From Torsors to Topoi: An Introduction with a View Toward $Σ$-Protocols in Cryptography

F rom T orsors to T op oi: An In tro duction with a View T o w ard Σ -Proto cols in Cryptograph y T akao Inoué F aculty of Informatics, Y amato Univ ersity , Osaka, Japan ∗ Marc h 17, 2026 Abstract This pap er pro vides a preparatory in tro duction to shea v es and top oi, written as a conceptual con tin uation of the author’s earlier introduction to torsors and as preparatory background for the author’s arXiv pap er Gr othendie ck T op olo gies and She af-The or etic F oundations of Crypto gr aphic Se curity: A ttacker Mo dels and Σ -Pr oto c ols as the First Step [ 2 ]. Rather than attempting an encyclop edic surv ey of all of top os theory , the exp osition dev elops those parts of the sub ject that are most relev ant for passing from torsor-based local-to-global reasoning to sheaf-theoretic and topos-theoretic reasoning: Grothendiec k top ologies, shea ves, torsors ov er a site, descent, sheaf top oi, elemen tary top oi, Cartesian closed structure, sub ob ject classifiers, and internal logic. The goal is not merely motiv ational. W e try to dev elop enough gen uine top os theory that the reader can understand, not only heuristically but structurally , wh y the later cryptographic framew ork of [ 2 ] uses Grothendieck top ologies and sheaf-theoretic language. T o make the note more self-con tained, we also include substantial appendices on basic category theory , Y oneda’s lemma, limits and colimits, equalizers and coequalizers, Kan extensions, the relation b etw een in ternal logic and intuitionistic logic, and exercises with solutions. In the final part, we explain ho w these ideas prepare the ground for a conceptual understanding of Σ -proto cols, esp ecially in connection with lo cal consistency , simulabilit y , and the passage from compatible lo cal data to global structure. k eywords : top os theory , sheaf theory , Grothendieck top ology , torsors, descent, internal logic, in tuitionistic logic, category theory , Y oneda lemma, Σ -proto cols, cryptographic security , smulabilit y , lo cal-to-global principle MSC2020 : 18F10, 18F20, 18N10, 03G30, 94A60 Con ten ts 1 In tro duction 3 ∗ Email: inoue.takao@y amato-u.ac.jp; P ersonal Email: takaoapple@gmail.com [I prefer m y p ersonal email address for correspondence.] 1 2 Sites, Grothendiec k top ologies, and shea ves 4 3 T orsors, co cycles, and descent on a site 5 4 F rom sheaf categories to top oi 6 5 Elemen tary top oi 6 6 Sub ob ject classifiers and in ternal logic 7 7 Practice: w orking with shea ves, top oi, and internal logic 9 8 T orsors and lo cal symmetry inside a top os 10 9 T ow ard Σ -proto cols and cryptographic securit y 11 10 Outlo ok 12 11 Conclusion 12 A A brief in tro duction to category theory 13 A.1 Categories, morphisms, and commutativ e diagrams . . . . . . . . . . . . . . . . . . . 13 A.2 F unctors and natural transformations . . . . . . . . . . . . . . . . . . . . . . . . . . 14 A.3 Univ ersal prop erties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 A.4 Limits and colimits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 A.5 Exercises on limits and colimits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 A.6 Kan extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 A.7 Represen table functors and the Y oneda p oin t of view . . . . . . . . . . . . . . . . . . 19 A.8 Cartesian closed categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 A.9 Why these notions matter for the main text . . . . . . . . . . . . . . . . . . . . . . . 21 B In ternal logic and intuitionistic logic 21 C Solutions to the exercises 22 D A p ossible lecture plan 25 D.1 Recommended ov erall length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 D.2 A standard 10-lecture plan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 D.3 A p ossible 12-lecture expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 2 D.4 P edagogical remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 1 In tro duction This pap er is intended as the middle term of a three-step line of exp osition. The first step is the author’s earlier note on torsors [ 1 ], where free transitive group actions, transp ort, lo cal triviality , and co cycle-based gluing w ere emphasized. The third step is the author’s later pap er [ 2 ], which prop oses a sheaf-theoretic approach to cryptographic securit y based on Grothendieck top ologies, attac ker mo dels, and Σ -proto cols. The presen t pap er is written to fill the conceptual gap b etw een these tw o w orks. Its purp ose is to explain, in a mathematically serious w a y , why the route from torsors to top oi is natural, and wh y a reader who wishes to understand [ 2 ] should first understand at least the basic architecture of top os theory . There are t wo reasons why a sup erficial bridge is not enough. First, torsor theory already con tains a lo cal-to-global philosophy: a torsor is lo cally indistinguishable from the acting group, yet globally it may fail to admit an y distinguished origin. This leads naturally to lo cal trivializations, transition functions, compatibility on ov erlaps, and descent. Second, the cryptographic framework of [ 2 ] do es not merely b orro w the language of shea ves as a metaphor. It treats Grothendieck top ologies, lo cal observ ability , and gluing of information as structural ingredien ts. A ccordingly , the reader must understand not only what a sheaf is, but also what a top os is, how internal logic works, and why suc h a setting supp orts disciplined lo cal reasoning. F or this reason, the presen t note is more substantial than a short conceptual bridge. It is still in tro ductory in st yle, but it aims to provide enough gen uine top os theory that the later cryptographic applications can b e understo o d from the inside. W e therefore discuss not only shea v es and torsors o ver a site, but also the passage from sheaf categories to top oi, the elementary definition of a top os, the role of exp onentials and sub ob ject classifiers, and the basic idea of internal logic. The discussion is selective rather than encyclop edic, but it is meant to b e mathematically real. A guiding principle throughout the pap er is that the mov e from torsors to top oi is not a change of sub ject. Rather, it is a conceptual enlargemen t of structures already presen t in torsor theory . The lac k of a canonical origin in a torsor, its reco v ery from lo cal trivializations, and the co cycle relations on ov erlaps all p oint tow ard a setting in which lo cal data and gluing are primary . Sheaf theory is the first systematic language for suc h phenomena. T op os theory then enlarges this language into a full mathematical univ erse with its own in ternal notion of truth, existence, and function. F rom that p ersp ectiv e, understanding top oi is part of understanding what torsors were already trying to say . The pap er is organized as follows. Section 2 reviews Grothendieck top ologies and shea v es. Section 3 revisits torsors as sheaf-theoretic and site-theoretic ob jects, with emphasis on co cycles and descent. Section 4 explains why sheaf categories lead naturally to top oi. Section 5 introduces elementary top oi. Section 6 discusses sub ob ject classifiers and in ternal logic in greater detail. Section 8 places torsors inside a top os. Section 7 pro vides a practice section before the cryptographic transition. Section 9 then explains why this entire progression matters for the conceptual study of Σ -proto cols and for [ 2 ]. Section 11 concludes the pap er by summarizing the passage from torsors to top oi and its relev ance to the later cryptographic framework. The app endices provide substantial supp orting material: a brief introduction to category theory , Y oneda’s lemma, Cartesian closed categories, limits and colimits, equalizers and co equalizers, Kan extensions, the relation b etw een internal logic and intuitionistic logic, and solutions to the exercises. 3 2 Sites, Grothendiec k top ologies, and shea v es The first serious step from torsors to top oi is the passage from ordinary op en co vers to Grothendieck top ologies. This makes it p ossible to sp eak of lo cality and gluing not only on top ological spaces, but on muc h more general categories. Definition 2.1. L et C b e a c ate gory. A sieve S on an obje ct U ∈ Ob ( C ) is a c ol le ction of morphisms with c o domain U that is close d under pr e c omp osition: if f : V → U lies in S and g : W → V is any morphism, then f ◦ g : W → U also lies in S . Definition 2.2. A Grothendiec k top ology J on a c ate gory C assigns to e ach obje ct U a c ol le ction J ( U ) of sieves on U , c al le d co v ering sieves , such that: (i) the maximal sieve b elongs to J ( U ) ; (ii) if S ∈ J ( U ) and f : V → U is any morphism, then the pul lb ack sieve f ∗ S b elongs to J ( V ) ; (iii) if S is a sieve on U and ther e exists R ∈ J ( U ) such that for every f : V → U in R , the pul lb ack sieve f ∗ S b elongs to J ( V ) , then S ∈ J ( U ) . The p air ( C , J ) is c al le d a site . These axioms abstract the familiar b eha vior of op en cov ers. Condition (ii) says that co v erings pull bac k. Condition (iii) is a transitivit y condition: a family that b ecomes cov ering after refinement b y a cov er was already cov ering. Example 2.3. L et X b e a top olo gic al sp ac e and let O ( X ) b e its c ate gory of op en sets, with morphisms given by inclusions. A family { U i ⊆ U } is c overing if U = S i U i . The sieves gener ate d by such families define a Gr othendie ck top olo gy on O ( X ) . The asso ciate d she af the ory is the usual she af the ory on X . Definition 2.4. L et ( C , J ) b e a site. A pr eshe af of sets on C is a c ontr avariant functor F : C op → Set . It is a sheaf if for every obje ct U and every c overing sieve S ∈ J ( U ) , the natur al map F ( U ) − → Hom( S, F ) is a bije ction. Equivalently, se ctions over U ar e uniquely determine d by c omp atible se ctions on every c over of U . In elemen tary terms, a sheaf is a presheaf for which lo cal compatible data glue uniquely . The mo dern form of the definition matters b ecause it sho ws that “lo calit y” dep ends only on the c hosen Grothendiec k top ology , not on any ambien t notion of p oint-set space. This flexibility is exactly what will later matter in cryptographic applications, where the “lo cal” pieces are not literally op en subsets of a top ological space, but information states organized b y a cov erage relation [ 2 ]. Prop osition 2.5. L et ( C , J ) b e a site. If F is a she af and { U i → U } is a c overing family, then any family of lo c al se ctions s i ∈ F ( U i ) that agr e e on al l overlaps U i × U U j glues to a unique glob al se ction s ∈ F ( U ) . Pr o of sketch. The family { U i → U } generates a cov ering sieve on U . Compatibilit y on the pullbac k o verlaps says precisely that the induced map from the siev e to F is natural. By the sheaf prop erty , suc h a natural transformation comes from a unique section of F ( U ) . 4 Th us sheaf theory is not simply a technical add-on. It is the formal expression of the principle that lo cal information, if compatible, should reconstruct global information. That is the precise direction in which torsor theory already p oin ts. 3 T orsors, co cycles, and descen t on a site In the earlier torsor note [ 1 ], torsors were introduced through free transitive group actions. On a site, this idea p ersists, but lo cal trivialit y now replaces the existence of an honest global basep oint. Definition 3.1. L et ( C , J ) b e a site and let G b e a she af of gr oups on it. A G -torsor is a she af P with a right action of G such that: (i) P is lo c al ly nonempty: ther e exists a c overing family { U i → U } such that e ach P ( U i ) is nonempty; (ii) the action is lo c al ly fr e e and tr ansitive: for every obje ct U and every p, q ∈ P ( U ) , ther e exists a unique g ∈ G ( U ) such that q = p · g . This definition captures the same geometry as in the classical case. A torsor lo oks like the acting group once a lo cal section has b een chosen, bu t no such c hoice is canonical globally . The resulting failure of global triviality is measured by co cycles. Prop osition 3.2. L et P b e a G -torsor and let { U i → U } b e a c overing family with chosen lo c al se ctions s i ∈ P ( U i ) . Then on e ach overlap U ij := U i × U U j ther e exists a unique element g ij ∈ G ( U ij ) such that s j = s i · g ij on U ij , and these elements satisfy the c o cycle c ondition g ij g j k = g ik on U i × U U j × U U k . Pr o of sketch. Uniqueness and existence of each g ij come from free transitivity . Ev aluating s k through s i either directly or via s j giv es the co cycle identit y on triple ov erlaps. Remark 3.3. The c o cycle of Pr op osition 3.2 do es not dep end on the chosen lo c al trivializations in an absolute sense, but only up to the exp e cte d change-of-trivialization r elations. Thus the torsor is enc o de d by desc ent data r ather than by a pr eferr e d glob al p oint. This is the first place where descen t b ecomes unav oidable. A descent datum is, roughly sp eaking, lo cal data together with compatibility on ov erlaps, sub ject to higher coherence on multiple ov erlaps. T orsors are among the cleanest elementary examples of this principle. Definition 3.4. L et { U i → U } b e a c overing family in a site. A descen t datum for pr eshe aves or structur e d obje cts over the c over c onsists of lo c al obje cts on e ach U i to gether with isomorphisms on overlaps satisfying a c o cycle c ondition on triple overlaps. Theorem 3.5 (Descen t principle for torsors, informal form) . A G -torsor over U may b e r e c onstructe d fr om lo c al trivial torsors on a c over of U to gether with a c o cycle satisfying the c omp atibility c ondition on triple overlaps. 5 Pr o of sketch. One starts with the trivial torsors G | U i and uses the co cycle to identify them on o verlaps. The sheaf condition ensures that the glued ob ject exists, and the cocycle iden tity guaran tees asso ciativity of the gluing. In this sense, torsor theory already lives naturally on a site. The later mov e to top oi do es not discard this p ersp ectiv e; it gives it a larger ambien t universe. 4 F rom sheaf categories to top oi F or a site ( C , J ) , the category Sh ( C , J ) of sheav es of sets is not only a con venien t receptacle for glued data. It is itself a highly structured mathematical world. This is the first app earance of the w ord top os . Definition 4.1. A Grothendieck top os is a c ate gory e quivalent to Sh ( C , J ) for some site ( C , J ) . The motiv ating example is Sh ( X ) for a top ological space X . But the definition is broader: a Grothendiec k top os is a category of shea ves on some generalized notion of space. It should therefore b e thought of b oth as (a) a category of lo cally v arying sets, and (b) a generalized space enco ded by the wa y those sets v ary and glue. Prop osition 4.2. F or every site ( C , J ) , the c ate gory Sh ( C , J ) has finite limits. Mor e over, the inclusion Sh ( C , J )  → Set C op pr eserves finite limits. Pr o of sketch. Finite limits of presheav es are computed p oin twise. The sheaf condition is stable under finite limits, so p oint wise finite limits of sheav es are again sheav es. This already giv es a great deal of structure. Pro ducts, equalizers, terminal ob jects, and many familiar set-like constructions exist internally . But a top os has even more: roughly sp eaking, it also has function ob jects and an internal ob ject of truth v alues. That is what mak es it a genuine replacemen t for a univ erse of sets. Remark 4.3. The p assage fr om a sp ac e X to the top os Sh ( X ) may b e r e ad in two opp osite ways. One may say that Sh ( X ) is simply the c ate gory of she aves on X . But one may also say that Sh ( X ) is the mathematic al ly me aningful avatar of X , sinc e it r ememb ers how obje cts vary lo c al ly over X and how they glue. This se c ond viewp oint is one of the c onc eptual sour c es of mo dern top os the ory. 5 Elemen tary top oi Grothendiec k top oi arise from sites, but there is also an intrinsic, axiomatic notion of top os. This is the notion of an elementary top os . It isolates the structural features that make a category resemble Set . Definition 5.1. A n elementary top os is a c ate gory E that has (i) finite limits, 6 (ii) exp onentials, and (iii) a sub obje ct classifier. Let us briefly recall these ingredients. Finite limits provide terminal ob jects, pro ducts, and equalizers. Exp onen tials mean that for ob jects A, B ∈ E , there is an ob ject B A represen ting morphisms from A to B . Th us one can sp eak of an in ternal function ob ject. The third ingredient, the sub ob ject classifier, is one of the most characteristic features of top os theory . Definition 5.2. A sub ob ject classifier in a c ate gory E with finite limits is an obje ct Ω to gether with a morphism true : 1 → Ω fr om the terminal obje ct, such that for every monomorphism m : A  → X ther e exists a unique morphism χ m : X → Ω for which the squar e A − → 1 m ↓ ↓ true X χ m − − → Ω is a pul lb ack. In Set , the sub ob ject classifier is the tw o-element set { 0 , 1 } . A subset A ⊆ X is classified by its c haracteristic function χ A : X → { 0 , 1 } . In a general top os, Ω plays the role of an ob ject of truth v alues, but its truth v alues need not b e merely Bo olean and global. This is precisely where lo cal logic enters. Example 5.3. F or a top olo gic al sp ac e X , the top os Sh ( X ) is an elementary top os. Its sub obje ct classifier Ω is the she af of op en subsets: for e ach op en set U ⊆ X , Ω( U ) is the set of op en subsets of U . Thus truth over U is r epr esente d not by a single glob al bit, but by an op en r e gion of U on which a statement holds. Prop osition 5.4. Every Gr othendie ck top os is an elementary top os. Pr o of sketch. F or Sh ( C , J ) , finite limits are inherited from the presheaf category , exp onentials exist b ecause sheafification preserves the needed structure, and a sub ob ject classifier can b e constructed b y sheafifying the presheaf of locally co vering sieves. F ull proofs ma y be found in standard references [ 3 , 4 ]. Th us the passage from sites to sheav es has brought us not merely to a category of glued ob jects, but to a set-like universe with its own in ternal structure. 6 Sub ob ject classifiers and in ternal logic One of the deep est insights of top os theory is that a top os carries its own in ternal logic. This logic is not imp osed from outside; it arises from the category’s structural features, esp ecially the sub ob ject classifier. 7 In ordinary set theory , a predicate on a set X is represented by a subset of X , or equiv alently by a characteristic function X → { 0 , 1 } . In a top os, a predicate on an ob ject X is represented by a sub ob ject of X , and hence by a classifying arro w X → Ω . Th us Ω serv es as an object of truth v alues. The difference is that these truth v alues ma y v ary lo cally . Remark 6.1. In Sh ( X ) , the truth value of a statement over an op en set U is gener al ly not simply “true” or “false” . It is an op en subset of U on which the statement holds. Henc e truth is lo c al and ge ometric. This is one of the c entr al r e asons top oi ar e so wel l adapte d to lo c al-to-glob al r e asoning. The slogan that a top os has an in ternal logic means at least three things. First, sub ob jects b ehav e lik e predicates. Second, Ω b ehav es like an internal algebra of truth v alues. Third, quantifiers can b e in terpreted by categorical adjunctions, so that statements ab out existence and universalit y can b e expressed internally . Ev en when one do es not formalize everything, these three p oints already give a serious mathematical conten t to the phrase “reasoning inside a top os. ” Example 6.2 (T ruth v alues in a sheaf top os) . L et X b e a top olo gic al sp ac e and let Ω ∈ Sh ( X ) b e the sub obje ct classifier. F or an op en set U ⊆ X , the set Ω( U ) is the set of op en subsets of U . If A  → F is a sub obje ct of a she af F , then the classifying map χ A : F → Ω assigns to a se ction s ∈ F ( U ) the lar gest op en subset of U on which s b elongs to A . Thus pr e dic ates do not r eturn a glob al bit; they r eturn a r e gion of validity. Example 6.3 (A concrete lo cal predicate) . T ake the she af C 0 ( − , R ) of c ontinuous r e al-value d functions on a sp ac e X . L et P  → C 0 ( − , R ) b e the sub obje ct c onsisting of strictly p ositive functions. Then for f ∈ C 0 ( U, R ) , the truth value χ P ( f ) ∈ Ω( U ) is the op en subset { x ∈ U | f ( x ) > 0 } . So the internal statement “ f > 0 ” is not simply true or false on U ; it is true pr e cisely on the op en r e gion wher e p ositivity holds. This is an elementary but very instructive mo del of lo c al truth. One ma y push this muc h further. In an elemen tary top os, one can in terpret conjunction, implication, quan tification, equality , and existence. This yields a higher-order intuitionistic logic internal to the top os. F or the presen t pap er, the full formal apparatus is unnecessary , but the following heuristic is indisp ensable: Statemen ts in a top os are ev aluated relative to the lo cal context, and existential claims amoun t to the existence of compatible lo cal data. This b ecomes esp ecially transparent in Kripke–Jo yal semantics. In a sheaf top os, to say that an op en set U forces an existential statement means that U can b e cov ered by smaller op ens on which witnesses exist, with the exp ected compatibilit y conditions. Th us the logical meaning of existence is already sheaf-theoretic. Prop osition 6.4 (Kripke–Jo y al heuristic for existence) . L et Sh ( X ) b e a she af top os and let ϕ ( x ) b e an internal statement. Then the assertion that “ U ⊩ ∃ x ϕ ( x ) ” should b e r e ad as fol lows: ther e exists an op en c over U = S i U i and se ctions x i over U i such that e ach U i for c es ϕ ( x i ) , with c omp atibility on overlaps whenever r e quir e d. Prop osition 6.5 (Kripk e–Jo yal heuristic for universal statements) . L et Sh ( X ) b e a she af top os and let ϕ ( x ) b e an internal statement ab out se ctions of a she af F . Then “ U ⊩ ∀ x ϕ ( x ) ” me ans that for every op en subset V ⊆ U and every se ction x ∈ F ( V ) , one has V ⊩ ϕ ( x ) . Thus universal truth is stable under r estriction to smal ler op ens. 8 Remark 6.6. This differ enc e b etwe en existential and universal statements is crucial. Existential truth may have to b e establishe d only after p assing to a c over, wher e as universal truth must survive every further lo c alization. That asymmetry is one of the signatur es of intuitionistic and ge ometric r e asoning. Example 6.7 (Lo cal existence versus global existence) . L et X = S 1 and let F b e the she af of c ontinuous r e al-value d functions. Consider the statement ∃ g ∈ F ( g 2 = f ) for a fixe d se ction f ∈ F ( X ) . If f is everywher e p ositive, then this statement is glob al ly true. If f changes sign, it is glob al ly false. But even when no glob al squar e r o ot exists, one may stil l have lo c al squar e r o ots on a suitable op en c over. In internal language, this me ans that the existential statement c an b e for c e d lo c al ly without b eing for c e d glob al ly. Example 6.8 (Implication as lo cal stabilit y) . F or sub obje cts A, B  → F in a top os, the implic ation A ⇒ B is another sub obje ct of F , char acterize d internal ly by the pr op erty that a se ction b elongs to it exactly wher e memb ership in A lo c al ly for c es memb ership in B . In a she af top os, implic ation is ther efor e not a pur ely p ointwise op er ation; it r e c or ds lo c al stability of entailment. Remark 6.9. The truth values in an elementary top os form internal ly a Heyting algebr a r ather than, in gener al, a Bo ole an algebr a. F or this r e ason, exclude d midd le ne e d not hold. The lo gic al me aning of this fact is discusse d in A pp endix B . F or now, the imp ortant p oint is that lo c al truth is usual ly subtler than classic al binary truth. Prop osition 6.10 (Internal logic as organized lo cal reasoning) . L et E b e a Gr othendie ck top os. Then the interpr etation of pr e dic ates by sub obje cts, of truth values by Ω , and of quantifiers by the internal adjoint structur e of the top os or ganizes r e asoning in such a way that lo c al c omp atibility data c an b e expr esse d and manipulate d internal ly. Pr o of sketch. Predicates are represented by monomorphisms and therefore b y classifying arrows in to Ω . Products and exp onen tials supp ort conjunction and function spaces. The existence of appropriate adjoints for pullbac k along pro jections supplies the categorical conten t of quan tification. In a Grothendieck top os, Kripke–Jo yal semantics mak es these constructions explicitly lo cal. Remark 6.11. This do es not me an that al l r e asoning in a top os is mer ely lo c al p atchwork. R ather, it me ans that glob al truth is me diate d by lo c al truth and gluing. The p oint is not the denial of glob al structur e, but the discipline d r e c onstruction of glob al structur e fr om lo c al data. This p ersp ective is crucial for the in tended cryptographic application. The later pap er [ 2 ] uses Grothendiec k top ologies and sheaf-lik e organization precisely b ecause attack er knowledge, observ- abilit y , and compatibility of information are not simply global yes/no matters. The internal logic of a top os is therefore not an ornamental abstraction; it is a mo del for lo cal reasoning with controlled gluing. 7 Practice: w orking with shea v es, top oi, and in ternal logic The follo wing exercises are in tended to help the reader consolidate the main ideas of the pap er b efore passing to the explicitly cryptographic discussion. They are not logically necessary for the later sections, but they are p edagogically useful. Solutions are collected in App endix C . 9 Exercise 1. Let X b e a top ological space and let Ω ∈ Sh ( X ) b e the sub ob ject classifier. Show that for each op en set U ⊆ X , the set Ω( U ) may b e iden tified with the set of op en subsets of U . Explain why this means that truth v alues in Sh ( X ) are lo cal. Exercise 2. Let G b e a sheaf of groups on a site ( C , J ) and let P b e a G -torsor. Choose lo cal sections on a cov ering family and derive the corresp onding co cycle g ij . V erify formally that the co cycle condition on triple ov erlaps expresses compatibility of transp ort. Exercise 3. In a sheaf top os, explain in words the difference b etw een the meanings of U ⊩ ∃ x ϕ ( x ) and U ⊩ ∀ x ϕ ( x ) . Wh y is the first statemen t allow ed to pass to a cov er while the second must hold after ev ery restriction? Exercise 4. Let F = C 0 ( − , R ) on a top ological space X and let P  → F b e the sub ob ject of p ositiv e functions. F or a section f ∈ F ( U ) , describ e concretely the classifying map χ P ( f ) ∈ Ω( U ) . What do es this say ab out the in ternal meaning of the predicate “ f > 0 ”? Exercise 5. Let E b e an elementary top os. Explain why the data of finite limits, exp onentials, and a sub ob ject classifier make E resem ble a universe of sets. Whic h of these ingredients is most directly resp onsible for internal truth v alues? 8 T orsors and lo cal symmetry inside a top os W e now return to torsors, but from the more mature viewp oint provided by top os theory . Once one is working in a top os E , it makes sense to sp eak of in ternal groups, internal actions, and therefore in ternal torsors. Definition 8.1. A n internal group in a c ate gory with finite pr o ducts is an obje ct G e quipp e d with multiplic ation, unit, and inversion morphisms satisfying the usual gr oup diagr ams internal ly. Definition 8.2. L et G b e an internal gr oup in a top os E . A G -torsor in E is an obje ct P with a right G -action such that internal ly: (i) P is inhabite d lo c al ly, (ii) the action is fr e e, (iii) the action is tr ansitive. Equivalently, the c anonic al map P × G − → P × P, ( p, g ) 7→ ( p, p · g ) , is an isomorphism, and P → 1 is an epimorphism. This definition shows clearly how torsors b elong to the top os world. The absence of a global origin is not a defect. It is a legitimate internal form of symmetry . What matters is not the existence of a preferred p oint, but the existence of lo cal p oin ts and the ability to transp ort among them. 10 Prop osition 8.3. L et ( C , J ) b e a site and let G b e a she af of gr oups. Then G -torsors on the site ar e pr e cisely torsors for the internal gr oup G in the top os Sh ( C , J ) . Pr o of sketch. The site-theoretic definition of a G -torsor is exactly the sheaf-theoretic v ersion of lo cal inhabitation together with free transitiv e action. Inside the top os, these b ecome the internal epimorphism condition and the isomorphism P × G ∼ = P × P . The tw o definitions translate directly in to one another. Remark 8.4. This pr op osition explains why the r oute taken in this p ap er is c onc eptual ly c oher ent. The first torsor note [ 1 ] intr o duc e d torsors external ly. The pr esent p ap er shows how the same ide a is natur al ly internalize d in a top os. That is alr e ady enough to suggest why a later she af-the or etic ac c ount of crypto gr aphic se curity [ 2 ] should tr e at lo c al know le dge and c omp atibility structur al ly r ather than ad ho c. 9 T o ward Σ -proto cols and cryptographic securit y W e no w explain, at a conceptual lev el, why the progression from torsors to top oi is relev ant to Σ -proto cols. The p oint is not that a standard proto col transcript is literally a sheaf or a torsor in any naive sense. The p oint is that the mathematics needed to understand lo cal consistency , sim ulability , and global kno wledge already has a natural language in sheaf and top os theory . In the torsor note [ 1 ], a recurring theme was the p ossibility of coheren t transport without a distinguished origin. In [ 2 ], the analogous cryptographic theme is that local views, attac k er observ ations, and sim ulated pieces of evidence can carry structure without there being a single naiv e global witness visible everywhere. The relev an t mathematics is therefore not merely the mathematics of global ob jects, but the mathematics of lo cally av ailable compatible ob jects. This is where Grothendieck top ologies matter. A Grothendieck top ology formalizes what coun ts as a cov ering family , and hence what coun ts as a lo cal view sufficien t for gluing. In the cryptographic setting of [ 2 ], one studies information through attack er mo dels and cov erings generated by accessible observ ations. Shea v es then enco de consistency of information across such lo cal observ ations. The role of top os theory is to provide the ambien t logic in which such lo cal reasoning can b e carried out coheren tly . Remark 9.1. One may think heuristic al ly of a simulator as pr oviding lo c al trivializations, not of a ge ometric bund le, but of an information structur e. A suc c essful simulation says that c ertain lo c al observables c an b e pr o duc e d c omp atibly. The p assage fr om these lo c al pie c es to a me aningful glob al statement is ther efor e natur al ly r eminisc ent of desc ent. This is not a the or em of pr oto c ol the ory by itself, but it is pr e cisely the kind of structur al analo gy that makes she af-the or etic and top os-the or etic language fruitful. The author’s pap er [ 2 ] should b e understo o d against this background. That pap er do es not emerge suddenly from ordinary set-based proto col notation. It emerges from a sustained lo cal-to-global viewp oin t: first torsors, then sheav es and top oi, and only then cryptographic securit y . Accordingly , the present pap er and the earlier torsor note are mean t as genuine preparation, not as unrelated exp ository side remarks. The in tended line is torsors and co cycles − → shea ves and descent − → topoi and internal logic − → cryptographic lo cal reasoning . 11 10 Outlo ok Sev eral directions remain op en. One ma y first strengthen the present introdu ctory discussion by giving a fuller account of geometric morphisms, sites of definition, and the relation b etw een internal and external viewp oints. Second, one may study torsors not merely in an ordinary sheaf top os, but in more refined settings in volving stacks or higher top oi. Third, and most relev ant for the author’s curren t program, one may try to make the cryptographic heuristics of Section 9 more precise by iden tifying exact sheaf-theoretic structures attached to attack er mo dels and classes of proto cols. F or the momen t, how ever, the aim of this pap er is preparatory . T ogether with the earlier torsor note [ 1 ], it is in tended to mak e the later pap er [ 2 ] conceptually approachable. If the presen t discussion succeeds, the reader should come aw a y with the sense that top os theory is not an optional abstract sup erstructure, but a natural enlargement of the local-to-global mathematics already presen t in torsors, and therefore a serious candidate language for the structural study of cryptographic security . 11 Conclusion W e hav e argued that the passage from torsors to top oi is not an accidental c hange of language, but a structural enlargement of ideas already present in torsor theory . Lo cal trivialit y , compatibilit y on o verlaps, co cycle data, and descent already p oin t tow ard a world in whic h lo cal information and its gluing are primary . Sheaf theory organizes this viewp oint systematically , and top os theory extends it further into a setting with its o wn internal notions of truth, existence, and function. F or that reason, the presen t pap er has pursued t wo goals at once. On the one hand, it has served as an introduction to sites, shea v es, top oi, internal logic, and related categorical to ols. On the other hand, it has b een written as explicit preparation for the author’s cryptographic pap er [ 2 ], in which Grothendiec k top ologies and sheaf-theoretic reasoning are used as genuine structural ingredients. The earlier torsor note [ 1 ] supplied the first step of this line of thought; the present pap er develops the second step by making the top os-theoretic bac kground more fully visible. The app endices were included for a p edagogical reason. A reader who is not yet fully comfortable with category theory , internal logic, or limits and colimits should still b e able to use this pap er as a working preparation for later study . In that sense, the pap er is mean t not only as a conceptual bridge, but also as a practical companion. If it succeeds, the reader should b e able to approach [ 2 ] with a clearer understanding of why lo cal reasoning, gluing, and sheaf-theoretic structure b elong naturally to the mathematical study of cryptographic securit y and Σ -proto cols. References [1] T. Inoué, A n I ntr o duction to T orsors in Mathematics with a V iew T owar d Σ -Pr oto c ols in Crypto gr aphy , preprint, arXiv, 2026. [2] T. Inoué, Gr othendie ck T op olo gies and She af-The or etic F oundations of Crypto gr aphic Se curity: A ttacker Mo dels and Σ -Pr oto c ols as the First Step , preprint, arXiv, 2026. [3] S. Mac Lane and I. Mo erdijk, She aves in Ge ometry and L o gic: A First Intr o duction to T op os The ory , Springer, 1992. 12 [4] P . T. Johnstone, Sketches of an Elephant: A T op os The ory Comp endium, V ol. 1 , Oxford Univ ersity Press, 2002. [5] P . T. Johnstone, Sketches of an Elephant: A T op os The ory Comp endium, V ol. 2 , Oxford Univ ersity Press, 2002. [6] M. Artin, A. Grothendiec k, and J. L. V erdier, Thé orie des top os et c ohomolo gie ètale des schémas (SGA 4) , Lecture Notes in Mathematics 269, 270, 305, Springer, 1972–1973. [7] S. Goldw asser, S. Micali, and C. Rack off, The K now le dge Complexity of Inter active Pr o of Systems , SIAM Journal on Computing 18 (1989), 186–208. [8] I. Damgård, On Σ -Pr oto c ols , Lecture notes, Aarhus Universit y , 2002. [9] C. P . Sc hnorr, Efficient Identific ation and Signatur es for Smart Car ds , In: A dvanc es in Cryptolo gy – CR YPTO ’89 , Lecture Notes in Computer Science 435, Springer, 1990, pp. 239– 252. T akao Inoué F acult y of Informatics Y amato Univ ersit y Kata yama-c ho 2-5-1, Suita, Osaka, 564-0082, Japan inoue.takao@y amato-u.ac.jp (P ersonal) takaoapple@gmail.com (I prefer my p ersonal mail) A A brief in tro duction to category theory This app endix is included for readers who are not yet comfortable with category theory but would still like to follow the main text. The aim is not to replace a full course, but to provide enough language and technique for basic w ork with sheav es, topoi, and internal logic. In particular, we review categories, functors, natural transformations, universal constructions, th e Y oneda lemma, and Cartesian closed categories. A.1 Categories, morphisms, and commutativ e diagrams A c ate gory C consists of: (i) a class of ob jects A, B , C, . . . ; (ii) for each pair of ob jects A, B , a set Hom C ( A, B ) of morphisms A → B ; (iii) for each triple A, B , C , a comp osition law Hom C ( B , C ) × Hom C ( A, B ) → Hom C ( A, C ) , ( g , f ) 7→ g ◦ f ; (iv) for each ob ject A , an iden tit y morphism id A : A → A . 13 These data satisfy asso ciativity ( h ◦ g ) ◦ f = h ◦ ( g ◦ f ) and the identit y la ws id B ◦ f = f , f ◦ id A = f . T ypical examples are the category Set of sets, the category Grp of groups, and the category Sh ( X ) of sheav es on a space X . In all of these, one studies not only ob jects in isolation, but also the maps b et w een them and the wa y constructions b ehav e functorially . A diagram in a category is said to c ommute if every pair of morphism comp osites with the same source and target are equal. F or example, A f − → B ↘ h ↓ g C comm utes if and only if g ◦ f = h . Many categorical definitions are nothing but the statement that a certain diagram commutes and is universal among all such diagrams. A.2 F unctors and natural transformations A functor F : C → D sends each ob ject A of C to an ob ject F ( A ) of D and each morphism f : A → B to a morphism F ( f ) : F ( A ) → F ( B ) in such a w a y that F ( g ◦ f ) = F ( g ) ◦ F ( f ) , F (id A ) = id F ( A ) . Th us a functor preserves categorical structure. F or example, the p o wer-set construction X 7→ P ( X ) is a functor Set → Set , and the assignmen t U 7→ F ( U ) for a sheaf F is a contra v ariant functor from op en sets to sets. If F , G : C → D are functors, a natur al tr ansformation η : F ⇒ G is a family of morphisms η A : F ( A ) → G ( A ) indexed by ob jects A of C , such that for ev ery morphism f : A → B , the square F ( A ) η A − → G ( A ) ↓ F ( f ) ↓ G ( f ) F ( B ) η B − − → G ( B ) comm utes. Naturality expresses the idea that the maps η A are compatible with all structure maps in sight. This is one of the basic wa ys in which category theory formalizes “the same construction in all contexts. ” 14 A.3 Univ ersal prop erties A large part of category theory is the study of ob jects defined by univ ersal prop erties. The p oint is that an ob ject is characterized not b y the material from which it is built, but by the maps into or out of it. F or instance, a pro duct of ob jects A and B is an ob ject A × B together with pro jections π 1 : A × B → A, π 2 : A × B → B , suc h that for eve ry ob ject T and ev ery pair of maps f : T → A, g : T → B , there exists a unique map ⟨ f , g ⟩ : T → A × B with π 1 ◦ ⟨ f , g ⟩ = f and π 2 ◦ ⟨ f , g ⟩ = g . The relev ant diagram is T ↙ f ↓ ⟨ f , g ⟩ ↘ g A π 1 ← − A × B π 2 − → B The dashed arro w is determined uniquely b y the univ ersal prop erty . This uniqueness is often more imp ortan t than any explicit formula. Another basic univ ersal construction is the pullback. Giv en f : A → C and g : B → C , a pullbac k is an ob ject A × C B fitting into a commutativ e square A × C B − → B ↓ ↓ g A f − → C that is universal among all such squares. Pullbacks o ccur constan tly in sheaf theory , for example when one restricts lo cal data to ov erlaps. A.4 Limits and colimits T w o of the most basic families of univ ersal constructions are limits and c olimits . A limit is a univ ersal wa y of receiving compatible maps from an outside ob ject into a diagram, whereas a colimit is a universal w ay of sending compatible maps out of a diagram. Pro ducts, pullbacks, and equalizers are examples of limits. Copro ducts, pushouts, and co equalizers are examples of colimits. It is useful to b egin with examples rather than the full abstract definition. W e hav e already seen that a pro duct A × B is characterized b y maps in to A and B . This makes it a limit of the discrete diagram consisting of A and B . Similarly , the pullback A × C B is the limit of the cospan A f − → C g ← − B . 15 The pullbac k is the most basic w ay to enforce compatibility conditions. In geometry and sheaf theory , it app ears whenev er one restricts tw o pieces of data to a common o v erlap. Another imp ortant limit is the e qualizer of tw o parallel arro ws A ⇒ f g B . An equalizer is an ob ject E with a map e : E → A such that f ◦ e = g ◦ e, and which is univ ersal with this prop erty . In Sets , one may think of it concretely as the subset E = { a ∈ A : f ( a ) = g ( a ) } . So the equalizer is the universal ob ject on whic h the t w o maps b ecome equal. It is a limit b ecause it receives maps from every ob ject on which f and g already agree. Dually , a copro duct A ⨿ B is characterized by inclusions A → A ⨿ B , B → A ⨿ B , suc h that maps out of A ⨿ B are exactly pairs of maps out of A and B . Likewise, a pushout is the colimit of a span A f ← − C g − → B , and can b e thought of as the univ ersal wa y of gluing A and B along the common piece C . Dually to the equalizer, one has the c o e qualizer of tw o parallel arrows A ⇒ f g B . A co equalizer is an ob ject q : B → Q such that q ◦ f = q ◦ g, and whic h is univ ersal with this prop erty . In Sets , one may think of it as the quotient of B obtained b y forcing f ( a ) ∼ g ( a ) for all a ∈ A. So the co equalizer is the universal ob ject in which the t w o maps b ecome equal after passing to a quotien t. It is a colimit b ecause ev ery map out of B that already identifies f ( a ) and g ( a ) factors uniquely through Q . More generally , let D : J → C b e a diagram. A c one from an ob ject L to D is a family of maps L → D ( j ) compatible with all arrows in J . A limit of D is a universal cone to D . Dually , a c o c one from D to an ob ject M is a family of maps D ( j ) → M compatible with all arrows in J , and a c olimit of D is a universal co cone. In practice, one do es not need the whole abstract definition at once. What matters is to learn the logic of universal problems. T o construct a limit, one asks: “What ob ject receiv es compatible maps 16 from everything else in the most universal wa y?” T o construct a colimit, one asks: “What ob ject is obtained by forcing compatible outgoing maps in the most universal w a y?” This wa y of thinking is cen tral in category theory and app ears constantly in sheaf theory , descent, and top os theory . It is also helpful to pause ov er the traditional names pr oje ctive limit (also called inverse limit ) and inductive limit (also called dir e ct limit ). An inv erse system is typically a diagram X 0 ← − X 1 ← − X 2 ← − · · · in which information is pushed backw ards along transition maps. Its limit is the ob ject of all compatible families ( x 0 , x 1 , x 2 , . . . ) with each x n matc hing the image of x n +1 . So an inv erse limit expresses the idea of keeping trac k of data at every stage while enforcing p erfect compatibility across all stages. One may think of it as an ob ject of coherent approximations. Dually , a direct system is typically a diagram X 0 − → X 1 − → X 2 − → · · · in which information mo ves forward. Its colimit identifies elemen ts that b ecome equal at some later stage. Thus a direct limit expresses the idea of building a larger object b y rep eatedly adjoining data and imp osing the iden tifications generated by the transition maps. One may think of it as the ev entual ob ject obtained from a pro cess of accum ulation. In Sets , for example, the in verse limit of a sequence of quotient maps records sequences that remain compatible through ev ery stage, whereas the direct limit of a c hain of inclusions is just the union of all stages. These are excellent mental pictures to keep in mind: in verse limits capture c omp atible families , and direct limits capture eventual identific ation and ac cumulation . A.5 Exercises on limits and colimits Because limits and colimits are often conceptually difficult at first, it is helpful to practice them in concrete situations. Exercise C1. Let C b e a category with pullbacks. Giv en arrows A f − → C g ← − B , sho w that a morphism X → A × C B is equiv alent to the data of morphisms u : X → A, v : X → B suc h that f ◦ u = g ◦ v . Explain why this justifies the slogan that the pullbac k is the ob ject of compatible pairs. Exercise C2. In the category of sets, let A = { 1 , 2 } , B = { a, b } , C = {∗} with the unique maps A → C and B → C . Compute the pullback A × C B explicitly . Then compute the pushout of the span A id A ← − − A h − → B for a map h : A → B with h (1) = a and h (2) = b . 17 Exercise C3. Let D : J → C b e a diagram and let K : J → 1 b e the unique functor to the terminal category . Explain why a left Kan extension Lan K D is the same thing as a colimit of D , and wh y a right Kan extension Ran K D is the same thing as a limit of D . Y ou do not need to prov e the most general theorem formally; it is enough to unpac k the corresp onding universal prop erties carefully . Exercise C4. Consider the inv erse system in Sets · · · π 4 − → Z / 16 Z π 3 − → Z / 8 Z π 2 − → Z / 4 Z π 1 − → Z / 2 Z , where each map is reduction mo dulo a smaller p o w er of 2 . Describ e concretely what an element of the inv erse limit of this system is. Explain why this inv erse limit should b e thought of as a coheren t system of appro ximations rather than as a single residue class mo dulo one fixed p ow er of 2 . Exercise C5. Consider the direct system in Sets { 1 , . . . , n }  → { 1 , . . . , n + 1 }  → { 1 , . . . , n + 2 }  → · · · , where each map is the evident inclusion. Compute its colimit. Then explain in w ords wh y this example captures the intuitiv e meaning of a direct limit as an “even tual union” of stages. A.6 Kan extensions Kan extensions are among the most imp ortant general constructions in category theory . They are, in a precise sense, the universal wa y of extending a functor along another functor. Many familiar constructions are sp ecial cases of Kan extensions, including several limits, colimits, sheafification-type pro cedures, and adjoin ts. Supp ose we are given categories A , B , C , a functor K : A → B , and another functor F : A → C . W e w ould like to replace F b y a functor defined on B . In other words, we seek a functor G : B → C whic h, in some univer sal sense, b est approximates the idea that G ◦ K ≈ F. There are tw o v ersions of this problem. A left K an extension of F along K , written Lan K F , is a functor B → C equipp ed with a natural transformation η : F ⇒ (Lan K F ) ◦ K 18 whic h is universal among all such pairs. This means that if H : B → C is any other functor and α : F ⇒ H ◦ K is any natural transformation, then there exists a unique natural transformation α : Lan K F ⇒ H suc h that α = ( αK ) ◦ η. Dually , a right K an extension of F along K , written Ran K F , is a functor B → C equipp ed with a natural transformation  : (Ran K F ) ◦ K ⇒ F whic h is universal among all suc h pairs. Thus right Kan extensions solv e the b est universal extension problem on the limiting side, while left Kan extensions solv e it on the colimiting side. One reason Kan extensions matter is that they unify many definitions. F or example, if K : A → 1 is the unique functor to the terminal category , then Lan K F is essentially the colimit of F , while Ran K F is essentially the limit of F , provided these exist. So limits and colimits can themselves b e regarded as Kan extensions to a p oint. A second reason is that adjoints may b e characterized by Kan extensions. If K : A → B has a left adjoin t, then that left adjoint may b e describ ed as a suitable Kan extension of the identit y functor. Th us Kan extensions sit very near the center of category theory: they enco de the general pattern of extending structure universally . F or b eginners, the slogan is: A Kan extension is the universal b est w a y to contin ue a functor b ey ond the category where it was originally defined. This slogan is not a definition, but it captures the right intuition. Left Kan extensions are built from colimit-t yp e information, and right Kan extensions from limit-type information. F or that reason, they form a bridge b etw een functorialit y and the universal constructions discussed ab ov e. A.7 Represen table functors and the Y oneda p oint of view Fix a category C . F or any ob ject A , one obtains a contra v ariant functor h A := Hom C ( − , A ) : C op → Set , called the functor r epr esente d by A . It sends an ob ject X to the set of morphisms X → A . Morphisms in to A therefore define a functorial “profile” of A as seen from the rest of the category . The basic insigh t of Y oneda theory is that an ob ject is completely determined by this profile. That is, to understand A , it is often enough to understand all maps into A . This principle is so imp ortant that it is worth stating in full. 19 Theorem A.1 (Y oneda lemma) . L et C b e a lo c al ly smal l c ate gory, let A b e an obje ct of C , and let F : C op → Set b e a pr eshe af. Then ther e is a natur al bije ction Nat( h A , F ) ∼ = F ( A ) . Her e Nat( h A , F ) denotes the set of natur al tr ansformations fr om h A to F . Pr o of. Given a natural transformation η : h A ⇒ F , consider the element η A (id A ) ∈ F ( A ) . This defines a function Φ : Nat( h A , F ) → F ( A ) , Φ( η ) = η A (id A ) . Con versely , giv en x ∈ F ( A ) , define for eac h ob ject X a map η x X : h A ( X ) = Hom C ( X , A ) → F ( X ) b y η x X ( f ) = F ( f )( x ) . W e m ust chec k that η x is natural. If u : Y → X , then for any f : X → A we ha v e η x Y ( f ◦ u ) = F ( f ◦ u )( x ) = F ( u )( F ( f )( x )) = F ( u )( η x X ( f )) , so the naturality square commutes. Th us x determines a natural transformation η x . This gives a function Ψ : F ( A ) → Nat( h A , F ) , Ψ( x ) = η x . No w Φ(Ψ( x )) = η x A (id A ) = F (id A )( x ) = x, so Φ ◦ Ψ = id F ( A ) . Conv ersely , if η : h A ⇒ F , then for each f : X → A , naturality applied to f giv es η X ( f ) = F ( f )( η A (id A )) =  η η A (id A )  X ( f ) . Hence Ψ(Φ( η )) = η . So Φ and Ψ are in v erse bijections. A particularly imp ortan t consequence is the Y one da emb e dding C  → [ C op , Set ] , A 7→ h A , whic h is fully faithful. Thus one may regard ev ery category as living inside a category of set-v alued functors. This p ersp ective lies in the bac kground of sheaf theory and top os theory , since a sheaf is precisely a presheaf satisfying an additional gluing condition. 20 A.8 Cartesian closed categories A category C is called Cartesian close d if it has finite pro ducts and, for every pair of ob jects A, B , an exp onential ob ject B A . The exp onen tial is characterized by a natural bijection Hom C ( X × A, B ) ∼ = Hom C ( X , B A ) for all ob jects X . Equiv alently , maps out of X × A in to B are represented by maps out of X in to B A . This is the categorical form of currying. In the category Set , the exp onential B A is the set of functions from A to B . The adjunction ab ov e sa ys exactly that a function X × A → B is the same thing as a function X → B A . Th us Cartesian closed categories are categories in whic h internal function ob jects exist. The reason CCCs matter here is that elemen tary top oi are, by definition, finite-limit categories that are Cartesian closed and p ossess a sub ob ject classifier. So if one wishes to understand top oi, one m ust b ecome comfortable with the logic of pro ducts and exp onentials. The exp onential ob ject is what allows implication and function t yp es to b e represented internally . This is one of the bridges from categorical structure to internal logic. A.9 Wh y these notions matter for the main text The main b o dy of this pap er uses all of the ideas ab ov e. Sites and sheav es are formulated in categories. T orsors on a site are functorial ob jects defined b y lo cal universal prop erties. T op oi are categories with enough internal structure to supp ort logic. The Y oneda lemma lies silen tly b ehind the passage from ob jects to presheav es, and Cartesian closedness lies b ehind in ternal function ob jects and implication. Th us category theory is not merely background notation here. It is the language in which the lo cal-to-global structures of torsors, sheav es, and top oi b ecome visible and manageable. B In ternal logic and in tuitionistic logic The internal logic of an elementary top os is, in general, not classical but in tuitionistic. This statemen t deserves a careful explanation, since it is one of the most imp ortant conceptual lessons of top os theory . In classical logic, every prop osition is either true or false, and the law of excluded middle P ∨ ¬ P is alwa ys v alid. Equiv alently , truth v alues are organized by a Bo olean algebra. In the category Set , this is reflected in the fact that the sub ob ject classifier is the tw o-elemen t set { 0 , 1 } , and sub ob jects of a set corresp ond to characteristic functions with Bo olean v alues. In a general top os, the ob ject of truth v alues is Ω . The collection of its global or lo cal truth v alues is not, in general, Bo olean. Rather, it carries the structure of a Heyting algebra. A Heyting algebra 21 still supp orts conjunction, disjunction, implication, and a notion of negation, but negation is weak er than in Bo olean logic. The formula P ∨ ¬ P need not hold. F or this reason, the internal logic is usually intuitionistic. This is not a defect or an omission. It is a faithful reflection of the fact that truth in a top os may b e lo cal. In a sheaf top os Sh ( X ) , a prop osition ov er an op en set U ma y hold on one part of U and fail on another. Its truth v alue is therefore represen ted by an op en subset of U . Suc h lo cal truth v alues naturally form a Heyting algebra of op ens, not in general a Bo olean algebra. Thus intuitionistic logic app ears b ecause the geometry of lo cality is itself non-Bo olean. One useful wa y to remember the difference is this: classical logic asks whether a prop osition has already b een globally decided, whereas in tuitionistic logic asks whether one can construct or verify it in the av ailable context. In a top os, the av ailable context is often lo cal. Th us intuitionistic logic is the natural logic of constructive lo cal reasoning. The relation with Kripke–Jo y al seman tics makes this especially transparen t. T o say that U ⊩ ∃ x ϕ ( x ) means that one can pass to a cov er of U and find lo cal witnesses there. T o say that U ⊩ ∀ x ϕ ( x ) means that the statement must contin ue to hold after every restriction to smaller opens. These are precisely the kinds of rules one exp ects in intuitionistic semantics, where truth is stable under refinemen t but may require lo cal construction. It is therefore reasonable to summarize the situation as follo ws. T op os theory provides an ambien t categorical setting in which geometric lo cality , sheaf-theoretic gluing, and intuitionistic reasoning b elong together. The in ternal logic of a top os is in tuitionistic b ecause truth is mediated by context, restriction, and compatibility rather than b y a single global Bo olean decision pro cedure. This observ ation matters for the broader aims of the present pap er. The later cryptographic applications of [ 2 ] are concerned with lo cal observ ability , lo cal consistency , controlled gluing of information, and the disciplined passage from partial data to global structure. F or such purp oses, in tuitionistic logic is not merely philosophically in teresting; it is structurally appropriate. It supplies a logical language in which the difference b etw een global kno wledge and lo cally forced information can b e expressed without collapse. C Solutions to the exercises Solution to Exercise 1. F or a sheaf top os Sh ( X ) , the sub ob ject classifier Ω assigns to eac h op en set U the set of op en subsets of U . Indeed, a sub ob ject of the terminal sheaf restricted to U is determined by the op en region where it is inhabited, so Ω( U ) is naturally iden tified with the lattice of op ens of U . Hence a truth v alue o v er U is not just a bit but a region of U on which a statement holds. This is why truth is lo cal. Solution to Exercise 2. Cho ose lo cal sections s i ∈ P ( U i ) . On each o verlap U ij = U i × U U j , free transitivit y gives a unique element g ij ∈ G ( U ij ) such that s j = s i · g ij . On a triple ov erlap U ij k , one may compare s k with s i directly and via s j : s k = s i · g ik = s j · g j k = s i · g ij g j k . By uniqueness of transp ort, g ik = g ij g j k . This is the co cycle condition. 22 Solution to Exercise 3. The statement U ⊩ ∃ x ϕ ( x ) means that after refining U b y a cov er, one can find lo cal witnesses on the pieces. Th us existence is allow ed to b e lo cal. By con trast, U ⊩ ∀ x ϕ ( x ) means that ev ery section ov er every smaller op en V ⊆ U satisfies ϕ . Universal truth m ust therefore surviv e every further lo calization. This difference reflects the in tuitionistic meaning of quantifiers. Solution to Exercise 4. F or f ∈ C 0 ( U, R ) , the classifying map sends f to the op en subset χ P ( f ) = { x ∈ U | f ( x ) > 0 } . So the internal predicate “ f > 0 ” is interpreted as the op en region where p ositivity holds. This is a concrete mo del of a truth v alue in the sheaf top os. Solution to Exercise 5. Finite limits pro vide pro ducts, terminal ob jects, pullbac ks, and equalizers, so one can form man y familiar set-like constructions. Exp onentials pro vide in ternal fun ction ob jects. The sub ob ject classifier provides an in ternal ob ject of truth v alues and allo ws predicates to b e represen ted categorically . Among these ingredien ts, the sub ob ject classifier is the one most directly resp onsible for internal truth v alues, although the others are needed to supp ort the broader logical and categorical structure. Solution to Exercise C1. By the universal prop erty of the pu llbac k, to give a map φ : X → A × C B is equiv alen t to giving maps u = π A ◦ φ : X → A, v = π B ◦ φ : X → B suc h that f ◦ u = g ◦ v . Con versely , an y such compatible pair ( u, v ) induces a unique map X → A × C B . Th us the pullbac k is precisely the univ ersal ob ject parameterizing compatible pairs of maps into A and B ov er C . Solution to Exercise C2. Since b oth maps A → C and B → C are unique and C has only one elemen t, the compatibility condition is automatic. Hence A × C B ∼ = A × B as sets, so explicitly A × C B = { (1 , a ) , (1 , b ) , (2 , a ) , (2 , b ) } . F or the pushout of A id A ← − − A h − → B , one starts from the disjoint union A ⨿ B and identifies eac h elemen t x ∈ A on the left with its image h ( x ) ∈ B on the right. Since h (1) = a and h (2) = b , the relations identify 1 with a and 2 with b . After imp osing these iden tifications, no new p oints remain b eyond the tw o classes represented by a and b . So the pushout is canonically isomorphic to B . 23 Solution to Exercise C3. A functor from the terminal category 1 to C is just the choice of a single ob ject of C . Thus a functor Lan K D : 1 → C is exactly an ob ject L of C . The natural transformation D ⇒ (Lan K D ) ◦ K is then precisely a co cone from the diagram D to the ob ject L . The universal prop erty of the left Kan extension says that this co cone is universal among all co cones out of D . But that is exactly the definition of a colimit. Dually , a right Kan extension Ran K D : 1 → C amoun ts to an ob ject R together with a natural transformation (Ran K D ) ◦ K ⇒ D , whic h is exactly a cone from R to the diagram D . Its universal prop erty says that this cone is univ ersal among all cones into D . That is exactly the definition of a limit. Solution to Exercise C4. An elemen t of the in v erse limit is a sequence ( x 1 , x 2 , x 3 , . . . ) , x n ∈ Z / 2 n Z , suc h that each x n +1 reduces to x n mo dulo 2 n . So one is not choosing a single residue class once and for all; one is choosing one residue class mo dulo 2 , one mo dulo 4 , one mo dulo 8 , and so on, in suc h a w ay that all of them fit together compatibly . This is wh y the inv erse limit is b est view ed as a coherent family of approximations. Each stage gives finite information, and the limit remem b ers all stages simultaneously . This is exactly the pattern that later leads to ob jects suc h as the 2 -adic in tegers. Solution to Exercise C5. Since every map in the system is an inclusion, no new identifications are imp osed b eyond those already present in the sets themselves. Therefore the colimit is simply the union of all stages: lim − → n { 1 , . . . , n } ∼ = N > 0 . This illustrates the intuitiv e meaning of a direct limit very well. One starts with a sequence of ob jects that grow by adjoining new elemen ts, and the colimit is the even tual ob ject containing ev erything that app ears at some stage. In this example, ev ery p ositive in teger app ears after finitely man y steps, so the direct limit is the set of all p ositive in tegers. So limits and colimits are Kan extensions to the terminal category: colim D ≃ Lan K D , lim D ≃ Ran K D . 24 D A p ossible lecture plan This note can b e taught in several w a ys, dep ending on the background of the audience. If one aims at a serious but still in tro ductory course for studen ts who are not yet fully comfortable with category theory , a natural format is a course of ab out 10–12 lectures , eac h of length 90 minutes . A shorter course of 8 lectures is p ossible if some of the app endices are treated as background reading, while a more leisurely 14-lecture version would allow more examples and exercises. F or most purp oses, ho wev er, the 10–12 lecture format seems the most realistic. The guiding principle of the course is to mo ve from concrete lo cal-to-global intuition to the more abstract language of shea ves, top oi, and internal logic, and only then to explain why this matters for the sheaf-theoretic approac h to cryptographic security and Σ -proto cols. In this sense, the course is not merely a survey of topics, but a structured preparation for the pap er Gr othendie ck T op olo gies and She af-The or etic F oundations of Crypto gr aphic Se curity: A ttacker Mo dels and Σ -Pr oto c ols as the First Step . D.1 Recommended o verall length A reasonable recommendation is the following. • Minim um v ersion: 8 lectures of 90 minutes, if the app endices are assigned for self-study . • Standard version: 10 lectures of 90 minutes, for a balanced introductory course. • Expanded version: 12 lectures of 90 minutes, if one wan ts to sp end time on exercises, examples, and the logical asp ects of top oi. Among these, the 10-lecture plan is probably the b est default choice. It is long enough to mak e the theory intelligible, but short enough to b e used as a fo cused reading course or graduate seminar. D.2 A standard 10-lecture plan W e no w describ e a concrete 10-lecture plan. Lecture 1. F rom torsors to lo cal-to-global thinking. Review the basic idea of a torsor: free transitiv e action, lack of a preferred origin, transp ort b etw een p oints, and the role of lo cal triviality . Explain again why torsors naturally suggest co cycle data and gluing. The goal of the lecture is to mak e clear that the passage to shea v es and top oi is not a change of sub ject, but a contin uation of the same lo cal-to-global philosophy . Lecture 2. Sites, Grothendieck top ologies, and sheav es. Introduce categories as generalized indexing worlds, then define siev es and Grothendieck top ologies. Giv e concrete examples: op en co vers on a top ological space and simple small-site examples. Then explain presheav es and sheav es, emphasizing gluing and lo cality . The p edagogical goal is to make the sheaf condition feel natural rather than formal. 25 Lecture 3. T orsors on a site, co cycles, and descent. Return to torsors, now in the language of sheav es of groups. Explain lo cal triviality on a site, Cech co cycles, and descen t data. This lecture is esp ecially imp ortan t b ecause it ties the earlier torsor note to the present text and makes precise the sense in which torsors already live naturally in a sheaf-theoretic world. Lecture 4. Sheaf categories and Grothendiec k top oi. Explain wh y categories of sheav es are not merely con v enient containers, but genuine mathematical universes. In tro duce the slogan that a Grothendiec k top os is a generalized space view ed through its sheav es. Compare Set with Sh ( X ) and explain why the latter b eha v es like a v ariable version of the former. Lecture 5. Elementary top oi. In tro duce finite limits, exp onentials, and sub ob ject classifiers. Presen t the definition of an elemen tary top os and explain why this definition abstracts the crucial structural features of categories of sheav es. It is helpful here to compare the notions with familiar constructions in Set . Lecture 6. Sub ob ject classifiers and in ternal logic. Develop the idea that truth v alues in a top os may v ary lo cally . Explain the meaning of Ω , characteristic morphisms, and the interpretation of predicates. Introduce the basic in tuition of in ternal logic and its lo cal c haracter. If time p ermits, illustrate the discussion with a sheaf example where lo cal truth and global truth differ. Lecture 7. Intuitionistic logic and lo cal reasoning. Build on the previous lecture and explain wh y the internal logic of a top os is generally intuitionistic. Discuss conjunction, implication, and quan tifiers in an informal but mathematically resp onsible w a y . This is the place where students b egin to see that logical structure is not external decoration, but part of the geometry of a top os itself. Lecture 8. Category-theoretic to ols in practice. Use selected material from the app endix: Y oneda lemma, limits and colimits, equalizers and co equalizers, pullbacks and pushouts, and the meaning of cartesian closedness. W ork through at least one exercise in class. The goal is not to turn the course into a general category theory course, but to provide enough op erational fluency for the earlier lectures to b ecome usable. Lecture 9. T orsors and lo cal symmetry inside a top os. Explain what it means to sp eak ab out group ob jects, torsors, and lo cal symmetry internally to a top os. This lecture should make clear that torsors are not left b ehind when one en ters top os theory; rather, they b ecome more conceptually transparent. Use this to revisit descent and lo cal triviality from the internal p oint of view. Lecture 10. T o ward cryptographic securit y and Σ -proto cols. Finally explain the conceptual bridge to the cryptographic pap er. Discuss wh y lo cal consistency , simulabilit y , witness data, and gluing are naturally illuminated by sheaf-theoretic and top os-theoretic reasoning. The aim is not y et to repro duce the full technical argumen ts of the cryptographic pap er, but to leav e students in a p osition where that pap er b ecomes meaningfully readable. 26 D.3 A p ossible 12-lecture expansion If tw o additional lectures are a v ailable, a useful expansion is the following. • Add one lecture entirely devoted to exercises on limits, colimits, Y oneda, and internal logic. • A dd one lecture on reading selected passages of the cryptographic pap er together with the presen t note, so that the preparatory nature of this introduction b ecomes fully explicit. This 12-lecture version is esp ecially appropriate for a graduate seminar or sup ervised reading course. D.4 P edagogical remarks The most difficult conceptual jump for b eginners usually o ccurs at tw o p oin ts. First, they may understand sheav es as a technical gadget without seeing why they lead naturally to top oi. Second, they may understand the formal definition of a top os without seeing why in ternal logic matters. F or that reason, one should rep eatedly return to a small n um b er of central themes: • lo cal data and gluing, • absence of preferred global c h oices, • descent and compatibilit y , • lo cal truth versus global truth, • the role of these ideas in the structural understanding of cryptographic securit y . In short, this material is probably to o ric h for a very short mini-course, but it is w ell suited to a 10–12 lecture course of 90 min utes eac h . Such a course would give students not only a first acquain tance with sheav es and top oi, but also a serious conceptual preparation for the later study of Σ -proto cols. 27