indeed, no map Z ⊕ Z → Z ∗ Z is as since Z ⊕ Z is countable and Z ∗ Z is uncountable. It follows from (b) that Grp does not satisfy the axiom of choice. Exercise 5.2.27. First we analyse stability under pullbacks. By Exercise 5.1.42, monics are stable under pullbacks. Now consider the category generated by the graph • ๐ • ๐ ๐ ๐ • ๐ • • ๐ with the relations ๐ ๐ = ๐๐, ๐๐ = ๐๐ and ๐ ๐๐ = ๐๐๐. Then ๐ is the equaliser of ๐ and ๐, the square is a pullback, but ๐ is not an equaliser, as can be check. It follows that regular monics are not stable 48 Solutions by positrón0802 5.2 Colimits: definition and example under pullbacks. Finally, note that split monics are not stable under pullbacks either: let ๐ and ๐ be disjoint empty subsets of a set ๐ . Then ∅ ๐ ๐ ๐ is a pullback on Set, where ๐ โฉ− → ๐ is split monic but ∅ โฉ− → ๐ is not. Next, epics are not stable under pullbacks. Consider the category Haus of Hausdorff topological spaces. If ๐ด ⊂ ๐ต is an inclusion of a dense subspace in Haus, it is epic, for if two maps into a Hausdorff space agree on a dense subset, they are equal. It follows that in the pullback diagram ∅ Q R\Q R in Haus the map Q ← −โช R is epic, but ∅ → R \ Q is clearly not. Now, an example similar to the one of regular monics above (where now ๐ and ๐ are maps into the left-bottom corner) shows that regular epics are not stable under pullbacks either. Finally, note that split epics are stable under pullbacks. Indeed, let ๐0 ๐ ๐ ๐0 ๐ ๐ ๐ ๐ be a pullback square in some category, where ๐ is split epic. Then there exists ๐ : ๐ → ๐ a right inverse of ๐ . Then the maps 1๐ : ๐ → ๐ and ๐ ๐ : ๐ → ๐ satisfy ๐1๐ = ๐ ๐ ๐, so by the universal property of the pullback there is a map ๐ 0 : ๐ → ๐ such that ๐ 0๐ 0 = 1๐ . Thus ๐ 0 is split epic. Now we analyse stability under composition. If ๐ and ๐ are monic and โ 1, โ 2 are maps such that ๐ ๐โ 1 = ๐ ๐โ 2, we have ๐โ 1 = ๐โ 2 since ๐ is monic, and in turn โ 1 = โ 2 since ๐ is monic. Thus monics are stable under composition. Dually, epics are stable under composition too. Finally, regular monics are not stable under compositions. Consider the full subcategory FHaus of Haus spanned by the functionally Hausdorff spaces (or completely Hausdorff spaces): those spaces ๐ such that for any ๐ฅ, ๐ฆ ∈ ๐, there exists a continuous map ๐ : [0, 1] → ๐ such that ๐ (0) = ๐ฅ and ๐ (1) = ๐ฆ. Let ๐ด = {1/๐ | ๐ ∈ Z+ }, ๐ต = ๐ด ∪ {0}, and ๐ถ the subspace whose underlying set is R and having basis ๐ฏ ∪ {R\๐ด} for its topology, where ๐ฏ is a basis for the standard topology on R. Then it can be prove that the inclusions ๐ด ⊂ ๐ต and ๐ต ⊂ ๐ถ are regular monics in FHaus, but their composition is not. Details would take us too far into analysis and hence are omitted. Since regular epics in a category ๐ are precisely regular monics in ๐ op, it follows that regular epics are not stable under composition either. 49 Solutions by positrón0802 5.3 5.3 Interactions between functors and limits Interactions between functors and limits Exercise 5.3.8. Let ๐น : ๐ × ๐ be given on objects by ๐น (๐, ๐ ) = ๐ × ๐ (for our previously made choice). For a morphism (๐ , ๐) : (๐, ๐ ) → (๐ 0, ๐ 0) in ๐ × ๐ define ๐น (๐ , ๐) : ๐ ×๐ → ๐ 0 ×๐ 0 to be the unique map ๐ ×๐ → ๐ 0 ×๐ 0 induced by the maps ๐ โฆ๐ 1๐ ,๐ : ๐ ×๐ → ๐ 0, ๐ โฆ๐ 2๐ ,๐ : ๐ ×๐ → ๐ 0 using the universal property of the product ๐ 0 × ๐ 0 . Then clearly ๐น (1๐ , 1๐ ) = 1๐ ,๐ . Consider two maps (๐ , ๐) : (๐, ๐ ) → (๐ 0, ๐ 0) and (โ, ๐) : (๐ 0, ๐ 0) → (๐ 00, ๐ 00) in ๐ × ๐. Then ๐น (โ๐ , ๐๐) is the 00 00 00 00 unique map ๐ ×๐ → ๐ 00 ×๐ 00 such that ๐ 1๐ ,๐ โฆ ๐น (โ๐ , ๐๐) = โ โฆ ๐ โฆ ๐ 1๐ ,๐ and ๐ 2๐ ,๐ โฆ ๐น (โ๐ , ๐๐) = 00 00 0 0 ๐ โฆ ๐ โฆ ๐ 2๐ ,๐ . Since ๐ 1๐ ,๐ โฆ ๐น (โ, ๐) = โ โฆ ๐ 1๐ ,๐ we have ๐ 1๐ 00 00,๐ 00 0 0 โฆ ๐น (โ, ๐) โฆ ๐น (๐ , ๐) = โ โฆ ๐ 1๐ ,๐ โฆ ๐น (๐ , ๐) = โ โฆ ๐ โฆ ๐ 1๐ ,๐ . 00 Similarly ๐ 2๐ ,๐ โฆ ๐น (โ, ๐) โฆ ๐น (๐ , ๐) = ๐ โฆ ๐ โฆ ๐ 1๐ ,๐ . By uniqueness it follows that ๐น (โ๐ , ๐๐) = ๐น (โ, ๐) โฆ ๐น (๐ , ๐). Thus ๐น is indeed a functor. Exercise 5.3.9. Given ๐ด, ๐, ๐ ∈ ๐ define ๐๐ด,๐ ,๐ : ๐(๐ด, ๐ × ๐ ) → ๐(๐ด, ๐ ) × ๐(๐ด, ๐ ) ๐ โฆ→ (pr๐ โฆ๐ , pr๐ โฆ๐ ). Then ๐๐ด,๐ ,๐ is bijective with inverse given by (๐, โ) โฆ→ ๐ , where ๐ is the unique map ๐ด → ๐ × ๐ induced by the universal property of the product. Given ๐ : ๐ด 0 → ๐ด, โ : ๐ → ๐ 0 and ๐ : ๐ → ๐ 0 consider the diagram ๐(๐ด, ๐ × ๐ ) ๐๐ด,๐ ,๐ ๐(๐ด, ๐ ) × ๐(๐ด, ๐ ) ๐∗ (โ×๐)∗ (๐∗โ ∗ ,๐∗๐ ∗ ) ๐(๐ด 0, ๐ 0 × ๐ 0) ๐๐ด0,๐ 0,๐ 0 ๐(๐ด 0, ๐ 0) × ๐(๐ด 0, ๐ 0) . For any ๐ : ๐ด → ๐ × ๐ we have (๐∗โ ∗, ๐∗๐ ∗ ) โฆ ๐๐ด,๐ ,๐ (๐ ) = (โ โฆ pr๐ โฆ๐ โฆ ๐, ๐ โฆ pr๐ โฆ๐ โฆ ๐) and ๐๐ด0,๐ 0,๐ 0 โฆ ๐∗ (โ × ๐)∗ (๐ ) = (pr๐ 0 โฆ(โ × ๐) โฆ ๐ โฆ ๐, pr๐ 0 โฆ(โ × ๐) โฆ ๐ โฆ ๐). Thus, proving naturality of ๐ reduces to proving commutativity of the diagrams ๐ ×๐ โ×๐ ๐0 ×๐0 ๐ ×๐ pr๐ 0 pr๐ ๐ โ โ×๐ pr๐ 0 pr๐ ๐0, ๐0 ×๐0 ๐ ๐ ๐0. But โ × ๐ is by definition the unique map such that the above diagrams commute. It follows that ๐๐ด,๐ ,๐ is an isomorphism ๐(๐ด, ๐ × ๐ ) ๐(๐ด, ๐ ) × ๐(๐ด, ๐ ) natural in ๐ด, ๐, ๐ ∈ ๐. 50 Solutions by positrón0802 5.3 Interactions between functors and limits Exercise 5.3.10. Let ๐น : ๐ → โฌ create limits. Let ๐ท : I → ๐ be a diagram. Then, for any limit ๐๐ผ ๐๐ผ cone (๐ต −→ ๐น ๐ท (๐ผ ))๐ผ ∈I on ๐น ๐ท, there is a unique cone (๐ด −→ ๐ท (๐ผ ))๐ผ ∈I on ๐ท such that ๐น (๐ด) = ๐ต and ๐๐ผ ๐๐ผ ๐น (๐ ๐ผ ) = ๐๐ผ for all ๐ผ ∈ I, and the cone (๐ด −→ ๐ท (๐ผ ))๐ผ ∈I is a limit cone. In particular, if (๐ด −→ ๐ท (๐ผ ))๐ผ ∈I ๐น (๐ ๐ผ ) ๐๐ผ is a cone on ๐ท such that (๐น (๐ด) −−−−→ ๐น ๐ท (๐ผ ))๐ผ ∈I is a limit cone on ๐น ๐ท, then (๐ด −→ ๐ท (๐ผ ))๐ผ ∈I is the unique cone given by the property of creating limits, and hence it is a limit cone. Therefore ๐น reflects limits. Exercise 5.3.11. (a) If ๐ท : I → Set is a diagram then ๐ข − ๐ฝ in I} ๐ฟ = lim ๐ท {(๐ฅ ๐ผ )๐ผ ∈I | ๐ฅ ๐ผ ∈ ๐ท (๐ผ ) for all ๐ผ ∈ I and (๐ท๐ข) (๐ฅ ๐ผ ) = ๐ฅ ๐ฝ for all ๐ผ → →I ๐๐ฝ is the limit cone on ๐ท with projections ๐ฟ −−→ ๐ท (๐ฝ ) given by ๐ ๐ฝ ((๐ฅ ๐ผ )๐ผ ∈I ) = ๐ฅ ๐ฝ for all ๐ฝ ∈ I. This formula was given in Example 5.1.22 and proven in Exercise 5.1.37. ๐๐ผ Let ๐ท : I → Grp be a diagram in Grp and (๐ต −→ ๐ ๐ท (๐ผ ))๐ผ ∈I be a limit cone on ๐ ๐ท in Set. Then ๐ข − ๐ฝ in I} ∈ Set ๐ต 0 = {(๐ฅ ๐ผ )๐ผ ∈I | ๐ฅ ๐ผ ∈ ๐ ๐ท (๐ผ ) for all ๐ผ ∈ I and (๐ ๐ท๐ข) (๐ฅ ๐ผ ) = ๐ฅ ๐ฝ for all ๐ผ → ๐0๐ฝ is also a limit cone on ๐ ๐ท with projections −−→ ๐ ๐ท (๐ฝ ) given by ๐ 0๐ฝ ((๐ฅ ๐ผ )๐ผ ∈I ) = ๐ฅ ๐ฝ for all ๐ฝ ∈ I, so there is a unique isomorphism โ : ๐ต → ๐ต 0 such that ๐ 0๐ฝ โ = ๐ ๐ฝ for all ๐ฝ ∈ I. Î Endow the set ๐ต 0 with a canonical group structure as a subgroup of ๐ ∈I ๐ท (๐ผ ) ∈ Grp, call this group ๐ด 0, and let ๐ ๐ฝ0 : ๐ด 0 → ๐ท (๐ฝ ), for ๐ฝ ∈ I, denote the projection homomorphism whose underlying set function is ๐ 0๐ฝ . Now, the bijection โ : ๐ต → ๐ต 0 endows ๐ต with a group structure, and call this group ๐ด, so that โ : ๐ด → ๐ด 0 is an isomorphism of groups. For ๐ฝ ∈ I, let ๐ ๐ฝ : ๐ด → ๐ท (๐ฝ ) be the projection, so that its underlying set function is ๐ ๐ฝ . Then ๐ (๐ด) = ๐ต and ๐ (๐ ๐ผ0 ) = ๐ ๐ผ for all ๐ผ ∈ I, and clearly (๐ด, (๐ ๐ผ0 )๐ผ ∈I ) is unique as such. ๐ต0 ๐ ๐ผ0 ๐๐ผ It remains to show that (๐ด −→ ๐ท (๐ผ ))๐ผ ∈I is a limit cone on ๐ท. So let (๐ถ − → ๐ท (๐ผ ))๐ผ ∈I be a cone ๐ (๐๐ผ ) on ๐ท. Then (๐ (๐ถ) −−−−→ ๐ ๐ท (๐ผ ))๐ผ ∈I is a cone on ๐ ๐ท, so there is a unique map ๐ : ๐ (๐ถ) → ๐ต such that ๐ ๐ผ ๐ = ๐ (๐๐ผ ) for all ๐ผ ∈ I. What we must prove is that ๐ is a group homomorphism as a map ๐ถ → ๐ด. But composing with โ gives a map โ๐ : ๐ (๐ถ) → ๐ต 0 which is the unique one given by the universal property. This โ๐ is a group homomorphism when viewed as a map ๐ถ → ๐ด 0 . Since we defined the group structure on ๐ด declaring โ to be an isomorphism of groups, it follows that ๐ : ๐ถ → ๐ด is indeed a homomorphism, and is of course the unique such that ๐ ๐ผ0๐ = ๐๐ผ for all ๐ผ ∈ I. ๐ ๐ผ0 Thus (๐ด −→ ๐ท (๐ผ ))๐ผ ∈I is a limit cone on ๐ท. We conclude that ๐ : Grp → Set creates limits. (b) The above proof works when Grp is replaced by Ab, Ring or Vect๐ . ๐๐ผ Exercise 5.3.12. Let ๐ท : I → ๐ be a diagram. Then there exists a limit cone (๐ต −→ ๐น ๐ท (๐ผ ))๐ผ ∈I on ๐๐ผ ๐น ๐ท. Since ๐น creates limits there is a unique cone (๐ด −→ ๐ท (๐ผ ))๐ผ ∈I such that ๐น (๐ด) = ๐ต, ๐น (๐ ๐ผ ) = ๐๐ผ ๐๐ผ for all ๐ผ ∈ I, and (๐ด −→ ๐ท (๐ผ ))๐ผ ∈I is a limit cone on ๐ท. It follows ๐ has limits of shape I. 51 Solutions by positrón0802 5.3 Interactions between functors and limits ๐๐ผ ๐๐ผ Now suppose (๐ด −→ ๐ท (๐ผ ))๐ผ ∈I is a limit cone on ๐ท. There exists a limit cone (๐ต −→ ๐น ๐ท (๐ผ ))๐ผ ∈I ๐ ๐ผ0 ๐ ๐ผ0 on ๐น ๐ท, and a unique cone (๐ด 0 −→ ๐ท (๐ผ ))๐ผ ∈I such that ๐น (๐ด 0) = ๐ต, ๐น (๐ ๐ผ0 ) = ๐๐ผ for all ๐ผ ∈ I, and (๐ด 0 −→ ๐ท (๐ผ ))๐ผ ∈I is a limit cone on ๐ท. Thus there is a unique isomorphism โ : ๐ด 0 → ๐ด such that ๐ ๐ผ โ = ๐ ๐ผ0 ๐น (๐ ๐ผ ) for all ๐ผ ∈ I. Then ๐น (โ) : ๐ต → ๐น (๐ด) is an isomorphism. Consider the cone (๐น (๐ด) −−−−→ ๐น ๐ท (๐ผ ))๐ผ ∈I . ๐๐ผ Given any cone (๐ถ − → ๐น ๐ท (๐ผ ))๐ผ ∈I, there is a unique ๐ : ๐ถ → ๐ต such that ๐๐ผ ๐ = ๐๐ผ for all ๐ผ ∈ I. Then ๐น (โ)๐ : ๐ถ → ๐น (๐ด) satisfies ๐น (๐ ๐ผ )๐น (โ)๐ = ๐น (๐ ๐ผ0 ) ๐ = ๐๐ผ for all ๐ผ ∈ I. Moreover, if โ 0 : ๐ถ → ๐น (๐ด) satisfies ๐น (๐ ๐ผ )โ 0 = ๐๐ผ for each ๐ผ then ๐น (โ) −1โ 0 satisfies ๐๐ผ ๐น (โ) −1โ 0 = ๐น (๐ ๐ผ0โ −1 )โ 0 = ๐น (๐ ๐ผ )โ 0 = ๐๐ผ , so we must have ๐น (โ) −1โ 0 = ๐ by uniqueness of ๐ , i.e. โ 0 = ๐น (โ)๐ is unique. It follows that ๐น (๐ ๐ผ ) (๐น (๐ด) −−−−→ ๐น ๐ท (๐ผ ))๐ผ ∈I is a limit cone on ๐น ๐ท. We conclude that ๐น preserves limits. Exercise 5.3.13. (a) Let ๐ ∈ Set be arbitrary and let ๐ : ๐ต → ๐ต 0 be epic in โฌ. Then ๐บ (๐ ) : ๐บ (๐ต) → ๐บ (๐ต 0) is epic in Set. By Exercise 5.2.26(b), ๐บ (๐ ) has a right inverse, say โ : ๐บ (๐ต 0) → ๐บ (๐ต). Then the induced map ๐บ (๐ )∗ : Set(๐, ๐บ (๐ต)) → Set(๐, ๐บ (๐ต 0)) has a right inverse, namely โ ∗ : Set(๐, ๐บ (๐ต 0)) → Set(๐, ๐บ (๐ต)). In particular, ๐บ (๐ )∗ is surjective. Since ๐บ a ๐น there is an isomorphism ๐๐,๐ต : โฌ(๐น (๐), ๐ต) → ๐(๐, ๐บ (๐ต)) natural in ๐ ∈ Set and ๐บ (๐ต), so that we have a commutative square โฌ(๐น (๐), ๐ต) ๐๐,๐ต Set(๐, ๐บ (๐ต)) ๐บ (๐ )∗ ๐∗ โฌ(๐น (๐), ๐ต 0) ๐๐,๐ต0 Set(๐, ๐บ (๐ต 0)) . Since ๐๐,๐ต , ๐๐,๐ต0 are isomorphisms and ๐บ (๐ )∗ is surjective, it follows that โฌ(๐น (๐), ๐ ) is surjective, i.e. it is epic. Therefore ๐น (๐) is projective. We conclude that ๐น (๐) is projective for all sets ๐. (b) Recall that a map in Ab is epic if and only if it is surjective (Example 5.2.19). Consider Z/2Z ∈ Ab. Then the unique non-trivial map ๐ : Z → Z/2Z is epic in Ab, but ๐∗ : Ab(Z/2, Z) → Ab(Z/2Z, Z/2Z) is not epic, for Ab(Z/2Z, Z) = 0 and Ab(Z/2Z, Z/2Z) Z/2Z. Thus Z/2Z is not projective in Ab. op (c) Let ๐ be a vector space and ๐ ∈ Vect๐ arbitrary. Let ๐ op : ๐ → ๐ be epic in Vect๐ , that is, ๐ : ๐ → ๐ is monic (i.e. injective) in Vect๐ . We shall prove that ๐ ∗ : Vect๐ (๐ , ๐ ) → Vect๐ (๐ , ๐ ) is surjective. Given a linear map ๐ฟ : ๐ → ๐ we need e ๐ฟ : ๐ → ๐ such that the diagram ๐ ๐ฟ ๐ ๐ ๐ e ๐ฟ commutes. Taking a basis {๐ฃ๐ }๐ ∈๐ฝ for ๐ then {๐ (๐ฃ๐ )}๐ ∈๐ฝ is linear independent in ๐ , so extends to a basis for ๐ . Then we may define e ๐ฟ by sending each ๐ (๐ฃ๐ ) to ๐ฟ(๐ฃ๐ ) and all other basis elements to 0. Thus ๐ is injective. Since ๐ was arbitrary, it follows that every ๐-vector space is injective. Now consider Z ∈ Ab and ๐ : Z → Q the inclusion map, which is monic. Then ๐ ∗ : Ab(Q, Z) → Ab(Z, Z) is not epic, for Ab(Q, Z) = 0 and Ab(Z, Z) = Z. It follows that Z is not injective in Ab. 52 Solutions by positrón0802 6. Adjoints, representables and limits 6 6.1 Adjoints, representables and limits Limits in terms of representables and adjoints Exercise 6.1.5. Let I be the discrete category with two object ๐ 0 and ๐ 1 . Let ๐ be a category. A functor ๐ท : I → ๐ is simply a pair of objects ๐ท (๐ 0 ) and ๐ท (๐ 1 ) of ๐. For ๐ด ∈ ๐, the diagonal functor Δ๐ → [I, ๐] = ๐ × ๐ is simply given by Δ(๐ด) = (๐ด, ๐ด). For ๐ท : I → ๐ and ๐ด ∈ ๐, a cone on ๐ท with vertex ๐ด is a pair of maps ๐ด → ๐ท (๐ 0 ) and ๐ด → ๐ท (๐ 1 ). Proposition 6.1.1 tells us that a limit of ๐ท is a product (๐ท (๐ 0 ) × ๐ท (๐ 1 ), (๐๐ : ๐ท (๐ 0 ) × ๐ท (๐ 1 ) → ๐ท (๐๐ ))๐=0,1 ); this is Example 5.1.21(a). Corollary 6.1.2 tells us that products are unique up to isomorphism. Lemma 6.1.3(a) tells us that given two products (๐ท (๐ 0 )×๐ท (๐ 1 ), (๐๐ )๐=0,1 ) and (๐ท 0 (๐ 0 )×๐ท 0 (๐ 1 ), (๐๐0 )๐=0,1 ) in ๐ and maps ๐ผ๐ : ๐ท (๐๐ ) → ๐ท 0 (๐๐ ), ๐ = 0, 1, there is a unique map ๐ผ : ๐ท (๐ 0 ) × ๐ท (๐ 1 ) → ๐ท 0 (๐ 0 ) × ๐ท 0 (๐ 1 ) such that ๐ผ๐๐0 = ๐๐ for ๐ = 0, 1. (cf. Exercise 5.3.8). Lemma 6.1.3(b) says that if we further have objects ๐ด, ๐ด 0 ∈ ๐ with maps ๐๐ : ๐ด → ๐ท (๐๐ ) and ๐๐0 : ๐ด 0 → ๐ท 0 (๐๐ ) for ๐ = 0, 1, and a map ๐ : ๐ด → ๐ด 0 such that ๐ผ๐ ๐๐ = ๐๐0๐ for ๐ = 0, 1, then ๐ผ ๐ = ๐ 0 ๐ , where ๐ and ๐ 0 are induced by the ๐๐ and the ๐๐0 respectively. Proposition 6.1.4 says that taking products is a functor (after taking choices, see Exercise 5.3.8) is right adjoint to the diagonal functor ๐ → ๐ × ๐. This was shown in Exercise 3.1.1. Exercise 6.1.6. If I = ๐บ is a group (with unique object ∗), then [๐บ, Set] is the category of left ๐บ-sets and ๐บ-equivariant maps (Example 1.3.4). The diagonal functor Δ : Set → [๐บ, Set] equips a set ๐ด with the trivial left ๐บ-action, such that ๐ · ๐ = ๐ for all ๐ ∈ ๐บ and ๐ ∈ ๐ด. Let ๐ท : ๐บ → Set be a left ๐บ-set and write ๐ = ๐ท (∗). A cone on ๐ท consists of a set ๐ together with a function ๐ : ๐ → ๐ such that ๐ · ๐ (๐ฅ) = ๐ (๐ฅ) for all ๐ฅ ∈ ๐ and ๐ ∈ ๐บ, that is, the image of ๐ is contained in the ๐บ-fixed point subset ๐ ๐บ = {๐ฅ ∈ ๐ | ๐ · ๐ฅ = ๐ฅ for all ๐ ∈ ๐บ } of ๐ . It follows that ๐ ๐บ together with the inclusion ๐ : ๐ ๐บ → ๐ is a limit cone on ๐ท. Proposition 6.1.4 tells us that the functor (−)๐บ : [๐บ, Set] → Set sending a ๐บ-set ๐ to its ๐บ-fixed point subset ๐ ๐บ is right adjoint to the functor Δ : Set → [๐บ, Set] which equips a set with the trivial left ๐บ-action. Now we consider the dual. A cocone on ๐ท consists of a set ๐ together with a function ๐ : ๐ → ๐ such that ๐ (๐ · ๐ฅ) = ๐ฅ for all ๐ ∈ ๐บ and ๐ฅ ∈ ๐, that is, ๐ factors through the set ๐ /๐บ = {๐บ · ๐ฅ | ๐ฅ ∈ ๐ } of orbits of ๐ under ๐บ . It follows that ๐ /๐บ together with the quotient ๐ : ๐ → ๐ /๐บ is a colimit of ๐ท. The dual of Proposition 6.1.4 tells us that the functor (−)/๐บ : [๐บ, Set] → Set sending a ๐บ-set ๐ to its set ๐ /๐บ of orbits under ๐บ is left adjoint to the functor Δ : Set → [๐บ, Set] which endows a set with the trivial left ๐บ-action. 6.2 Limits and colimits of presheaves Exercise 6.2.20. (a) Let ๐ผ : A → ๐ฎ be a natural transformation. First assume that each ๐ผ๐ด, ๐ด ∈ A, is monic in ๐ฎ. If ๐ฝ, ๐พ : A → ๐ฎ are natural transformations such that ๐ผ๐ฝ and ๐ผ๐พ are defined and equal, so that ๐ผ๐ด ๐ฝ๐ด = ๐ผ๐ด๐พ๐ด for all ๐ด ∈ ๐, we have ๐ฝ๐ด = ๐พ๐ด for all ๐ด as each ๐ผ๐ด is monic. Thus ๐ฝ = ๐พ and it follows that ๐ผ is monic in [A, ๐ฎ]. 53 Solutions by positrón0802 6.2 Limits and colimits of presheaves Conversely, assume that ๐ผ is monic in [A, ๐ฎ]. By Lemma 5.1.32, the square ๐ 1๐ ๐ 1๐ ๐ผ ๐ ๐ผ ๐ is a pullback in [A, ๐ฎ]. Since ๐ฎ has pullbacks, by Corollary 6.2.6 it follows that ๐ (๐ด) 1๐ (๐ด) 1๐ (๐ด) ๐ (๐ด) ๐ผ๐ด ๐ (๐ด) ๐ผ๐ด ๐ (๐ด) is a pullback in ๐ฎ for all ๐ด ∈ ๐. By Lemma 5.1.32 again, we deduce that each ๐ผ๐ด, ๐ด ∈ ๐, is monic in ๐ฎ. (b) By part (a), ๐ผ ∈ [Aop, Set] is monic (respectively epic) if and only if ๐ผ๐ด is monic (respectively epic) in Set for all ๐ด ∈ A. (c) Let ๐ผ : ๐ ⇒ ๐ be a natural transformation between functors Aop → Set. If each ๐ผ๐ด is monic (epic), then the proof given in (a) that ๐ผ is monic (epic) in [Aop, Set] did not rely on the fact that (co)limits of presheaves are computed pointwise. It remains to prove that if ๐ผ is monic (epic), then so is ๐ผ๐ด for each ๐ด ∈ Aop, without relying on this fact. We consider the case of ๐ผ monic, the case of ๐ผ epic is dual. Suppose for the sake of contradiction that there exists ๐ด ∈ A such that ๐ผ๐ด is not monic. Then there exist ๐, ๐ 0 ∈ ๐ (๐ด) such that ๐ ≠ ๐ 0 and ๐ผ๐ด (๐) = ๐ผ๐ด (๐ 0) ∈ ๐ (๐ด). By the Yoneda Lemma, we have an isomorphism ๐ (๐ด) [Aop, Set] (๐ป๐ด, ๐ ) under which ๐ and ๐ 0 correspond to functors ๐ฝ : ๐ป๐ด ⇒ ๐ and ๐ฝ 0 : ๐ป๐ด ⇒ ๐, respectively, determined uniquely by the conditions ๐ฝ๐ด (1๐ด ) = ๐ and ๐ฝ๐ด0 (1๐ด ) = ๐ 0 . Thus ๐ฝ ≠ ๐ฝ 0 and ๐ผ๐ฝ = ๐ผ๐ฝ 0, contradicting the fact that ๐ผ is monic. We deduce that if ๐ผ is monic in [Aop, Set], then ๐ผ๐ด is monic in Set for all ๐ด ∈ Aop . Exercise 6.2.21. We will write the sum ๐ + ๐ in coproduct notation ๐ q ๐ . By definition, given functor ๐, ๐ : ๐ op → Set then ๐ q๐ is the functor ๐ op → Set given by (๐ q๐ ) (๐ด) = ๐ (๐ด)q๐ (๐ด), the disjoint union of ๐๐ด and ๐๐ด, for all ๐ด ∈ ๐, and similarly (๐ q ๐ ) (๐ ) = ๐ (๐ ) q ๐ (๐ ) on morphisms. Suppose that ๐ด ∈ ๐ is such that there exists an isomorphism ๐ผ : ๐ป๐ด → ๐ q ๐ . Then for every ๐ต ∈ ๐, the function ๐ผ ๐ต : ๐(๐ต, ๐ด) → ๐ (๐ต) q ๐ (๐ต) is bijective. Considering 1๐ด ∈ ๐(๐ด, ๐ด), we must have either 1๐ด ∈ ๐ผ −1 (๐ (๐ด)) or 1๐ด ∈ ๐ผ๐ด−1 (๐ (๐ด)). Suppose that 1๐ด ∈ ๐ผ๐ด−1 (๐ (๐ด)); we will prove that ๐ is constant with value ∅. Let ๐ต ∈ ๐. If ๐(๐ต, ๐ด) = ∅, then ๐ (๐ต) = ∅ as ๐ผ ๐ต is bijective. So assume ๐(๐ต, ๐ด) ≠ ∅ and let ๐ : ๐ต → ๐ด be any map. SBy naturality of ๐ผ, the following diagram is commutative: ๐ผ๐ด ๐(๐ด, ๐ด) ๐ (๐ด) q ๐ (๐ด) ๐∗ ๐(๐ต, ๐ด) ๐ (๐ ) q๐ (๐ ) ๐ผ๐ด ๐ (๐ต) q ๐ (๐ต) . 54 Solutions by positrón0802 6.2 Limits and colimits of presheaves Since 1๐ด ∈ ๐ผ๐ด−1 (๐ (๐ด)), it follows that ๐ผ๐ด (๐ ) ∈ ๐ (๐ต). Since ๐ผ๐ด is a bijection and ๐ ∈ ๐(๐ต, ๐ด) was arbitrary, we deduce that ๐ (๐ต) = ∅. We conclude that ๐ is the constant functor ∅. Similarly, if 1๐ด ∈ ๐ผ๐ด−1 (๐ (๐ด)) then ๐ is constant with value ∅. (b) By part (a), it suffices to prove that no representable functor is constant with value ∅. For this, note that for an object ๐ด in a locally small category ๐ we have ๐ป๐ด (๐ด) = ๐(๐ด, ๐ด) ≠ ∅ as 1๐ด ∈ ๐(๐ด, ๐ด). Exercise 6.2.22. Let A be a category and ๐ : Aop → Set a functor. Let {∗} be a one-point set and ๐น : 1 → Set the functor sending the unique object of 1 to {∗}. Then the category of element E(๐ ) of ๐ is precisely the comma category (๐น ⇒ ๐ ). Indeed, an object of (๐น ⇒ ๐ ) is a pair consisting of an object ๐ด ∈ ๐ and an arrow ๐ฅ : {∗} → ๐ (๐ด), the latter being the same as an object ๐ฅ ∈ ๐ (๐ด). An arrow in (๐น ⇒ ๐ ) from (๐ด 0, ๐ฅ 0) to (๐ด, ๐ฅ) is an arrow ๐ : ๐ด 0 → ๐ด in ๐ such that ๐ (๐ ) โฆ๐ฅ = ๐ฅ 0, that is, when ๐ฅ and ๐ฅ 0 are considered as objects of ๐ (๐ด) and ๐ (๐ด 0) respectively, we have ๐ (๐ ) (๐ฅ) = ๐ฅ 0 . Exercise 6.2.23. Let ๐ be a presheaf on a locally small category ๐. First assume that ๐ is representable; we will prove that E(๐ ) h as a terminal object. It suffices to consider the case ๐ = ๐ป๐ด for some ๐ด ∈ ๐. In this case (๐ด, 1๐ด ) is terminal in E(๐ป๐ด ), for given (๐ต, ๐ : ๐ต → ๐ด) ∈ E(๐ป๐ด ) then ๐ : ๐ต → ๐ด is an arrow such that ๐ ∗ (1๐ด ) = ๐ , the unique as such. Conversely, suppose that E(๐ ) has a terminal object (๐ด, ๐ฅ). This is an object ๐ด ∈ ๐ and an element ๐ฅ ∈ ๐ (๐ด) such that for any ๐ต ∈ ๐ and ๐ฆ ∈ ๐ (๐ต), there exists a unique map ๐ : ๐ต → ๐ด such that ๐ (๐ ) (๐ฅ) = ๐ฆ. It follows from Corollary 4.3.2 that ๐ is representable. Exercise 6.2.24. Let A be a small category and ๐ a presheaf on A. Consider the category B = E(๐ ) of elements of ๐ . Since A is small, so is E(๐ ). We claim there is an equivalence of categories [Aop, Set]/๐ ' [E(๐ ) op, Set]. By Proposition 1.3.18, it suffices to find a functor ๐น : [Aop, Set]/๐ → [E(๐ ) op, Set] full, faithful and essentially surjective on objects. Let (๐ , ๐ผ) be an object of [Aop, Set]/๐, that is, ๐ is a presheaf on A and ๐ผ is a natural transformation ๐ ⇒ ๐ . Define ๐น (๐, ๐ผ) : E(๐ ) op → Set as follows. Given (๐ด, ๐ฅ) ∈ E(๐ ), let ๐น (๐, ๐ผ) send (๐ด, ๐ฅ) to the subset ๐ผ๐ด−1 (๐ฅ) of ๐ (๐ด). If ๐ : (๐ด, ๐ฅ) → (๐ต, ๐ฆ) is an arrow in E(๐ ) and ๐ง ∈ ๐ผ ๐ต−1 (๐ฆ) ⊂ ๐ (๐ต), we claim ๐ (๐ ) : ๐ (๐ต) → ๐ (๐ด) sends ๐ง to an element of ๐ผ๐ด−1 (๐ฅ). Indeed, ๐ผ๐ด โฆ ๐ (๐ ) (๐ง) = ๐ (๐ ) โฆ ๐ผ ๐ต (๐ง) = ๐ (๐ ) (๐ฆ) = ๐ฅ by naturality of ๐ผ and the fact that ๐ is an arrow (๐ด, ๐ฅ) → (๐ต, ๐ฆ) in E(๐ ). It follows that ๐น (๐ , ๐ผ) is a functor E(๐ ) op → Set, given on morphisms by sending ๐ : (๐ด, ๐ฅ) → (๐ต, ๐ฆ) to the function ๐ผ ๐ต−1 (๐ฆ) → ๐ผ๐ด−1 (๐ฅ) obtained by restricting the domain and codomain of ๐ (๐ ). This defines a map ๐น : [Aop, Set]/๐ → [E(๐ ) op, Set] on objects, now we define ๐น on morphisms. Given a map ๐ : (๐ , ๐ผ) → (๐ 0, ๐ผ 0) in [Aop, Set]/๐, then for each ๐ด ∈ ๐, ๐ ๐ด is a function ๐ (๐ด) → ๐ 0 (๐ด). Let (๐ด, ๐ฅ) ∈ E(๐ ) and ๐ง ∈ ๐ผ๐ด−1 (๐ฅ). Then ๐ผ๐ด0 โฆ ๐ ๐ด (๐ง) = ๐ผ๐ด (๐ง) = ๐ฅ, so that ๐ ๐ด (๐ง) ∈ (๐ผ๐ด0 ) −1 (๐ฅ). Thus, we obtain a function ๐น (๐ )(๐ด,๐ฅ) : ๐ผ๐ด−1 (๐ฅ) → (๐ผ๐ด0 ) −1 (๐ฅ) by restricting the domain and codomain of ๐ ๐ด . Moreover, given a map ๐ : (๐ด, ๐ฅ) → (๐ต, ๐ฆ) in E(๐ ) then the square on the left below commutes, being obtained from the diagram on the right by restricting 55 Solutions by positrón0802 6.2 Limits and colimits of presheaves domains and codomains: ๐ผ ๐ต−1 (๐ฆ) ๐น (๐ ) (๐ต,๐ฆ) (๐ผ ๐ต0 ) −1 (๐ฆ) ๐น (๐ 0,๐ผ 0 ) (๐ ) ๐น (๐ ,๐ผ) ( ๐ ) ๐ผ๐ด−1 (๐ฅ) ๐ (๐ต) ๐น (๐ ) (๐ด,๐ฅ ) ๐๐ด ๐ 0 (๐ ) ๐ (๐ ) (๐ผ๐ด0 ) −1 (๐ฅ) , ๐ 0 (๐ต) ๐ (๐ด) ๐๐ต ๐ 0 (๐ด) . Thus ๐น (๐ ) is a natural transformation ๐น (๐ , ๐ผ) ⇒ ๐น (๐ 0, ๐ผ 0). It follows that ๐น is a functor from [Aop, Set]/๐ to [E(๐ ) op, Set]. First we show that ๐น is essentially surjective on objects. Let ๐ be a presheaf on E(๐ ). Define ๐ : Aop → Set as follows. Let ๐ (๐ด) = q๐ฅ ∈๐ (๐ด)๐ (๐ด, ๐ฅ) on objects ๐ด ∈ ๐. If ๐ : ๐ด → ๐ต is a morphism in ๐, then for each ๐ฆ ∈ ๐ (๐ต) it is a morphism ๐ : (๐ด, ๐ฅ) → (๐ต, ๐ฆ), where ๐ฅ = ๐ (๐ ) (๐ฆ), so let ๐ (๐ ) : q๐ฆ ∈๐ (๐ต) ๐ (๐ต, ๐ฆ) → q๐ฅ ∈๐ (๐ด)๐ (๐ด, ๐ฅ) be the function whose restriction summand ๐ (๐ต, ๐ฆ), ๐ฆ ∈ ๐ (๐ต), is the map ๐ (๐ ) : ๐ (๐ต, ๐ฆ) → ๐ (๐ด, ๐ (๐ ) (๐ฆ)) follows by the inclusion into q๐ฅ ∈๐ (๐ด)๐ (๐ด, ๐ฅ). Since ๐ is a functor, so is ๐ . Moreover, for each ๐ด ∈ ๐ we have a function ๐ผ๐ด : ๐ (๐ด) → ๐ (๐ด) whose restriction to the summand๐ (๐ด, ๐ฅ), ๐ฅ ∈ ๐ (๐ด), is constant with value ๐ฅ . By definition of ๐ on morphisms, for ๐ : ๐ด → ๐ต a map in ๐, the diagram ๐ผ๐ต q๐ฆ ∈๐ (๐ต)๐ (๐ต, ๐ฆ) ๐ (๐ต) ๐ (๐ ) ๐ (๐ ) q๐ฅ ∈๐ (๐ด)๐ (๐ด, ๐ฅ) ๐ผ๐ด ๐ (๐ด) commutes. Thus ๐ผ is a natural transformation ๐ ⇒ ๐, and therefore (๐ , ๐ผ) is an object of [Aop, Set]/๐ . Moreover, by definition ๐น (๐ , ๐ผ) sends an object (๐ด, ๐ฅ) ∈ E(๐ ) to ๐ผ๐ด−1 (๐ฅ) = ๐ (๐ด, ๐ฅ) and a morphism ๐ : (๐ด, ๐ฅ) → (๐ต, ๐ฆ) to ๐ (๐ ), so that ๐น (๐ , ๐ผ) = ๐ . It follows that ๐น is surjective on objects. It is left to prove that ๐น is full and faithful. For this purpose, let (๐ , ๐ผ), (๐ 0, ๐ผ 0) ∈ [Aop, Set]/๐ . We shall prove that given a natural transformation ๐ : ๐น (๐ , ๐ผ) ⇒ ๐น (๐ 0, ๐ผ 0), there is a unique map ๐ : (๐, ๐ผ) → (๐ 0, ๐ผ 0) such that ๐น (๐ ) = ๐. To show this, note that for ๐ด ∈ ๐, the set ๐ (๐ด) is the disjoint union q๐ฅ ∈๐ (๐ด) ๐ผ๐ด−1 (๐ฅ) (and similarly for ๐ 0) so there is precisely one way to define ๐ ๐ด : ๐ (๐ด) → ๐ 0 (๐ด) so that ๐น (๐ )(๐ด,๐ฅ) = ๐ (๐ด,๐ฅ) for all (๐ด, ๐ฅ) ∈ E(๐ ), namely the restriction of ๐ ๐ด to each summand ๐ผ๐ด−1 (๐ฅ), ๐ฅ ∈ ๐ (๐ด), of ๐ (๐ด) must be the function ๐ (๐ด,๐ฅ) : ๐ผ๐ด−1 (๐ฅ) → (๐ผ๐ด0 ) −1 (๐ฅ) followed by the inclusion (๐ผ๐ด0 ) −1 (๐ฅ) ⊂ ๐ 0 (๐ด). Moreover, ๐ : ๐ ⇒ ๐ 0 is natural as ๐ is natural, and is a map (๐ , ๐ผ) → (๐ 0, ๐ผ 0). It follows that ๐น is full and faithful. We conclude that ๐น is an equivalence of categories [Aop, Set]/๐ ' [E(๐ ) op, Set]. Exercise 6.2.25. (a) We must define Lan๐น ๐ on morphisms. Let ๐ : ๐ต → ๐ต 0 be a map in B. For (๐ด, โ) ∈ (๐น ⇒ ๐ต), let ๐ (๐ด,โ) : ๐ (๐ด) → Lan๐น ๐ (๐ต) be the canonical map into the colimit, and similarly for (๐ด, ๐) ∈ (๐น ⇒ ๐ต 0) let ๐ (๐ด,๐) : ๐ (๐ด) → Lan๐น ๐ (๐ต 0) be the canonical map into 56 Solutions by positrón0802 6.2 Limits and colimits of presheaves the colimit. Given (๐ด, โ) ∈ (๐น ⇒ ๐ต), then (๐ด, ๐ โ : ๐น (๐ด) → ๐ต 0) ∈ (๐น ⇒ ๐ต 0). Furthermore, if ๐ : (๐ด, โ) → (๐ด 0, โ 0) is a map in (๐น ⇒ ๐ต), then ๐ is also a map (๐ด, ๐ โ) → (๐ด 0, ๐ โ 0), so ๐ (๐ด0,๐ โ0) โฆ ๐ (๐) = ๐ (๐ด,๐ โ) . Thus Lan๐น ๐ (๐ต 0) together with the maps ๐ (๐ด,๐ โ) is a cocone on ๐ โฆ ๐๐ต . By the universal property of the colimit, there is a unique map Lan๐น ๐ (๐ ) : Lan๐น ๐ (๐ต) → Lan๐น ๐ (๐ต 0) such that Lan๐น ๐ (๐ ) โฆ ๐ (๐ด,โ) = ๐ (๐ด,๐ โ) for all (๐ด, โ) ∈ (๐น ⇒ ๐ต). Now let ๐ : B → ๐ฎ be a functor. We want to prove that there is a canonical bijection [B, ๐ฎ] (Lan๐น ๐, ๐ ) [A, ๐ฎ] (๐, ๐ ๐น ). Let ๐ผ : Lan๐น ๐ ⇒ ๐ be a natural transformation. Then, for each ๐ด ∈ A, we have a map ๐ผ ๐น (๐ด) : Lan๐น ๐ (๐น (๐ด)) → ๐ ๐น (๐ด), or to say the same thing, a cocone on ๐ โฆ ๐ ๐น (๐ด) with vertex ๐ ๐น (๐ด), say with structure maps ๐ฝ ๐ด(๐ถ,โ) : ๐ (๐ถ) → ๐ ๐น (๐ด) for (๐ถ, โ) ∈ (๐น ⇒ ๐น (๐ด)). In particular we have a map ๐ผe๐ด = ๐ฝ ๐ด(๐ด,1 ) : ๐ (๐ด) → ๐ ๐น (๐ด). We want to ๐น (๐ด) show that the family ๐ผe๐ด, ๐ด ∈ A, defines a natural transformation. So let ๐ : ๐ด → ๐ด 0 be a map in A. Then ๐ is a map (๐ด, ๐น (๐ )) → (๐ด 0, 1๐น (๐ด0) ), so by definition of cocone the following diagram commutes: ๐ (๐ด) ๐ (๐ ) ๐ (๐ด 0) 0 ๐ฝ๐ด (๐ด0,1 0 ๐ฝ๐ด (๐ด,๐น (๐ ) ) ๐น (๐ด0 ) ) ๐ ๐น (๐ด 0) , 0 that is, ๐ฝ ๐ด(๐ด,๐น (๐ )) = ๐ผe๐ด0 โฆ ๐ (๐ ). Now consider the diagram ๐ฝ๐ด (๐ด,1 ๐ (๐ด) ๐น (๐ด) ) Lan๐น ๐ (๐น (๐ด)) ๐ผ ๐น (๐ด) Lan๐น ๐ (๐น (๐ )) ๐ (๐ด) ๐ ๐น (๐ด) ๐ ๐น (๐ ) Lan๐น ๐ (๐น (๐ด 0)) ๐ผ ๐น (๐ด0 ) ๐ ๐น (๐ด 0) 0 ๐ฝ๐ด (๐ด,๐น ( ๐ ) ) where the two horizontal morphisms on the left are the canonical inclusions into the colimit. The left-hand side square commutes by definition of Lan๐น ๐ on morphisms, and the right-hand side 0 square commutes by naturality of ๐ผ . Thus ๐ฝ ๐ด(๐ด,๐น (๐ )) = ๐ ๐น (๐ ) โฆ ๐ฝ ๐ด(๐ด,1 ) = ๐ ๐น (๐ ) โฆ ๐ผe๐ด . It follows ๐น (๐ด) that ๐ผe is a natural transformation ๐ ⇒ ๐ ๐น . We thus have defined a map [B, ๐ฎ] (Lan๐น ๐, ๐ ) → [A, ๐ฎ] (๐, ๐ ๐น ). We now define a map in the other direction. Let ๐ : ๐ ⇒ ๐ ๐น be a natural transformation. For ๐ต ∈ โฌ, we shall define a map e ๐๐ต : Lan๐น ๐ (๐ต) → ๐ (๐ต). Given (๐ด, โ) ∈ (๐น ⇒ ๐ต), consider ๐ (โ) โฆ ๐๐ด : ๐ (๐ด) → ๐ (๐ต). Given a map ๐ : (๐ด, โ) → (๐ด 0, โ 0) in (๐น ⇒ ๐ต), the triangle in the 57 Solutions by positrón0802 6.2 Limits and colimits of presheaves diagram ๐ (๐ด) ๐๐ด ๐ ๐น (๐ด) ๐ (๐ ) ๐ (โ) ๐ ๐น (๐ ) ๐ (๐ด 0) ๐ ๐น (๐ด 0) ๐๐ด0 ๐ (๐ต) ๐ (โ0 ) commutes by definition, and the square commutes by naturality of ๐. It follows that ๐ (๐ต) together with the collection of maps ๐ (โ) โฆ ๐๐ด, for (๐ด, โ) ∈ (๐น ⇒ ๐ต), is a cocone on ๐ โฆ ๐๐ต , and we let e ๐๐ต : Lan๐น ๐ (๐ต) → ๐ (๐ต) be the unique map induced by the universal property of the colimit Lan๐น ๐ (๐ต). We claim that e ๐ is a natural transformation. Indeed, to prove this it suffices to prove that the diagram ๐ (๐ด) ๐ (โ)โฆ๐๐ด ๐ (๐ต) ๐ (๐) ๐ (๐โ)โฆ๐๐ด ๐ (๐ต 0) commutes for all ๐ : ๐ต → ๐ต 0 in B and (๐ด, โ) ∈ (๐น ⇒ ๐ต), and this is clear. We thus have defined a map [A, ๐ฎ] (๐, ๐ ๐น ) → [B, ๐ฎ] (Lan๐น ๐, ๐ ), ๐ โฆ→ e ๐ , and looking at the definition it is easily checked that this map is a two-sided inverse of the map [B, ๐ฎ] (Lan๐น ๐, ๐ ) → [A, ๐ฎ] (๐, ๐ ๐น ), ๐ผ โฆ→ ๐ผe, defined above. (b) First, note that Lan๐น is a functor [A, ๐ฎ] → [B, ๐ฎ] as follows. Let ๐ ⇒ ๐ 0 → ๐ be a natural transformation between functors A → ๐ฎ. Let ๐ต ∈ A. For each (๐ด, โ) ∈ (๐น ⇒ ๐ต), let ๐ (๐ด,โ) : ๐ (๐ด) → Lan๐น ๐ (๐ต) denote the inclusion into the colimit. Given a map ๐ : (๐ด, โ) → (๐ด 0, โ 0) in (๐น ⇒ ๐ต), we have a commutative diagram ๐ 0 (๐ด) ๐๐ด ๐ 0 (๐ ) ๐ (๐ด) ๐ (๐ด,โ) ๐ (๐ ) ๐ 0 (๐ด 0) ๐๐ด0 ๐ (๐ด 0) ๐ (๐ด0,โ0 ) Lan๐น ๐ (๐ต) . Thus Lan๐น ๐ (๐ต) together with the collection of maps ๐ (๐ด,โ) โฆ๐๐ด, for (๐ด, โ) ∈ (๐น ⇒ ๐ต), is a cocone on ๐ 0 โฆ ๐๐ต . Let Lan๐น (๐)๐ต : Lan๐น ๐ 0 (๐ต) → Lan๐น ๐ (๐ต) be the unique map induced by the universal property of the colimit Lan๐น ๐ 0 (๐ต). For proving Lan๐น (๐) : Lan๐น ๐ 0 ⇒ Lan๐น ๐ is natural, for a map ๐ : ๐ต → ๐ต 0 in B we shall prove that the diagram Lan๐น ๐ 0 (๐ต) Lan๐น Lan๐น (๐)๐ต Lan๐น ๐ 0 (๐ต 0) Lan๐น ๐ (๐ต) Lan๐น ๐ (๐) ๐ 0 (๐) Lan๐น (๐)๐ต0 Lan๐น ๐ (๐ต 0) commutes. Let (๐ด, โ) ∈ (๐น ⇒ ๐ต), and let ๐ (๐ด,โ) : ๐ 0 (๐ด) → Lan๐น ๐ 0 (๐ต) and ๐ (๐ด,โ) : ๐ (๐ด) → Lan๐น ๐ (๐ต 0) denote the inclusions into the respective colimits. It then follows from the definitions 58 Solutions by positrón0802 6.2 that Limits and colimits of presheaves Lan๐น (๐)๐ต0 โฆ Lan๐น ๐ 0 (๐) โฆ ๐ (๐ด,โ) = ๐ (๐ด,โ) โฆ ๐๐ด = Lan๐น ๐ (๐) โฆ Lan๐น (๐) โฆ ๐ (๐ด,โ) and thus, as these holds for all (๐ด, โ), we deduce that the diagram above commutes. Therefore Lan๐น (๐) : Lan๐น ๐ 0 ⇒ Lan๐น ๐ is natural. Now, in part (a) we have defined for ๐ ∈ [A, ๐ฎ] and ๐ ∈ [B, ๐ฎ] a bijection [A, ๐ฎ] (๐, ๐ ๐น ) → [B, ๐ฎ] (Lan๐น ๐, ๐ ), ๐ โฆ→ e ๐. It remains to prove that this bijection, call it ๐๐ ,๐ , is natural in ๐ and ๐ . So let ๐ : ๐ 0 ⇒ ๐ and ๐ : ๐ ⇒ ๐ 0 be natural transformations. We shall prove that the diagram [A, ๐ฎ] (๐, ๐ ๐น ) ๐๐ ,๐ (๐ ๐น )∗ โฆ๐ ∗ [A, ๐ฎ] (๐ 0, ๐ 0๐น ) [B, ๐ฎ] (Lan๐น ๐, ๐ ) ๐ ∗ โฆLan๐น (๐) ∗ ๐๐ 0,๐ 0 [B, ๐ฎ] (Lan๐น ๐ 0, ๐ 0) is commutative. Let ๐ : ๐ ⇒ ๐ ๐น and let ๐ต ∈ B. Let (๐ด, โ) ∈ (๐น ⇒ ๐ต) and let ๐ (๐ด,โ) : ๐ 0 (๐ด) → Lan๐น ๐ 0 (๐ต) denote the inclusion. It follows from the definitions that (๐ โฆ ๐๐ ,๐ (๐) โฆ Lan๐น (๐))๐ต โฆ ๐ (๐ด,โ) = ๐ ๐ต โฆ ๐ (โ) โฆ ๐๐ด โฆ ๐๐ด and ๐๐ 0,๐ 0 (๐ ๐น โฆ ๐ โฆ ๐)๐ต โฆ ๐ (๐ด,โ) = ๐ 0 (โ) โฆ ๐ ๐น (๐ด) โฆ ๐๐ด โฆ ๐๐ด, and these two are equal by naturality of ๐ . It follows that the diagram above commutes, so that ๐๐ ,๐ is natural in ๐ and ๐ . We conclude that the functor − โฆ ๐น : [B, ๐ฎ] → [A, ๐ฎ] is right adjoint to the functor Lan๐น : [A, ๐ฎ] → [B, ๐ฎ]. (c) Let ๐ฎ = Set be the category of sets. First let ๐น be the unique functor 1 → ๐บ . Then − โฆ ๐น : [๐บ, Set] → [1, Set] = Set is precisely the forgetful functor ๐ from ๐บ-sets to sets. From part (b) we know ๐ has both left and right adjoints, and in Exercise 2.1.16 we described these adjoints explicitly. Its left adjoint ๐บ : Set → [๐บ, Set] is given on objects by sending ๐ ∈ Set to ๐บ × ๐ ∈ [๐บ, Set], where the ๐บ-action on ๐บ × ๐ is given by โ · (๐, ๐ฅ) = (โ๐, ๐ฅ) for all ๐, โ ∈ ๐บ and ๐ฅ ∈ ๐, and a map ๐ : ๐ → ๐ of sets to the map ๐บ (๐ ) : ๐บ × ๐ → ๐บ × ๐ of ๐บ-sets given by (๐, ๐ฅ) โฆ→ (๐, ๐ (๐ฅ)). Its right adjoint ๐ป : Set → [๐บ, Set] sends ๐ ∈ Set to ๐ป (๐ ) = ๐ ๐บ ∈ [๐บ, Set], the set of functions ๐บ → ๐, with ๐บ-action given by (๐ · ๐ ) (โ) = ๐ (โ๐) for all ๐, โ ∈ ๐บ, ๐ ∈ ๐ ๐บ , and a function ๐ : ๐ → ๐ to the map ๐ป (๐) : ๐ ๐บ → ๐ ๐บ given by [๐ป (๐) (๐ )] (โ) = ๐ โฆ ๐ (โ). Now consider ๐น to be the unique functor ๐บ → 1. Then − โฆ ๐น : Set → [๐บ, Set] is precisely the functor Δ that equips a set with the trivial ๐บ-action. From part (b) we know Δ has both left and right adjoints, and in Exercise 2.1.16 we described these adjoints explicitly. Its left adjoint is the functor (−)/๐บ : [๐บ, Set] → Set sending a ๐บ-set ๐ to the set ๐ /๐บ = {๐บ · ๐ฅ | ๐ฅ ∈ ๐ } of orbits of ๐ under ๐บ, and its right adjoint the functor (−)๐บ : [๐บ, Set] → Set sending a ๐บ-set ๐ to its ๐บ-fixed point subset ๐ ๐บ = {๐ฅ ∈ ๐ | ๐ · ๐ฅ = ๐ฅ for all ๐ ∈ ๐บ }. 59 Solutions by positrón0802 6.3 6.3 Interactions between adjoint functors and limits Interactions between adjoint functors and limits Exercise 6.3.21. (a) The initial object of Grp is the zero group, whose underlying set has precisely one element, and the initial object of Set is the empty set. Thus ๐ does not preserve colimits, so it has no right adjoint by Theorem 6.3.1. (b) Recall from Exercise 3.2.16 that ๐ถ : Cat → Set sends a small category ๐ be the quotient ๐ถ (๐) of the set of objects of ๐ by the equivalence relation generated by ๐ ∼ ๐ 0 if there exists an arrow ๐ → ๐ 0 in ๐. We shall prove that ๐ถ has no left adjoint. For this purpose, first we claim that right adjoints preserve monics. Indeed, suppose ๐บ : ๐ → โฐ is a functor right adjoint to a functor ๐น, with the adjunction given by the natural isomorphism ๐ ๐ธ,๐ท : ๐(๐น (๐ธ), ๐ท) → โฐ(๐ธ, ๐บ (๐ท)). Let ๐ : ๐ท → ๐ท 0 be monic in ๐ and โ 1, โ 2 : ๐ธ → ๐บ (๐ท) two maps in ๐ such that ๐บ (๐ ) โฆ โ 1 = ๐บ (๐ ) โฆ โ 2 . It follows from equation (2.2) that −1 ๐ โฆ ๐ ๐ธ,๐ท (โ 1 ) = ๐ ๐ธ,๐ท 0 (๐บ (๐ ) โฆ โ 1 ) = ๐ ๐ธ,๐ท 0 (๐บ (๐ ) โฆ โ 2 ) = ๐ โฆ ๐ −1 (โ 2 ), −1 (โ ) = ๐ −1 (โ ) as ๐ is monic, and therefore โ = โ . The claim follows. Thus, for so that ๐ ๐ธ,๐ท 1 2 1 2 proving that ๐ถ is not a right adjoint, we will prove that ๐ถ does not preserve monics. Let ๐ be a discrete category with two objects ๐ 1, ๐ 2 and ๐ be the indiscrete category with two objects ๐ 1, ๐ 2, that is, we have precisely two non-identity morphisms in ๐, one ๐ 1 → ๐ 2 and one ๐ 2 → ๐ 1 . We have a functor ๐น : ๐ → ๐ sending ๐๐ to ๐๐ , ๐ = 1, 2. Assume โฐ is a category and ๐บ, ๐บ 0 : โฐ → ๐ are functors such that ๐น๐บ = ๐น๐บ 0 . Then ๐บ and ๐บ 0 must the the same on objects, and must send a morphism ๐ → ๐ 0 in โฐ to the identity map of ๐บ (๐) = ๐บ (๐ 0) = ๐บ 0 (๐) = ๐บ 0 (๐ 0). It follows that ๐น is monic. But ๐ถ (๐น ) is not monic, for ๐ถ (๐) has two objects while ๐ถ (๐) has only one. We deduce that ๐ถ is not a right adjoint. Now, recall from Exercise 3.2.16 that ๐ผ : Set → Cat sends a set ๐ to the indiscrete category ๐ผ (๐) whose set of objects is ๐. We shall prove that ๐ผ is not a left adjoint. By Theorem 6.3.1, it suffices to note that ๐ผ does not preserve coproducts. If {๐} denotes a one-element set, note that {๐} q {๐} is simply a set with precisely two elements ๐ 1, ๐ 2, so that the category ๐ผ ({๐} q {๐}) has a morphism ๐ 1 → ๐ 2 . But, on the other hand, the category ๐ผ ({๐}) q ๐ผ ({๐}) has no non-identity morphisms. It follows that ๐ผ has no right adjoint. (c) Let ๐ be a topological space. If ๐ is empty, then in the chain of adjunctions in Exercise 2.1.17 we have Λ = Δ = ∇ and Π = Γ, so we have two functors which are left and right adjoint to each other and thus the chain extends infinitely at both ends. Now assume that ๐ is non-empty. Recall from Exercise 2.1.17 that Λ : Set → [๐ช(๐ ) op, Set] sends ๐ด ∈ Set to the functor Λ(๐ด) : ๐ช(๐ ) op → Set given on objects by Λ(๐ด) (∅) = ๐ด and Λ(๐ด)(๐ ) = ∅ if ๐ ≠ ∅, and on morphisms by Λ(๐ด) (1 ∅ ) = 1๐ด and Λ(๐ด) (๐๐ ,๐ ) = the empty function if ๐ ≠ ∅. We claim that Λ has no left adjoint. By Theorem 6.3.1, it suffices to prove that Λ does not preserve limits. For this purpose, consider the one-point set {∗}, which is terminal in Set. Since limits in [๐ช(๐ ) op, Set] are computed point-wise, if ๐น is a terminal object in [๐ช(๐ ) op, Set] then ๐น (๐ ) must be a singleton for all ๐ ∈ ๐ช(๐ ). As Λ({∗}) ({∗}) = ∅, it follows that Λ({∗}) is not terminal. We deduce that Λ has no left adjoint. 60 Solutions by positrón0802 6.3 Interactions between adjoint functors and limits Now, recall from Exercise 2.1.17 that ∇ : Set → [๐ช(๐ ) op, Set] is given on objects by sending ๐ด ∈ Set to the functor ∇๐ด : ๐ช(๐ ) op → Set given by ∇๐ด(๐ ) = ๐ด and ∇๐ด(๐ ) = {∗} for all ๐ ≠ ๐ . Now ∅ is the initial object of Set, but (as ๐ is non-empty) the set ∇∅(∅) = {∗} is non-empty. It follows from Theorem 6.3.1 that ∇ has no right adjoint. Exercise 6.3.22. (a) First assume that ๐ has a left adjoint ๐น : Set → ๐ and let ๐๐,๐ด : ๐(๐น (๐), ๐ด) → Set(๐, ๐ (๐ด)) be the isomorphism natural in ๐ ∈ Set and ๐ด ∈ ๐ determining the adjunction. In particular, by considering the one-element set ๐ = {∗}, we have an isomorphism ๐ {∗},๐ด : ๐(๐น ({∗}), ๐ด) → Set({∗}, ๐ (๐ด)) = ๐ (๐ด) natural in ๐ด ∈ ๐. It follows that ๐ is representable. Now, if ๐ is representable, then it preserves limits by Proposition 6.2.2. (b) Suppose that ๐ has sums and is representable. Without loss of generality, we may assume that ๐ = ๐ป ๐ด = ๐(๐ด, −) for some ๐ด ∈ ๐. Define a functor ๐น : Set → ๐ as follows. For ๐ ∈ Set, let ๐น (๐) = q๐ ∈๐ ๐ด (a chosen coproduct). For a function ๐ : ๐ 0 → ๐, we have a unique map ๐น (๐ ) : ๐น (๐ 0) → ๐น (๐) induced by the universal property of the coproduct. That is, for ๐ 0 ∈ ๐ 0, denote by ๐๐ 0 : ๐ด → q๐ 0 ∈๐ 0 ๐ด inclusion corresponding to the summand ๐ 0 ∈ ๐ 0, and for ๐ ∈ ๐ denote by ๐๐ : ๐ด → q๐ ∈๐ ๐ด the inclusion corresponding to the summand ๐ ∈ ๐. Then ๐น (๐ ) is the unique map such that ๐น (๐ ) โฆ ๐๐ 0 = ๐ ๐ (๐ 0) . We claim that ๐น is left adjoint to ๐ . For ๐ ∈ ๐ฎ and ๐ต ∈ ๐, define ๐๐,๐ต : ๐(๐น (๐), ๐ต) → Set(๐, ๐ (๐ต)) as follows. Let โ : ๐น (๐) → ๐ต. Let ๐๐,๐ต (โ) : ๐ → ๐ (๐ต) send ๐ to โ โฆ ๐๐ : ๐ด → ๐ต. Conversely, define ๐๐,๐ต : Set(๐, ๐ ๐ต) → ๐(๐น (๐), ๐ต) as follows. For ๐ : ๐ → ๐ (๐ต), let ๐๐,๐ต : ๐น (๐) → ๐ต be induced from the maps ๐ (๐ ) : ๐ด → ๐ต, ๐ ∈ ๐, by the universal property of the coproduct. It is clear that ๐๐,๐ต and ๐๐,๐ต are inverses to each other, so that ๐๐,๐ต is a bijection. It remains to prove that ๐๐,๐ต is natural in ๐ ∈ Set and ๐ต ∈ ๐. Let ๐ : ๐ 0 → ๐ be a function of sets and ๐ : ๐ต → ๐ต 0 be a map in ๐. We shall prove that the diagram ๐(๐น (๐), ๐ต) ๐๐,๐ต ๐∗ โฆ๐น (๐ ) ∗ ๐(๐น (๐ 0), ๐ต 0) Set(๐, ๐ (๐ต)) ๐∗ โฆ๐ (๐) ∗ ๐๐ 0,๐ต0 Set(๐ 0, ๐ (๐ต 0)) commutes. Let โ : ๐น (๐) → ๐ต. For ๐ ∈ ๐ and ๐ 0 ∈ ๐ 0, let ๐๐ and ๐๐ 0 be defined as in the first paragraph above. Then ๐ โฆ ๐๐,๐ต โฆ ๐ (๐) sends ๐ 0 ∈ ๐ 0 to ๐ โฆ โ โฆ ๐ ๐ (๐ 0) , and ๐๐ 0,๐ต0 (๐ โฆ โ โฆ ๐น (๐ )) sends ๐ 0 ∈ ๐ 0 to ๐ โฆ โ โฆ ๐น (๐ ) โฆ ๐๐ 0 , and these two are equal since ๐น (๐ ) โฆ ๐๐ 0 = ๐ ๐ (๐ 0) . It follows that the diagram above commutes, so that ๐๐,๐ต is natural. We conclude that ๐น is left adjoint to ๐ . Exercise 6.3.23. (a) Let ๐ be a preordered set. Let ๐ be the quotient of ๐ by the equivalence relation given by ๐ ∼ ๐ 0 if ๐ ≤ ๐ 0 and ๐ 0 ≤ ๐. For ๐ ∈ ๐, let [๐] ∈ ๐ denote its equivalence class. 61 Solutions by positrón0802 6.3 Interactions between adjoint functors and limits Then ๐ is an ordered set via [๐] ≤ [๐ 0] if ๐ ≤ ๐ 0 . Let ๐น : ๐ → ๐ be the quotient function. Then ๐น is a functor. Moreover, ๐น is clearly full, faithful and essentially surjective on objects, hence an equivalence by Proposition 1.3.18. (b) A map ๐ : ๐ด → ๐ต ๐ผ from ๐ด into the product of |๐ผ | copies of ๐ต is the same thing as a collection of maps ๐๐ : ๐ด → ๐ต, ๐ ∈ ๐ผ . Since there exist distinct maps ๐ , ๐ : ๐ด → ๐ต in ๐, there are at least two possibilities for each ๐๐ . Thus, we have at least 2 |๐ผ | maps ๐ด → ๐ต ๐ผ in ๐. In particular, it follows that ๐ is not small. (c) If ๐ is small with small products, it follows from part (b) that ๐ is a preoder, and in turn from part (a) that ๐ is equivalent to an ordered set ๐. Note that ๐ has all small products, being equivalent to ๐. Moreover, trivially any ordered set has equalisers. It follows from Proposition 5.1.26(a) that ๐ is complete. (d) Let ๐ be a finite category with finite products. From the analogous statement of (b) it follows that ๐ is a preorder. Thus, by part (a) we have that ๐ is equivalent to a finite ordered set ๐ with all finite products. Note that any finite ordered set has a maximal element, i.e. ๐ has a terminal object. We deduce from Proposition 5.1.26(b) that ๐ is complete. Exercise 6.3.24. (a) An element of the subgroup of ๐บ generated by {๐๐ | ๐ ∈ ๐ด} is a finite string of elements of the set {๐๐ , ๐๐−1 | ๐ ∈ ๐ด}. it follows that this subgroup has cardinality at most |N| · |๐ด| = max{|N|, |๐ด|}. (b) Let ๐ be a set. Let ๐บ be a group with cardinality |๐บ | ≤ |๐ |. Then there is an injection ๐ : ๐บ → ๐ which endows ๐ (๐บ) ⊂ ๐ with a group structure such that ๐บ ๐ (๐บ). Thus, we may assume that ๐บ ⊂ ๐. Now the group structure of ๐บ is determined by a function ๐บ × ๐บ → ๐บ, which is a subset of ๐บ × ๐บ × ๐บ ⊂ ๐ × ๐ × ๐. It follows that there is at most |๐ซ(๐)| · |๐ซ(๐ × ๐ × ๐)| isomorphism classes of groups of cardinality at most |๐ |, so this collection is small. (c) If ๐ด is infinite, let ๐ = ๐ด, otherwise, let ๐ = N. Consider the collection S of all elements of ∈ (๐ด ⇒ ๐ ) such that ๐ (๐บ 0) ⊂ ๐. It follows from part (b) that S is a set. Let (๐บ, โ) ∈ (๐ด ⇒ ๐ ) and let ๐ฟ denote the subgroup of ๐บ generated by {โ(๐) | ๐ ∈ ๐ด}. By part (a), ๐ฟ has cardinality at most |๐ |, so we have a bijection ๐ : ๐ 0 → ๐ (๐ฟ) from ๐ 0 a subset of ๐. This bijection endows the set ๐ 0 with a unique group structure such that ๐ is an isomorphism; call this group ๐บ 0, so that ๐ (๐บ 0) = ๐ 0 . If ๐ : ๐ฟ → ๐บ denotes the inclusion, then ๐ = ๐ โฆ ๐ : ๐บ 0 → ๐บ is a monomorphism of groups. Now let โ 0 : ๐ด → ๐ 0 be given by โ 0 (๐) = ๐ −1 โฆ โ(๐). Then ๐ : (๐บ 0, โ 0) → (๐บ, โ) is a map in (๐ด ⇒ ๐ ), where (๐บ 0, โ 0) ∈ S. It follows that S is a weakly initial set in (๐ด ⇒ ๐ ). (๐บ 0, โ 0) (d) Note that Set is complete and Grp is locally small. In Exercise 5.3.11 we proved that ๐ creates arbitrary limits. Thus, by Lemma 5.3.6 (proved in Exercise 5.3.12), Grp is complete and ๐ preserves limits. Finally, in (c) we have prove that (๐ด ⇒ ๐ ) has a weakly initial set for every ๐ด ∈ Set. We deduce from the General adjoint functor theorem (Theorem 6.3.10) that ๐ has a left adjoint. b = [Aop, Set] for the presheaf category. Let ๐ป • : A → A b denote the Exercise 6.3.25. Write A Yoneda embedding. By assumption, A has finite products and for each ๐ต ∈ A, the functor − × 62 Solutions by positrón0802 6.3 Interactions between adjoint functors and limits ๐ต : A → A has a right adjoint (−) ๐ต : ๐ → ๐. Given ๐, ๐ ∈ ๐, there is a natural isomorphism ๐ป๐ ×๐ = ๐(−, ๐ × ๐ ) ๐(−, ๐ ) × ๐(−, ๐ ) = ๐ป๐ × ๐ป๐ given at ๐ด ∈ A by sending a map ๐ : ๐ด → ๐ × ๐ into the pair (๐๐ โฆ ๐ , ๐๐ โฆ ๐ ), where ๐๐ and ๐๐ denote the projections. Thus ๐ป • preserves finite products; it remains to prove that it preserves b For ๐ด, ๐ต ∈ A, exponentials. (Recall from the proof of Theorem 6.3.20 the exponentials in A.) b • × ๐ป๐ต , ๐ป๐ด ) A(๐ป b •×๐ต , ๐ป๐ด ). (๐ป๐ด ) ๐ป๐ต = A(๐ป Since ๐ป • is fully faithful, we have b •×๐ต , ๐ป๐ด ) (๐ป๐ด ) ๐ป๐ต . ๐ป๐ด๐ต = A(−, ๐ด๐ต ) A(− × ๐ต, ๐ด) A(๐ป It follows that ๐ป • preserves the whole cartesian closed structure. Exercise 6.3.26. (a) Let ๐ : ๐ด 0 → ๐ด be a map in ๐. Let ๐ : ๐ → ๐ด be a monic into ๐ด, representing a class [๐] ∈ Sub(๐ด). By assumption, the pullback ๐0 ๐0 ๐ ๐0 ๐ ๐ด0 ๐ด ๐ exists in ๐. It follows from Exercise 5.1.42 that ๐ 0 : ๐ 0 → ๐ด 0 is a monic into ๐ด 0, and we set Sub(๐ ) [๐] = [๐ 0]. It is clear that this gives a well-defined map Sub(๐ ) : Sub(๐ด) → Sub(๐ด 0). (b) We follow the hint. It is clear that Sub(1๐ด ) = 1Sub(๐ด) for any ๐ด ∈ ๐. Now let ๐ : ๐ด 00 → ๐ด 0 and ๐ : ๐ด 0 → ๐ด and be composable maps in ๐. Let ๐ : ๐ → ๐ด be a monic into ๐ด. Consider a diagram ๐ 00 ๐0 ๐00 ๐0 ๐0 ๐ ๐0 ๐ด 00 ๐ด0 ๐ ๐ ๐ ๐ด where both squares are pullbacks. Then Sub(๐) โฆ Sub(๐ ) [๐] = Sub(๐) [๐ 0] = [๐ 00] by definition. On the other hand, the outer rectangle is also a pullback by Exercise 5.1.35, so that Sub(๐ โฆ ๐ ) [๐ 00] = [๐]. We deduce that Sub is a functor ๐ op → Set. (c) Consider the set 2 = {0, 1} with two elements. Given ๐ ∈ Set, recall from Exercise 5.1.40(a) that subobjects of ๐ are in canonical one-to-one correspondence with subsets of ๐, where the class of a monic ๐ : ๐ → ๐ into ๐ corresponds to the image ๐(๐ ) ⊂ ๐. Under this correspondence, for a function ๐ : ๐ 0 → ๐, the map Sub(๐ ) : Sub(๐) → Sub(๐ 0) is simply the inverse image functor ๐ซ(๐) → ๐ซ(๐ 0), ๐ โฆ→ ๐ −1 (๐ ). It follows that Sub ๐ป {0,1} . 63 Solutions by positrón0802 6.3 Interactions between adjoint functors and limits b = [Aop, Set] for the presheaf category. Exercise 6.3.27. Write A b Then Sub(๐ ) ๐ป Ω (๐ ) = (a) Suppose that Ω : Aop → Set is a subobject classifier of A. b b In particular, this holds for representable presheaves, so for all ๐ด ∈ A we A(๐, Ω) for all ๐ ∈ A. b have Sub(๐ป๐ด ) A(๐ป๐ด, Ω) Ω(๐ด), the last isomorphism by Yoneda. b op denote the Yoneda emdedding. Inspired from (a), set (b) Let ๐ป • : Aop → ( A) Ω = Sub โฆ๐ป • : Aop → Set. We shall prove that Ω is a subobject classifier. We shall prove that Ω is a subobject classifier. First b are precisely the natural transformations which recall from Exercise 6.2.20(b) that monics in A are pointwise monic. It follows from the description of subobjects in Set (see Exercise 5.1.40(a)) b Sub(๐ ) can be identified with collection of subpresheaves of ๐, that is, that for a presheaf ๐ ∈ A, b presheaves ๐น ∈ A such that ๐น (๐ด) ⊂ ๐ (๐ด) for all ๐ด ∈ A and such that for all maps ๐ : ๐ด 0 → ๐ด in A, the function ๐ (๐ ) : ๐ (๐ด) → ๐ (๐ด 0) restricts to a function ๐น (๐ด) → ๐น (๐ด 0). b and define a map ๐๐ : Sub(๐ ) → Now, we want a natural isomorphism Sub ๐ป Ω . Let ๐ ∈ A b A(๐, Ω) as follows. Let ๐น ∈ Sub(๐ ) be a subpresheaf of ๐ . Let ๐ด ∈ A. Given ๐ ∈ ๐ (๐ด), let ๐๐ : ๐ป๐ด ⇒ ๐ be the natural transformation corresponding to ๐ via the Yoneda Lemma, and define ๐๐ (๐น )๐ด : ๐ (๐ด) → Sub(๐ป๐ด ) to send ๐ ∈ ๐ (๐ด) to the subpresheaf ๐น 0 of ๐ป๐ด given by the pullback ๐น0 ๐น ๐๐ ๐ป๐ด ๐, where the right-hand side map is the inclusion ๐น ⇒ ๐ . Explicitly, ๐น 0 : Aop → Set is given by ๐น 0 (๐ต) = {๐ ∈ A(๐ต, ๐ด) | (๐๐ )๐ต (๐ ) = ๐ (๐ ) (๐) ∈ ๐น (๐ต)} ⊂ A(๐ต, ๐ด) on objects and it is the restriction of ๐ป๐ด on morphisms. We claim that ๐๐ (๐น )๐ด is natural in ๐ด ∈ A. Let ๐ : ๐ด 0 → ๐ด be a map in A. We shall prove that the diagram ๐ (๐ด) ๐๐ (๐น )๐ด Sub(๐ป ๐ ) ๐ (๐ ) ๐ (๐ด 0) Sub(๐ป๐ด ) ๐๐ (๐น )๐ด0 Sub(๐ป๐ด0 ) commutes. Let ๐ ∈ ๐ (๐ด). By definition Sub(๐ป ๐ ) โฆ ๐๐ (๐น )๐ด (๐) is the subpresheaf ๐น 0 ⊂ ๐ป๐ด0 given by ๐น 0 (๐ต) = {๐ ∈ A(๐ต, ๐ด 0) | ๐ โฆ ๐ ∈ ๐๐ (๐น )๐ด (๐) (๐ต)} = {๐ ∈ A(๐ต, 0 ๐ด) | ๐ (๐ โฆ ๐) (๐) ∈ ๐น (๐ต)}. On the other hand, by definition ๐๐ (๐น )๐ด0 โฆ ๐ (๐ ) (๐) is the subpresheaf ๐น 00 ⊂ ๐ป๐ด0 given by ๐น 00 (๐ต) = {๐ ∈ A(๐ต, ๐ด) | ๐ (๐) โฆ ๐ (๐ ) (๐) ∈ ๐น (๐ต)}. 64 Solutions by positrón0802 6.3 Interactions between adjoint functors and limits Thus ๐น 0 = ๐น 00 and it follows that ๐๐ (๐น ) : ๐ ⇒ Sub โฆ๐ป • is natural. b Conversely, define ๐๐ : A(๐, Ω) → Sub(๐ ) as follows. Let ๐ : ๐ ⇒ Ω be natural. Let ๐๐ (๐) be the subpresheaf of ๐ given by ๐๐ (๐) (๐ด) = {๐ ∈ ๐ (๐ด) | 1๐ ∈ ๐๐ด (๐) (๐ด) ⊂ A(๐ด, ๐ด)} ⊂ ๐ (๐ด) on objects. We shall prove that ๐๐ is a two-sided inverse of ๐๐ . Let ๐น ∈ Sub(๐ ). By definition, ๐๐ โฆ ๐๐ (๐น ) is the subpresheaf ๐น 0 of ๐ given by ๐น 0 (๐ด) = {๐ ∈ ๐ (๐ด) | 1๐ ∈ ๐๐ (๐น )๐ด (๐) (๐ด)}. Since ๐๐ (๐น )๐ด (๐) (๐ด) ⊂ A(๐ด, ๐ด) consists of those arrows ๐ : ๐ด → ๐ด such that ๐ (๐ ) (๐) ∈ ๐น (๐ด), it follows that ๐น 0 (๐ด) = ๐น (๐ด) for all ๐ด and therefore ๐๐ โฆ ๐๐ (๐น ) = ๐น . On the other hand, let ๐ : ๐ ⇒ Ω. By definition, ๐๐ (๐๐ (๐))๐ด (๐) (๐ต) = {๐ ∈ A(๐ต, ๐ด) | ๐ (๐ ) (๐) ∈ ๐๐ (๐) (๐ต)}. Unravelling the definitions, we thus see that for proving ๐๐ โฆ ๐๐ (๐) = ๐ it suffices to prove the following: for any ๐ : ๐ต → ๐ด in A and ๐ ∈ ๐ด, we have ๐ ∈ ๐๐ด (๐) (๐ต) if and only if 1๐ต ∈ ๐๐ต (๐ (๐ ) (๐)) (๐ต). This follows from naturality of ๐, by considering the commutative diagram ๐ (๐ด) Sub(๐ป๐ด ) ๐๐ด Sub(๐ป ๐ ) ๐ (๐ ) ๐ (๐ต) Sub(๐ป๐ต ) ๐๐ต evaluated at ๐ ∈ ๐ (๐ด). We deduce that ๐๐ is a two-sided inverse of ๐๐ . Lastly, we shall prove that ๐๐ is natural in ๐ . So let ๐ผ : ๐ 0 ⇒ ๐ be a natural transformations between presheaves on A and consider the diagram Sub(๐ ) ๐๐ Sub(๐ผ) Sub(๐ 0) b A(๐, Ω) b A(๐ผ,Ω) ๐๐ 0 b 0, Ω) . A(๐ Let ๐น be a subpresheaf of ๐ . Let ๐ด, ๐ต ∈ A and ๐ 0 ∈ ๐ 0 (๐ด). On the one hand, we have ๐๐ (๐น )๐ด (๐ผ๐ด (๐ 0)) (๐ต) = {๐ ∈ A(๐ต, ๐ด) | ๐ (๐ ) (๐ผ๐ด (๐ 0)) ∈ ๐น (๐ต)}. On the other hand, ๐๐ 0 (Sub(๐ผ) (๐น ))๐ด (๐ 0) (๐ต) = {๐ ∈ A(๐ต, ๐ด) | ๐ 0 (๐ ) (๐ 0) ∈ Sub(๐ผ) (๐น ) (๐ต)} = {๐ ∈ A(๐ต, ๐ด) | ๐ผ ๐ต โฆ ๐ 0 (๐ ) (๐ 0) ∈ ๐น (๐ต)}. It follows from naturality of ๐ผ that the above two sets are equal. We conclude that ๐ : Sub ⇒ ๐ป Ω b is a natural isomorphism, so that Ω is a subobject classifier for A. b is cartesian closed and has all (c) From Theorems 6.2.5 and 6.3.20 we now that the category A b is a topos. limits. By part (b), it has a subobject classifies. Thus A 65 Solutions by positrón0802 6.3 Interactions between adjoint functors and limits Appendix: Proof of the general adjoint functor theorem Exercise A.3. (a) For each ๐ต ∈ โฌ, let 0๐ต : 0 → ๐ต denote the unique map from 0 to ๐ต. Given a map 0๐ต ๐ : ๐ต → ๐ต 0, we have ๐ โฆ 0๐ต = 0๐ต0 : 0 → 0๐ต0 by uniqueness, so (0 −−→ ๐ต)๐ต ∈โฌ is a indeed a cone in โ๐ต 1โฌ : โฌ → โฌ. Now let ๐ถ ∈ โฌ and let (๐ถ −−→ ๐ต)๐ต ∈โฌ be a cone on 1โฌ . In particular, โ 0 : ๐ถ → 0 is a map such that 0๐ต โฆ โ 0 = โ๐ต for all ๐ต ∈ โฌ. Moreover, โ 0 is the unique such map, for if โ : ๐ถ → 0 satisfies 0๐ต โฆ โ = โ๐ต for all ๐ต ∈ โฌ, in particular for ๐ต = 0 we have 00 โฆ โ = โ 0 ; since 00 = 10, we 0๐ต have โ = โ 0 . It follows that (0 −−→ ๐ต)๐ต ∈โฌ is a limit cone on 1โฌ . ๐๐ต (b) Since (๐ฟ −−→ ๐ต)๐ต ∈โฌ is a cone on 1โฌ, for all ๐ต ∈ โฌ we have ๐ ๐ต โฆ ๐ ๐ฟ = ๐ ๐ต . Since we also have ๐ ๐ต โฆ 1๐ฟ = ๐ ๐ต for all ๐ต, it follows by the uniqueness on the universal property of the limit that ๐ ๐ฟ = 1๐ฟ . Now let ๐ต ∈ โฌ. If ๐ : ๐ฟ → ๐ต is a map in โฌ, we must have ๐ ๐ต = ๐ โฆ ๐ ๐ฟ = ๐ โฆ 1๐ฟ , so ๐ ๐ต is the unique map ๐ฟ → ๐ต. We deduce that ๐ฟ is initial. Exercise A.4. (a) By definition ๐ ⊂ ๐ถ is a weakly initial set if for all ๐ ∈ ๐ถ, there exists ๐ ∈ ๐ such that ๐ ≤ ๐. Ó (b) Let ๐ 0 = ๐ ∈๐ ๐ . By definition of meet (greatest lower bound), we have ๐ 0 ≤ ๐ for all ๐ ∈ ๐. Given ๐ ∈ ๐ถ, there exists ๐ ∈ ๐ such that ๐ ≤ ๐, and thus ๐ 0 ≤ ๐. It follows that ๐ 0 is the least element of ๐ถ. Exercise A.5. (a) (We do not need ๐บ to be limit-preserving in this part.) Let ๐ธ : I → (๐ด ⇒ ๐บ) be a diagram. By definition, for each ๐ ∈ I we are given an object (๐ธ๐ , โ๐ ) ∈ (๐ด ⇒ ๐บ), that is, an object ๐ธ๐ ∈ โฌ together with a morphism โ๐ : ๐ด → ๐บ (๐ธ๐ ). Furthermore, for each map ๐ : ๐ → ๐ in I we have a map (๐ธ๐ , โ๐ ) → (๐ธ ๐ , โ ๐ ), that is, a map ๐ธ (๐ ) : ๐ธ๐ → ๐ธ ๐ in โฌ such that ๐ธ (๐ ) โฆ โ๐ = โ ๐ : ๐ด → ๐บ (๐ธ ๐ ). This is precisely the information of a diagram ๐ธ : I → โฌ in โฌ together with a cone on ๐บ โฆ ๐บ with vertex ๐ด. ๐๐ (b) Let ๐ท : I → (๐ด ⇒ ๐บ) be a diagram. For each ๐ ∈ I, let ๐ท (๐) = (๐ต๐ , โ๐ ). Let (๐ต −→ ๐๐ด ๐ท (๐) = ๐บ (๐๐ ) ๐ต๐ )๐ ∈I be a limit cone on ๐๐ด ๐ท. Since ๐บ is limit-preserving, (๐บ (๐ต) −−−−→ ๐บ (๐ต๐ ))๐ ∈I is a limit cone โ๐ on ๐บ๐๐ด ๐ท. Now, from part (a), (๐ด −→ ๐บ (๐ต๐ ))๐ ∈I is a cone on ๐บ๐๐ด ๐ท, so there is a unique map ๐ : ๐ด → ๐บ (๐ต) such that ๐บ (๐๐ ) โฆ ๐ = โ๐ for all ๐ ∈ I. Now (๐ต, ๐ ) ∈ (๐ด ⇒ ๐บ), and for each ๐ ∈ ๐ผ, ๐๐ is ๐๐ a map (๐ต, ๐ ) → (๐ต๐ , โ๐ ) in (๐ด ⇒ ๐บ). So ((๐ต, ๐ ) −→ ๐ท (๐))๐ ∈I is a cone on ๐ท such that ๐๐ด (๐ต, ๐ ) = ๐ต ๐๐ and ๐๐ด (๐๐ ) = ๐๐ for all ๐ ∈ I. Moreover, ((๐ต, ๐ ) −→ ๐ท (๐))๐ ∈I is unique as such by uniqueness of ๐๐ ๐๐ ๐ . Finally, we shall prove that ((๐ต, ๐ ) −→ ๐ท (๐))๐ ∈I is a limit cone. So let ((๐ต 0, ๐ 0) −→ ๐ท (๐))๐ ∈I be ๐๐ a cone on ๐ท. Then (๐ต 0 −→ ๐ต๐ )๐ ∈I ) is a cone on ๐๐ด ๐ท, so there exists a unique map ๐ก : ๐ต → ๐ต 0 such ๐บ (๐๐ ) that ๐๐ โฆ ๐ก = ๐๐ for all ๐ ∈ I. It then follows from the fact that (๐บ (๐ต) −−−−→ ๐บ (๐ต๐ ))๐ ∈I is a limit cone that ๐ก is a map (๐ต 0, ๐ 0) → (๐ต, ๐ ), the unique such that ๐๐ โฆ ๐ก = ๐๐ (now in (๐ด ⇒ ๐บ)) for all ๐ ∈ I. We conclude that the projection ๐๐ด : (๐ด ⇒ ๐บ) → โฌ is limit-preserving. 66 Solutions by positrón0802 6.3 Interactions between adjoint functors and limits Please send comments, suggestions and corrections by e-mail, or at website. https://positron0802.wordpress.com positron0802@mail.com 67 Solutions by positrón0802