走向几何

The grammar of hindsight

Nearly thirty years later, Grothendieck retrospectively associated 1958 with the emergence of his new geometry and named schemes and topoi as its central instruments. The passage is evidence of his later self-periodization, not a contemporaneous 1958 announcement.11Grothendieck, Récoltes et Semailles (Université des Sciences et Techniques du Languedoc and Centre National de la Recherche Scientifique, 1986), P23; Jackson, “As If Summoned from the Void, I,” 1047–48.

That difference governs this chapter. By the end of 1958, Grothendieck had recast Riemann–Roch, delivered a plenary lecture on the cohomology of abstract algebraic varieties, visited Harvard, and begun the writing that became Éléments de géométrie algébrique. IHÉS had been founded, and discussions about his future there were under way. These are contemporary or closely dated thresholds.

Mature topos theory was not one of them. Sites and topoi emerged through later work on generalized sheaf theory and étale cohomology. The word topos and the full conceptual architecture belong to the 1960s.22Jackson, “As If Summoned from the Void, I,” 1047–48.

Schemes were closer to immediate construction, but here too the finished EGA foundation should not be moved backward. The writing began in autumn 1958; most publication and systematic development followed. The year can be called a birth if birth means the convergence of a program, not the completion of its tools.

Hindsight is not disqualified by compression. Grothendieck knew which ideas had unfolded from the year. His testimony identifies a subjective turning point inaccessible to publication lists. The historian’s task is to place a second sentence beside his: in 1958 the vision was generative because so much remained unmade.

A death before the year began

Hanka Grothendieck died in 1957. The sources consulted for this account do not support a narrative of Grothendieck’s private response, household, or mental state. Those claims are withheld. The 1958 account therefore resumes with public documents rather than a reconstructed passage from bereavement to mathematics.33Jackson, “As If Summoned from the Void, I,” 1047.

Edinburgh, August 1958

In August 1958, Grothendieck delivered a plenary lecture at the International Congress of Mathematicians in Edinburgh. The paper, “The Cohomology Theory of Abstract Algebraic Varieties,” appeared in the proceedings published in 1960, pages 103–118.44Jackson, “As If Summoned from the Void, I,” 1047–48 and bibliography (Edin); Grothendieck, “The Cohomology Theory of Abstract Algebraic Varieties”.

The title is a program. “Cohomology theory” denotes a graded, functorial system with products, cycle classes, duality, and compatibility. “Abstract algebraic varieties” signals a setting beyond the complex manifolds on which classical topology operated.

The lecture should not be summarized as the presentation of étale cohomology. That theory did not yet exist in mature form. Jackson later interpreted the talk as anticipating themes Grothendieck pursued over the next dozen years. The contemporaneous fact is that he publicly framed a broad cohomological research program.

An ICM plenary lecture differs from a technical seminar. It addresses an international mathematical audience and selects problems capable of orienting a field. Grothendieck had entered the 1950s known for functional analysis. At Edinburgh he spoke as a builder of algebraic geometry’s future cohomology.

The public shift did not erase earlier subjects. His 1958 second edition of the São Paulo course on topological vector spaces appeared in the same year. The bibliography moved on several clocks: an old course expanded, recent Riemann–Roch papers entered print, and a future program was announced.

What the Weil conjectures asked from a language

Let $X$ be a nonsingular projective variety over a finite field $\mathbb F_q$. For every $n\geq1$, count its points over $\mathbb F_{q^n}$. Package the counts into $$Z(X,t) =\exp\!\left(\sum_{n\geq1} \#X(\mathbb F_{q^n})\frac{t^n}{n}\right).$$ Weil conjectured that this zeta function is rational, satisfies a functional equation, and has zeros and poles constrained by analogues of the Riemann hypothesis. Its factorization should contain numbers analogous to the Betti numbers of a complex variety.55Jackson, “As If Summoned from the Void, I,” 1047–48.

For a complex projective variety, topology provides cohomology groups. A continuous self-map acts on them, and a Lefschetz trace formula relates fixed points to alternating traces. Over a finite field, the Frobenius endomorphism fixes exactly the rational points over suitable extensions. If an adequate cohomology theory existed, a trace formula for Frobenius could convert point counts into linear algebra.

This heuristic asks for more than groups. The theory should be finite-dimensional so traces and characteristic polynomials make sense. It should be contravariantly functorial. It should have cup products and a Künneth formula. Smooth proper varieties should satisfy Poincaré duality. Algebraic cycles should produce cohomology classes. Frobenius eigenvalues should enter the zeta function.

No single 1958 construction supplied all of this. Zariski sheaf cohomology was powerful for coherent sheaves but did not yield the desired Weil cohomology. Classical topology worked after embedding into the complex numbers, unavailable in characteristic $p$. The problem was to invent both coefficients and a notion of local covering fine enough to detect arithmetic information.

Tôhoku had solved one layer: given a suitable abelian category of sheaves and a global-sections functor, derived cohomology could be constructed. The missing geometric layer was the choice of sheaves and site capable of producing the right invariants. Homological algebra made the problem exact without answering it.

The distinction between formal desiderata and construction is central. One can list axioms a Weil cohomology should satisfy and derive consequences. Proving that a theory satisfying them exists is another matter. In the following years, Grothendieck and collaborators developed scheme- and site-theoretic settings in which the required sheaves could be defined, while the Weil-cohomology program clarified the properties such a theory had to satisfy.66Weibel, “History of Homological Algebra”.

A checklist is not a construction

A candidate Weil cohomology assigns to every smooth projective variety $X$ graded finite-dimensional vector spaces $H^i(X)$ over a coefficient field. A morphism $X\to Y$ induces maps in the opposite direction. Products of varieties should satisfy a Künneth formula, and cup product should give a graded ring structure.77Grothendieck, “The Cohomology Theory of Abstract Algebraic Varieties”.

If $X$ has dimension $d$, Poincaré duality should pair degree $i$ with degree $2d-i$. Algebraic cycles of codimension $r$ should have classes in degree $2r$. A suitable trace map should send the top-degree class to the coefficient field. These operations must commute with morphisms and products.

For a smooth projective curve, the checklist predicts three nonzero degrees. Degree zero records connected components, degree two contains the fundamental class, and degree one has dimension twice the genus in the classical complex case. Frobenius acting on degree one supplies the nontrivial factor of the zeta function.

This is a true anchor because complex curves already possess singular cohomology with these properties. The problem over a finite field is not to imagine the dimensions; it is to construct groups and an endomorphism with the same formal relations when no complex topology is present.

The checklist has a boundary. Coherent-sheaf cohomology $H^i(X,\mathcal O_X)$ is finite-dimensional for proper varieties under standard hypotheses, but it does not provide all Betti degrees or the required finite-field trace formula. Chow groups receive cycle classes directly but are not generally finite-dimensional vector spaces with the needed duality. Neither theory satisfies the whole list.

In characteristic zero, later algebraic de Rham cohomology supplies another candidate and comparison theorems connect it to Betti cohomology over the complex numbers. In finite characteristic, later $\ell$-adic étale cohomology supplies the principal Weil theory. These later examples show that the axioms admit different realizations and that comparison is a theorem, not a definition.

The axiomatic method therefore performs two jobs. It derives consequences common to every adequate theory, and it reveals exactly what a proposed construction lacks. It cannot produce the sheaves, prove finiteness, or establish a trace formula by itself.

Historically, this distinction prevents the Edinburgh lecture from being mistaken for the solution of the Weil conjectures. The lecture could state a program and organize its requirements while the principal construction remained unavailable. That was already consequential because it turned an analogy with topology into a list of verifiable mathematical obligations.

The same distinction links Edinburgh to Tôhoku. Tôhoku supplied general derived-functor machinery once an abelian category of sheaves existed. The Weil checklist specified what the resulting cohomology had to do. Scheme and site theory would supply the geometric category between them.

For the biography, the checklist shows why Grothendieck moved from isolated theorems to foundations. A missing cohomology theory could not be obtained by one calculation. It required coordinated changes to objects, covers, coefficients, and functors. The public program of 1958 named that scale without completing it.88Weibel, “History of Homological Algebra”; Jackson, “As If Summoned from the Void, I,” 1047–50.

The trace-formula dream

The point-counting series has a formal logarithm. If a linear operator $F$ acts on a finite-dimensional vector space $V$, then $$-\log\det(1-tF) =\sum_{n\geq1}\operatorname{Tr}(F^n)\frac{t^n}{n}.$$ Compare this with the definition of $Z(X,t)$. If point counts were alternating traces of powers of Frobenius on graded cohomology, $$\#X(\mathbb F_{q^n}) =\sum_i(-1)^i\operatorname{Tr} \left(F^n\mid H^i(X)\right),$$ then formally $$Z(X,t) =\prod_i \det(1-tF\mid H^i(X))^{(-1)^{i+1}}.$$99Grothendieck, “The Cohomology Theory of Abstract Algebraic Varieties,” 103–18; Jackson, “As If Summoned from the Void, I,” 1047–48.

Rationality would follow because each determinant is a polynomial. The functional equation would follow from a duality pairing relating degree $i$ to degree $2d-i$. The degree of the $i$-th polynomial would be the $i$-th Betti number. The analogue of the Riemann hypothesis would become a statement about the absolute values of Frobenius eigenvalues.

This chain turns several assertions about a generating function into consequences of finite-dimensionality, trace, duality, and eigenvalues. The conjectures specify both a proof strategy and the shape of the missing theory.

The word dream should not conceal rigor. Each arrow in the chain requires a theorem specifying the object on which Frobenius acts, its coefficient field, finite-dimensionality, a fixed-point formula, and constraints on eigenvalue sizes. A formal determinant identity supplies a target, not the construction.

The eventual solution separated these questions. Grothendieck and collaborators built étale cohomology and the trace formalism; Grothendieck proved rationality and the functional equation in this framework; Deligne later proved the deepest eigenvalue estimate. Those achievements lie after 1958 and involved a large community. The Edinburgh moment belongs to formulation and program.

The trace expression also demonstrates why ordinary coherent-sheaf cohomology was not the answer. Groups $H^i(X,\mathcal O_X)$ are important but do not provide the required Betti numbers or coefficient flexibility for the zeta function. The desired cohomology had to resemble topological cohomology while remaining algebraic.

The program therefore linked foundation to a famous problem without reducing foundation to problem solving. Even if the Weil conjectures had not existed, schemes and generalized sheaf theory would have transformed geometry. The conjectures supplied an exact test of whether the language could produce cohomology on which Frobenius behaved as topology predicted.

Coefficients that do not share the characteristic

The eventual étale solution used a prime $\ell$ different from the characteristic $p$ of the base field, finite coefficient sheaves, inverse systems, and the $\ell$-adic field $\mathbb Q_\ell$.1010Weibel, “History of Homological Algebra”.

The condition $\ell\neq p$ is structural. Torsion of order prime to the characteristic behaves like discrete topological covering data under étale maps. $p$-torsion has different and subtler behavior. A single coefficient theory does not treat every prime uniformly.

The resulting groups $$H^i_{\mathrm{\acute et}} (X_{\overline{\mathbb F}_q},\mathbb Q_\ell)$$ are finite-dimensional $\mathbb Q_\ell$-vector spaces in the relevant proper smooth setting and carry an action of Frobenius. Their alternating characteristic polynomials give the zeta function. Poincaré duality and cycle-class maps supply other parts of the Weil package.1111Weibel, “History of Homological Algebra,” 825–26.

This mature account must remain in the future tense relative to Edinburgh. The construction and proofs were collective achievements of the following years, involving Grothendieck, Artin, Verdier, and many others; Deligne later proved the eigenvalue estimate. The 1958 lecture announced the need, not this finished notation.

The later coefficients show how exact the missing problem was. A new topology was insufficient without compatible finite coefficient systems, inverse limits, continuous group actions, finiteness theorems, and independence questions as $\ell$ varied.1212Weibel, “History of Homological Algebra,” 825–26.

They also show the importance of Tôhoku’s abstract setting. Sheaves of abelian groups on a site form an abelian category. Derived functors produce cohomology. Inverse limits introduce further homological issues. The formal language written in Kansas became infrastructure for arithmetic geometry.

The choice of $\ell$ creates an apparent noncanonicity: the geometric zeta function has rational coefficients, while its cohomological factorization is computed over $\mathbb Q_\ell$. Proving that traces and characteristic polynomials have the expected independence is part of the theory. A successful auxiliary coefficient should reveal an invariant not dependent on arbitrary choice.

This pattern resembles other Grothendieck constructions. Enlarge the setting to make the object exist; use a coefficient or resolution convenient for construction; prove that the final invariant is canonical. The move from arbitrary scaffolding to natural output is homological and geometric at once.

The later solution therefore confirms the prescience of the 1958 program without making the solution contemporaneous. Edinburgh supplied the question’s architecture. The next decade supplied the objects.

Why the Zariski topology was not enough

On an irreducible algebraic variety, every two nonempty Zariski-open sets intersect. The topology is effective for regular functions and coherent sheaves but too coarse to trivialize many finite coverings.

Classical covering-space theory builds fundamental groups and local systems from coverings that are locally homeomorphisms. In algebraic geometry over a finite field, an analogous finite unramified covering should be visible to cohomology. The Zariski topology often sees no such local trivialization.

One possible response is to add more open subsets to the same point set. Grothendieck’s later response was more categorical: change what counts as a cover. Objects over $X$ can serve as neighborhoods, and families of étale maps can cover. A sheaf is then a functor on this category satisfying descent.

That idea developed after the threshold of this chapter. Its logic can be seen from Tôhoku. Once cohomology is the derived functor of sections in an abelian category of sheaves, changing the site changes the cohomology while preserving the homological machinery. The formal and geometric layers separate cleanly.

The separation was a major strategic gain. Grothendieck did not need a wholly new theory of resolutions for every topology. He needed to prove that sheaves on the new site formed a suitable category with enough injectives, then identify the geometric meaning of the resulting groups.

It also explains the later topos, a category behaving in specified ways like a category of sheaves on a space or site. The construction was incomplete in 1958; the public problem concerned the inadequacy of ordinary opens for the desired local language.

Saying “the Zariski topology was too coarse” is only the start. It was exactly right for many algebraic questions. Grothendieck did not replace it universally. He created a framework in which different topologies could coexist, each adapted to a class of local phenomena.

Varieties and their limits

Algebraic geometry before schemes possessed rigorous foundations. Oscar Zariski had built a local algebraic approach through valuations, ideals, and the Zariski topology. André Weil’s 1946 Foundations of Algebraic Geometry treated abstract varieties over arbitrary fields. Serre’s FAC and GAGA brought sheaves and cohomology into the field.1313Jackson, “As If Summoned from the Void, I,” 1048, on Weil’s and Zariski’s foundations; Serre, “Faisceaux algébriques cohérents”; Serre, “Géométrie algébrique et géométrie analytique”; Jackson, “As If Summoned from the Void, I,” 1048.

The existing frameworks nevertheless made some operations cumbersome. Products, specialization, reduction modulo primes, families over a base, nilpotent infinitesimal structure, and points over varying field extensions all pressed against a geometry conceived primarily as varieties over a field.

A variety’s visible points do not record every algebraic feature. The rings $$k[x]/(x) \quad\text{and}\quad k[x]/(x^2)$$ have the same ordinary zero set, one point, but the second contains a nilpotent infinitesimal thickening. If geometry identifies them, tangent and deformation information is lost.

Arithmetic supplies another pressure. The equation defining a variety may have coefficients in the integers. One wants to study its fibers over different primes and over the generic rational point in one family. A framework requiring a fixed ground field makes this relation external rather than intrinsic.

The word “scheme” would name Grothendieck’s reorganization of a concept with predecessors in algebra, number theory, and geometry. EGA made schemes the basic language of his relative, functorial theory; this role should not be reduced to a lone coinage.

In 1958 that reorganization was beginning. The most accurate transition is not “varieties were wrong.”” Varieties became a distinguished class inside a larger category. The enlargement retained their successful geometry while adding arithmetic bases, generic points, and infinitesimal structure.

Affine schemes

Given a commutative ring $A$, its spectrum $$\operatorname{Spec}A$$ is the set of prime ideals of $A$. For an ideal $I$, the set $$V(I)=\{\mathfrak p: I\subseteq\mathfrak p\}$$ is closed in the Zariski topology. Distinguished open sets $$D(f)=\{\mathfrak p:f\notin\mathfrak p\}$$ form a basis.

Topology alone is not enough. A structure sheaf assigns to $D(f)$ the localization $A_f$, with compatible restrictions. The pair $$(\operatorname{Spec}A,\mathcal O_{\operatorname{Spec}A})$$ is an affine scheme. General schemes are ringed spaces locally isomorphic to affine schemes.

Prime ideals include more than closed points. In an integral domain, the zero ideal is a generic point whose closure is the whole space. It represents the generic behavior of the variety. Closed points represent specializations. A single topological space records both general and special positions.

Nilpotents live in the structure sheaf even when they do not alter the prime spectrum as a set. Thus $\operatorname{Spec}k[x]/(x^2)$ remembers more than its one point. Infinitesimal neighborhoods become geometric objects rather than auxiliary algebra.

A ring homomorphism $A\to B$ induces a morphism $$\operatorname{Spec}B\to\operatorname{Spec}A$$ in the reverse direction. Algebra and geometry form a contravariant relation. Localization becomes restriction to an open set. Tensor product of rings becomes fiber product of affine schemes.

These definitions are presented here in mature modern form to explain the emerging language; they are not a claim that every axiom and convention had been fixed in the Edinburgh lecture. The precision window belongs to the threshold the chapter approaches. Much of its systematic exposition would appear in EGA after 1958.

The gain is relative geometry. A scheme $X$ over a base $S$ is a morphism $X\to S$. fibers over points of $S$ are obtained by fiber product. Changing the base is another fiber product. Properties such as proper, smooth, or flat can be defined for morphisms and tested under base change.

Riemann–Roch had already shown the power of a theorem about a proper map. Scheme language would make maps over changing bases the default setting rather than an extension of the absolute case.

The spectrum of the integers

The prime ideals of $\operatorname{Spec}\mathbb Z$ are $(0)$ and $(p)$ for each prime number $p$. The point $(0)$ is generic; its residue field is $\mathbb Q$. Each closed point $(p)$ has residue field $\mathbb F_p$.

A scheme of finite type over $\operatorname{Spec}\mathbb Z$ is an arithmetic family. Its generic fiber is a variety over $\mathbb Q$. Its fiber at $(p)$ is a variety over $\mathbb F_p$. Reduction modulo different primes is no longer a collection of external operations; it is taking fibers of one morphism.

This geometry makes specialization visible. The generic point specializes to every closed prime. A property true generically may fail at finitely many or infinitely many special fibers; the locus where it holds can itself be studied. Arithmetic becomes relative geometry over a one-dimensional base.

The example also revises “dimension.” The base $\operatorname{Spec}\mathbb Z$ has dimension one because chains of prime ideals have one strict step. A family of relative dimension $d$ over it has total dimension $d+1$. Geometric and arithmetic directions coexist.

Nothing in the set of ordinary complex points of an equation records these fibers. Scheme language does. It gives a common object in which characteristic zero and positive characteristic are related by base change.

This relation was essential to the Weil program. A variety in characteristic zero can have models over finitely generated rings and reductions modulo primes. Cohomological invariants should vary compatibly. Comparison theorems between different cohomologies would become a later central theme.

The spectrum also includes primes that are not maximal in more general rings. A prime ideal corresponds to an irreducible closed subset through its generic point. Thus schemes turn subvarieties into points at another level of granularity. Localization at that point studies the generic behavior along the subvariety.

This is one reason the scheme is a metamorphosis rather than a larger point set. It changes what a point can represent. Closed geometric points remain, but generic and arithmetic points join them, and the structure sheaf gives each a local ring.

In 1958 the full expository consequences were still being organized. Presenting $\operatorname{Spec}\mathbb Z$ here explains what the EGA writing would normalize, not what every listener in Edinburgh had already accepted.

Nilpotents and infinitesimal questions

The dual numbers $k[\varepsilon]/(\varepsilon^2)$ have the same underlying prime spectrum as $\operatorname{Spec}k$, but a different structure sheaf; $\varepsilon$ is nonzero and squares to zero.

A map $$\operatorname{Spec}k[\varepsilon]/(\varepsilon^2)\to X$$ extending a $k$-point of $X$ represents a first-order tangent direction. Thus tangent vectors can be described as points with values in a nonreduced test scheme. Nilpotents encode infinitesimal variation.

Classical varieties often imposed reducedness, erasing nilpotents. For intersection theory and deformation, this can lose multiplicity. Two subvarieties meeting tangentially may have the same set-theoretic intersection as a transverse meeting but a different scheme-theoretic intersection. The quotient ideal retains the difference.

Families can also acquire nonreduced fibers even when the total object is defined naturally. A foundational category closed under fiber products should not throw such fibers away. Schemes accept them as ordinary members.

This acceptance changes the relation between equations and points. The ideals $(x)$ and $(x^2)$ vanish on the same point set over an algebraically closed field but define different closed subschemes. Algebraic geometry records multiplicity as well as common zeros.

The move has a philosophical resonance: visible points are insufficient. Yet it remains a precise mathematical claim, not a metaphor about hidden human life. The structure sheaf contains algebra that the topology does not.

Grothendieck’s later relative geometry would make infinitesimal lifting criteria central. Smoothness, unramifiedness, and étaleness can be characterized through lifting maps from nilpotent thickenings under suitable finite-presentation hypotheses. This mature formulation belongs after 1958, but nilpotents explain why the enlarged category was necessary.

The same test objects enter deformation theory. A first-order family lives over the dual numbers; obstructions to extending it live in higher algebraic structures. Ext groups from the homological period become deformation spaces. The convergence of schemes and homological algebra was therefore substantive, not terminological.

Points with values

For a scheme $X$, its ordinary set of points is only one shadow. Each test scheme $T$ gives a set $$X(T)=\operatorname{Hom}(T,X)$$ of $T$-valued points. As $T$ varies, these sets form a contravariant functor. The functor of points determines $X$ up to unique isomorphism through the Yoneda principle.

When $T=\operatorname{Spec}K$ for a field $K$, $X(T)$ is the set of $K$-valued points. When $T$ contains nilpotents, a $T$-point probes infinitesimal directions. When $T$ varies over extensions or local rings, arithmetic and deformation data appear in one functor.

This viewpoint fits Grothendieck’s language of functors. A family of geometric objects over $T$ should pull back along every $T'\to T$. Such families form a functor on bases.1414Eilenberg and Mac Lane, “General Theory of Natural Equivalences”.

The mature representability program belongs to the following parts of the biography. In 1958 it was a direction becoming possible. A scheme makes base change and families native operations while adding generic and infinitesimal points.

Functoriality also changes equality. Two schemes can be shown isomorphic by naturally identifying all their $T$-points. A construction can be defined through the functor it represents even before explicit equations are found. This is the geometric counterpart of characterizing tensor products or adjoints by universal mapping properties.

The analogy has limits. A functor need not be representable. Automorphisms can require a stack rather than a scheme. Infinite-dimensional or badly separated moduli can resist geometric realization. Later theory would enlarge the targets. At the 1958 threshold, representability is a question-generating method, not a universal guarantee.

What schemes solve—and what they do not

Schemes join algebraic varieties and arithmetic spectra in one category, retain nilpotents, make fiber products reliable, and turn families over a base into morphisms. They do not by themselves provide the Weil cohomology.1515Jackson, “As If Summoned from the Void, I,” 1047–50.

The distinction is easy to lose because schemes and étale cohomology entered the same later program. One can define $\operatorname{Spec}A$ and its Zariski sheaves without knowing how to construct finite-dimensional cohomology groups carrying Frobenius. The category of schemes supplies the geometry on which a new topology and new coefficients can be built.

Nor does scheme language automatically make every moduli functor representable. A functor may fail the sheaf condition, have nontrivial automorphisms, or require infinite-dimensional parameters. Representability theorems need hypotheses. Grothendieck’s later FGA work would develop criteria and examples; stacks would handle some failures.

Schemes also do not eliminate singularities. They describe singular and nonreduced objects more faithfully. Faithful description can make difficulties sharper: local rings need not be regular, coherent sheaves may lack finite locally free resolutions, and intersection products require corrections.

The enlargement does not choose a cohomology coefficient field. Betti cohomology over $\mathbb C$, algebraic de Rham cohomology in characteristic zero, and later $\ell$-adic étale cohomology have different constructions and comparison theorems. The scheme is the shared geometric input, not a claim that all outputs are identical.

This limit is part of the strength. By separating object, topology, coefficients, and derived machinery, Grothendieck could vary each layer deliberately. A scheme can carry Zariski, étale, flat, or other sites. A sheaf can carry torsion or $\ell$-adic coefficients. Homological algebra can derive sections in each case.

The layered architecture also makes comparison a theorem. If two cohomologies act on the same schemes, one can ask for natural comparison maps, cycle-class compatibility, and agreement of characteristic polynomials. The common source category turns resemblance into a functorial question.

In 1958, these layers were not all available. The scheme foundation was beginning; the cohomological demand was public; Tôhoku supplied derived machinery. The gaps identified the work ahead. Calling the year a birth is credible precisely because the parts had begun to align without being collapsed.

This separation guards against technological history in which one invention solves every problem. Grothendieck’s later achievement was an ecosystem of concepts built with many collaborators. Schemes were the base language, not a universal answer.

Fiber products and the discipline of base change

Given maps $X\to S$ and $S'\to S$, the fiber product $$X_{S'}=X\times_S S'$$ is the result of changing the base from $S$ to $S'$. It is characterized by a universal property: maps into $X_{S'}$ correspond to compatible maps into $X$ and $S'$ over $S$.

For affine schemes, if $X=\operatorname{Spec}B$, $S=\operatorname{Spec}A$, and $S'=\operatorname{Spec}A'$, then $$X\times_S S' =\operatorname{Spec}(B\otimes_A A').$$ The tensor product that changes scalars in algebra becomes a geometric pullback.

Fibers are special cases. A point $s\in S$ has residue field $\kappa(s)$, and $$X_s=X\times_S\operatorname{Spec}\kappa(s).$$ One family contains all these fibers, while the morphism records how they vary.

Properties of morphisms can be tested for stability under base change. Proper maps remain proper. Smooth maps remain smooth. Flatness is itself designed to make tensoring preserve exact sequences, so flat base change avoids certain homological distortions. These stability statements became part of EGA’s systematic architecture.

Base change also has cohomological content. Given a proper $f:X\to S$ and $S'\to S$, compare the pullback of $R^if_\ast\mathcal F$ with the higher direct image on $X_{S'}$. A canonical base-change map exists; proving it is an isomorphism requires hypotheses. The natural map comes first, the theorem second.

This repeats the pattern of GAGA. Construct a comparison functorially, then find conditions under which it loses no information. The relative viewpoint produces families of such comparisons and makes their compatibility unavoidable.

Riemann–Roch also invites base change. A proper morphism has K-theoretic and cycle pushforwards. In a cartesian square one asks how pushforward and pullback interact. Transversality or flatness may remove excess intersection terms. Later intersection theory would encode the corrections.

The base is therefore not passive scenery. It controls coefficients, fibers, and specialization. Grothendieck’s “relative world” would treat an object without its structural map to a base as incomplete data.

In 1958 this discipline was emerging from several directions—fiber spaces, sheaves, proper maps, arithmetic reduction. EGA would make it the ordinary syntax of geometry.

One equation, several fibers

A simple family shows what the relative language contributes. Let the base be the affine line $$S=\operatorname{Spec}k[t],$$ and define $$X=\operatorname{Spec}k[t,x,y]/(xy-t).$$ The inclusion of $k[t]$ gives a morphism $X\to S$.

Over a point $t=a$, the fiber is defined by $$xy=a.$$ When $a\neq0$, this is a smooth hyperbola over a suitable field. At $a=0$, the equation becomes $$xy=0,$$ the union of the two coordinate axes meeting at the origin. One morphism contains smooth general fibers and a singular special fiber.

Without a base, these curves could be listed as separate varieties. The map to $S$ records their relation, and the coordinate $t$ records specialization. The generic fiber over $k(t)$ and the special fiber over $k$ become parts of one object.

Base change makes the construction portable. A map $S'\to S$ substitutes a new parameter for $t$, and the fiber product produces the corresponding family over $S'$. Passing to the point $t=a$ is one such base change. Passing to a field extension changes coefficients without rewriting the geometry.

The example also distinguishes properties of the total space from properties of the morphism and its fibers. The total ring is isomorphic to $k[x,y]$ after eliminating $t$, so the total space is an affine plane. The map nevertheless has a singular fiber at the origin. Smoothness of the total space does not imply smoothness of every fiber.

Flatness supplies another layer. Under the standard hypotheses, this family is flat over the base, so the singular fiber is a controlled degeneration rather than an arbitrary collapse of dimension. Flatness does not mean that all fibers are isomorphic or smooth. It preserves an algebraic form of continuous variation while allowing geometry to change.

This is a true modern scheme example, not a claim that Grothendieck used the equation in Edinburgh. Its historical role is explanatory: scheme language turns degeneration and specialization into properties of one morphism, while base change specifies how the family is tested.

The boundary is equally important. Not every map is flat, proper, or smooth. Each property has its own stability and local criteria. Relative geometry gains power by distinguishing them, not by treating every family as well behaved.

For the biography, the example identifies what EGA writing was beginning to normalize. Riemann–Roch had made a proper morphism the subject of a theorem. A scheme over a base now made families and special fibers ordinary objects of study. The passage from absolute varieties to relative geometry can be seen in one equation without being treated as a personal metaphor.1616Jackson, “As If Summoned from the Void, I,” 1047–50.

Topos: a later name in a memory of 1958

A topological space $X$ determines a category $\operatorname{Sh}(X)$ of sheaves. Grothendieck’s later work treated a category with the relevant sheaf-like properties as a generalized space, a topos.1717Marquis, “Category Theory”; Weibel, “History of Homological Algebra,” 825–26.

A site starts with a category and specified covering families satisfying axioms. A sheaf assigns data to objects of the site with compatibility and descent over covers. The resulting category has limits, colimits, exponentials, and a subobject classifier; it behaves in important ways like a universe of sets varying over a space.

This mature description synthesizes work from the early 1960s and after. It is not a 1958 definition.1818Jackson, “As If Summoned from the Void, I,” 1047–48.

What existed in 1958 was the pressure to alter space. Ordinary open sets were too restrictive for arithmetic coverings. Sheaf cohomology depended primarily on the category of sheaves and its global-sections functor. If another category of sheaves had the same formal properties, homological algebra could operate there.

The later topos makes that pressure conceptual. Space is no longer only a set of points with opens. It is a setting in which localization, gluing, and variable sets make sense. Points may be useful, sparse, or absent.

Schemes and topoi solve different problems. A scheme enlarges algebraic varieties through spectra and structure sheaves. A topos enlarges spaces through categories of sheaves. Étale cohomology uses both: schemes supply geometric objects and the étale site; the associated topos supplies sheaves and cohomology.

Grothendieck’s later statement gives topoi the deeper role. That retrospective judgment does not place the future SGA 4 in the Edinburgh lecture.

This is a model for using retrospective sources. Preserve the speaker’s own architecture of memory. Then annotate the development dates of the concepts named. The result neither dismisses memory nor lets memory overwrite chronology.

The year’s published mathematics

The Borel–Serre Riemann–Roch article and Grothendieck’s Chern-class appendix appeared in 1958. The Kansas fiber-space report and São Paulo topological-vector-space course received second editions. The Edinburgh paper appeared in 1960, while EGA writing began before any EGA volume appeared.1919Borel and Serre, “Le théorème de Riemann–Roch”; Grothendieck, “La théorie des classes de Chern”; Grothendieck, Espaces vectoriels topologiques; Grothendieck, A General Theory of Fibre Spaces with Structure Sheaf; Grothendieck, “The Cohomology Theory of Abstract Algebraic Varieties”.

This mixed bibliography resists the image of a clean disciplinary break. Functional analysis, fiber spaces, characteristic classes, and future foundations coexist. Some items are new research; others are revised exposition; one is a conference program awaiting publication.

The distinction among dates is especially important for the phrase “the 1958 work.” Grothendieck’s Riemann–Roch theorem had been presented and transmitted in 1957. Its article was received and printed in 1958. The Edinburgh lecture was delivered in 1958 and printed in 1960. The second editions repeated works first issued in 1954 and 1955. EGA began as writing in 1958 and became publication later.

An influence chronology is longer still. Schemes would become standard over years. Étale cohomology would be constructed collectively in the following decade. Topoi would transform geometry and logic. The 1966 Fields Medal would retrospectively group achievements that were only beginning in 1958.

Separating the clocks does not fragment the year. It shows what convergence meant. Earlier homological and sheaf work reached a theorem; the theorem displayed the power of morphisms; the Weil conjectures demanded a stronger cohomology; schemes offered a larger geometry; institutional support for sustained construction was forming.

Harvard in the autumn

In autumn 1958 Grothendieck made the first of several visits to Harvard. John Tate was on the faculty, and Oscar Zariski chaired the department. Mumford later connected the invitation with Grothendieck’s cohomological proof of a connectedness theorem and Zariski’s wish to expose students to the method.2020Jackson, “As If Summoned from the Void, I,” 1048; Jackson, “As If Summoned from the Void, I,” 1048–49, reporting David Mumford’s later recollection.

Zariski and Grothendieck represented different mathematical formations. Zariski’s geometry was deeply connected to local algebra and to the hard-won rigorization of algebraic varieties. Grothendieck’s emerging style pushed toward categories, sheaves, and relative foundations. The encounter should not be written as old against new. Zariski recognized power in methods different from his and created an audience for them.

Mumford remembered Grothendieck making an abstract argument for a concrete lemma that initially seemed unbelievable and then proved exactly right. He described a capacity to move to a level of abstraction an order of magnitude higher and still control the original question. The testimony is later and belongs to the Harvard visit; it should not be converted into a specific blackboard exchange unless the quoted recollection is used with its source.2121Jackson, “As If Summoned from the Void, I,” 1048–49.

The visit enlarged his American mathematical audience. It also placed him near Zariski’s students at a moment when the foundations of algebraic geometry were about to change. The later prominence of Mumford and others can make the meeting look like the origin of a school. In autumn 1958 it was a visit, not yet the institutional seminar machine of IHÉS.

Zariski’s foundation and Grothendieck’s enlargement

Zariski had helped establish rigorous algebraic foundations through local rings, valuations, ideals, and the Zariski topology. Scheme theory extended that foundation rather than replacing an absence of rigor.2222Jackson, “As If Summoned from the Void, I,” 1048.

The local ring at a prime ideal is already a scheme-theoretic object. Commutative algebra supplied dimension, regularity, integral dependence, and normalization. Grothendieck’s enlargement made these local rings the stalks of one structural sheaf across $\operatorname{Spec}A$ and placed ring maps into a global category.

Weil’s abstract varieties supplied another predecessor. They allowed varieties to be treated without one fixed affine or projective embedding and over arbitrary fields. Serre’s sheaves supplied local-to-global algebra. Chevalley and others developed schemes or related algebraic spaces in forms contributing to the emerging terminology.

Grothendieck’s originality was architectural and extensive. He took these materials as basic language, insisted on arbitrary bases and functoriality, admitted nilpotents and generic points, and built a theory whose definitions were optimized for base change.

Zariski’s invitation to Harvard can therefore be read as a meeting between foundation builders, not between rigor and abstraction. Their styles differed. Zariski often drew strength from explicit local algebra and geometric examples; Grothendieck sought universal settings. The later scheme theory needed both.

Mumford’s recollection records Zariski’s willingness to expose students to cohomological methods he did not adopt in his own work.

The connectedness theorem mentioned in the recollection is an example of the new relation. A result proved through local algebra or birational geometry could be rederived through coherent cohomology and duality. The second proof did not make the first false. It revealed a broader mechanism and suggested extensions.

Harvard also exposed Grothendieck to a strong American algebraic-geometry audience before IHÉS began. The visit belonged to an international circulation continuing the paths through São Paulo and Kansas, now with the new geometry rather than functional analysis at its center.

An institute founded

IHÉS was created in 1958 by Léon Motchane, who took inspiration from Princeton’s Institute for Advanced Study and sought an independent center for theoretical mathematics and physics. The formal establishment occurred in June according to Jackson’s reconstruction; the official institutional history confirms the year and Motchane’s model.2323Institut des Hautes Études Scientifiques, “The Foundation”; Jackson, “As If Summoned from the Void, I,” 1049; Jackson, “As If Summoned from the Void, I,” 1049.

Accounts differ on the route by which Grothendieck entered the plan. Jackson reports discussions around the Edinburgh Congress, perhaps earlier; Cartier reports that Dieudonné made an offer to Grothendieck a condition of his own acceptance. The exact negotiation remains unresolved.2424Jackson, “As If Summoned from the Void, I,” 1049; Jackson, “As If Summoned from the Void, I,” 1049, summarizing differing accounts from David Aubin and Pierre Cartier.

IHÉS was independent of the French state and could appoint a stateless mathematician outside the public-employment barrier described in the biographical sources.2525Jackson, “As If Summoned from the Void, I,” 1049.

Jackson’s secondary account places a formal appointment transition for Grothendieck and Dieudonné in March 1959 and the beginning of Grothendieck’s algebraic-geometry seminar in May. The current IHÉS profile, by contrast, lists Grothendieck as “Permanent Professor at IHES from 1958 to 1970.” No personnel record has been opened here to reconcile recruitment, acceptance, contract signature, effective start, or retrospective service range. Part IV therefore ends at the 1958 institutional threshold, not at a verified employment date.2626Jackson, “As If Summoned from the Void, I,” 1049, reports a March 1959 formal appointment transition and a May seminar; Institut des Hautes Études Scientifiques, “Alexander Grothendieck, Permanent Professor from 1958 to 1970,” https://www.ihes.fr/en/professeur/alexander-grothendieck-1/.

The distinction guards against a teleological biography. Grothendieck’s Brazil and Kansas posts were not rehearsals designed to lead to IHÉS. The institute did not exist when he accepted them. In 1958, after years of temporary arrangements, a new structure appeared capable of supporting the scale of work he was beginning.

Independence, status, and scale

IHÉS was conceived outside the ordinary university and state-research hierarchy. Its model promised a small permanent faculty, long research horizons, and a flow of visitors. For Grothendieck, institutional independence had a specific practical consequence: his statelessness did not bar appointment in the same way it had complicated a permanent French state post.

The official history describes a broader institutional aim in mathematics and theoretical physics, supported by public and private contributors with Robert Oppenheimer on the scientific committee. It does not present the institute as founded for Grothendieck.

The fit concerned scale as well as legal status. A program to rebuild algebraic geometry could not be completed between routine short appointments. It required sustained time, collaborators, students, visitors, seminar continuity, and publication channels. IHÉS would eventually provide these.

In 1958 the institute had no established Bois-Marie campus or mature seminar culture. Those belong to later chronology. Its creation offered a promise rather than the familiar institution seen from hindsight.

Dieudonné’s involvement was equally important. He brought foundational aims, Bourbaki experience, and editorial power. In Cartier’s account, Dieudonné made an offer to Grothendieck a condition of his own acceptance. The exact negotiation remains unresolved, but the account establishes the condition it attributes to Dieudonné.

Institutional independence did not mean material independence from funders. The source of IHÉS funding would become a decisive ethical issue for Grothendieck in 1970. That later conflict must not be planted in 1958 as foreknown. At foundation, the independent structure solved one problem and created conditions whose consequences would appear years later.

The new position also altered belonging without changing nationality. Grothendieck could have an institutional home while remaining stateless. The distinction between legal citizenship and scientific affiliation is factual; it need not be converted into a metaphor for schemes over a base.

Part V will show how the institute’s seminar and publications amplified his work. Here the narrower evidence is that EGA writing was under way by autumn 1958; its relation to the unreconciled appointment clocks is not established.

The first EGA pages

Jackson, citing the published Grothendieck–Serre correspondence, dates the beginning of EGA writing to autumn 1958 and separately reports formal appointments in March 1959. The resulting interval is his secondary chronology, not a verified contract-signature or effective-start date.2727Jackson, “As If Summoned from the Void, I,” 1049–50; Jackson, “As If Summoned from the Void, I,” 1049, reports a March 1959 formal appointment transition and a May seminar; Institut des Hautes Études Scientifiques, “Alexander Grothendieck, Permanent Professor from 1958 to 1970”; Jackson, “As If Summoned from the Void, I,” 1049–50, citing the published Grothendieck–Serre correspondence.

The title Éléments de géométrie algébrique echoed Bourbaki’s Éléments de mathématique and, behind it, Euclid. It announced foundations rather than a sequence of research papers. The first pages did not yet guarantee the eventual 1,800-page published monument or the still larger plan left incomplete.

Writing began before the institutional seminar. This matters because it places the foundational decision in 1958 itself. It also keeps publication in the future. EGA I appeared later; subsequent chapters and revisions stretched across the 1960s.

EGA’s title pages identify Dieudonné as collaborator. The surviving publication assigns Grothendieck authorship with Dieudonné’s collaboration; a detailed division of drafting labor requires manuscript evidence beyond this account.2828Jackson, “As If Summoned from the Void, I,” 1049–50. Detailed drafting labor remains unresolved without manuscripts.

The initial program extended beyond scheme definitions to sheaves, morphisms, base change, quasi-coherent modules, separation, properness, smoothness, dimension, and cohomology.

The sequence of exposition became part of the mathematics. A definition introduced too narrowly could block a later theorem. A lemma proved without base-change compatibility might have to be rebuilt. EGA aimed to make the language reliable enough that future geometers could reason inside it.

That ambition joined Grothendieck to Bourbaki while exceeding the group’s established domains. He had begun participating in Bourbaki in the later 1950s. The collective’s structural method and insistence on foundations offered a model. Algebraic geometry, however, became an enterprise centered on Grothendieck and Dieudonné rather than a Bourbaki volume.

Redaction as mathematical construction

EGA’s redaction ordered definitions, propositions, and compatibility results across a growing foundation. This statement concerns the structure of the published work, not a reconstructed drafting scene.2929Jackson, “As If Summoned from the Void, I,” 1049–50.

Grothendieck’s drafts supplied the movement of ideas. Dieudonné expanded arguments, standardized notation, checked dependencies, and produced polished text. EGA’s title-page credit identifying Jean Dieudonné as collaborator should be read literally. Global vision and line-by-line construction were different contributions to one work.

Modularity was essential. A proposition needed a stable number and exact hypotheses so later chapters could invoke it. Definitions had to avoid hidden assumptions of algebraic closure or noetherianity when the program intended greater scope. Variants had to be stated at the correct level rather than left to analogy.

The work’s title “Elements” can mislead modern readers into expecting elementary exposition. It meant foundational components arranged deductively. The reader was expected to acquire a language capable of supporting research, not a gentle introduction through examples.

Bourbaki supplied one precedent for collective structural exposition. Grothendieck’s participation and the echo between the titles Éléments de mathématique and Éléments de géométrie algébrique are documented; claims about private ambition are omitted.3030Jackson, “As If Summoned from the Void, I,” 1049–50, on Bourbaki participation and EGA’s title.

This simultaneity created risk. A later insight could force revision of an earlier chapter. Plans expanded. The published sequence eventually stopped far short of the total program. The incompletion should not be read as simple failure; the completed volumes transformed the field, while omitted topics moved into SGA seminars and other works.

In autumn 1958 none of that extent was guaranteed. The first writing marks commitment. Grothendieck and Dieudonné began the labor required to make scheme language public, checkable, and teachable.

The contrast with the São Paulo course is illuminating. There, a mature analytic theory was organized for students and reissued. Here, exposition and invention advanced together. Both required pedagogical architecture; only the second was also constructing the field’s base objects.

The author’s later impatience with unmotivated tricks also finds a redactional counterpart. If a proof requires a device, a foundational text asks whether the device follows from a more general lemma. Repeated devices become definitions or propositions. Writing searches for the natural sequence after discovery has been irregular.

EGA would acquire authority partly because this search was relentless. Its sentences often feel impersonal, but the impersonality was produced by collaboration and revision. The personal history survives in drafts, letters, and the division of labor behind the finished theorem.

The year 1958 in sequence

The year can be arranged without pretending every item has an exact month. The Riemann–Roch and Chern-class manuscripts were received in May and June. IHÉS was formally established in June. Grothendieck spoke in Edinburgh in August. Harvard received him in the autumn, when EGA writing also began.3131Jackson, “As If Summoned from the Void, I,” 1047–50.

Some events resist tighter placement. Motchane’s discussions with Grothendieck and Dieudonné may have occurred at Edinburgh or before. The precise period of mathematical cessation is unclear. The development of scheme terminology across drafts requires archival work.

Publication receipt dates are exact. Congress month is exact. Institutional foundation year and reported month are strong. Witness recollections are attributed. Conceptual births remain hindsight.

Dense years invite false causation when a narrative transition makes a June event explain an August event. The evidence here supports convergence, not every causal arrow.

Paris held correspondence and publication. Edinburgh held the public lecture. Harvard held a visiting audience. IHÉS existed as a new French institution. EGA began through collaboration not yet anchored in the later campus routine.

Grothendieck’s statelessness makes the geography biographically salient, but the mathematics was also intrinsically international. Weil’s conjectures, Japanese publication of Tôhoku, American seminars, German Arbeitstagung, French journals, and Brazilian course editions formed one network.

By 1958 his name traveled more securely than his legal nationality. This observation is factual so long as it is not romanticized. Statelessness imposed real limits; international prestige did not erase them. IHÉS offered a professional home through an institutional exception.

A program larger than one author

The Edinburgh program required scheme foundations, new topologies, homological algebra, finiteness theorems, trace formulas, comparison results, and calculations. No single paper supplied them.3232Grothendieck, “The Cohomology Theory of Abstract Algebraic Varieties,” 103–18; Weibel, “History of Homological Algebra,” 825–26.

Grothendieck’s later seminars distributed this labor among named students and colleagues. Those later contributions are not folded anonymously into 1958.3333Jackson, “As If Summoned from the Void, I,” 1049–50.

The need for collaboration arose from mathematics as well as logistics. New theories are tested by people who approach them from different fields. Number theorists ask whether Frobenius traces give the right counts. Topologists test analogies with classical cohomology. Algebraists examine exactness and local rings. Geometers supply families and singular examples.

Serre had already performed this testing role in correspondence. Borel and Serre had turned the Riemann–Roch proof into an exposition. Dieudonné was beginning foundational redaction. The collective future did not appear from nowhere when SGA began; its forms were accumulating.

Architecture can create asymmetry between credit for setting definitions and credit for proving components. Grothendieck’s later disputes over attribution do not establish motives in 1958; they reinforce the need to preserve named collaborators.

The phrase Grothendieck’s program is therefore useful if it names origin and coherence, misleading if it implies solitary execution. Weil’s conjectures were Weil’s. Serre transmitted their cohomological force. Many mathematicians constructed the eventual tools. Deligne completed the deepest estimate.

This division of labor also explains why an institution mattered. IHÉS could host visitors and a sustained seminar; publication series could preserve thousands of pages; a stable position allowed problems to continue across years. The program’s intellectual scale and institutional scale matched.

At the end of Part IV, the collective has not yet assembled in its later form. What exists is a public invitation: a language, a set of exact demands, and an architect beginning a foundation with one crucial collaborator. The open future belongs to many hands.

The threshold ends when exception becomes program. Jackson reports a March 1959 formal appointment transition and a May seminar opening, while the current IHÉS profile begins Grothendieck’s permanent-professor range in 1958. Neither record by itself identifies recruitment, acceptance, contract signature, or effective start. Those unresolved institutional clocks belong to Part V and should not be consumed by this chapter’s climax.3434Jackson, “As If Summoned from the Void, I,” 1049, reports a March 1959 formal appointment transition and a May seminar; Institut des Hautes Études Scientifiques, “Alexander Grothendieck, Permanent Professor from 1958 to 1970”.

Not yet the relative world

By autumn 1958 several elements of the relative viewpoint were present: Riemann–Roch for a proper morphism, higher direct images on a target, base-change questions, and early scheme foundations.3535Borel and Serre, “Le théorème de Riemann–Roch,” secs. 5–7; Jackson, “As If Summoned from the Void, I,” 1047–50.

Yet the relative world of the next part—the systematic doctrine that an object should be studied over a base and under every base change—was not finished. The terminology, theorems, and vast EGA apparatus were still being written. This chapter ends at the decision to make that world foundational.

The boundary rests on dated records for the institution’s foundation, the Edinburgh lecture, the Harvard visit, and the beginning of EGA writing. Jackson’s reported March 1959 appointment transition and the profile’s 1958–1970 service range remain attributed, non-equivalent records; the seminar chronology belongs to the following part.3636Jackson, “As If Summoned from the Void, I,” 1049–50; Institut des Hautes Études Scientifiques, “The Foundation”; Institut des Hautes Études Scientifiques, “Alexander Grothendieck, Permanent Professor from 1958 to 1970”.

The chronology creates the transition without an invented epiphany. A public program, a visiting audience, a manuscript project, and an institution converge. No one event alone is “the” turn.

Grothendieck’s later statement names schemes and topoi with nearly thirty years of hindsight. The contemporary 1958 vocabulary was more provisional: abstract varieties, cohomology, sheaves, proper maps, fiber spaces, and unpublished foundations.3737Jackson, “As If Summoned from the Void, I,” 1047–48, reproducing the later R&S self-periodization.

The rising language had already changed the scale of a theorem. It was about to change the scale of geometry. Part IV began with a course in topological vector spaces whose second edition fixed an existing theory. It ends with the first writing of a foundation whose theory did not yet exist in finished form.

Functional analysis remained active, including through Grothendieck’s work. Across 1953–58, São Paulo preserved an analytic course, Kansas produced the report and Tôhoku manuscript, Riemann–Roch reorganized a classical theorem around morphisms, Edinburgh stated a cohomological program, and EGA writing began.

The title of this part, A Language Begins, should therefore be read literally and modestly. A language begins when its basic distinctions become unavoidable: object and morphism, pullback and pushforward, exact functor and derived correction, local section and global obstruction, variety and scheme, work date and publication date. Fluency would take years and many speakers.

Serre’s coherent sheaves, Cartan–Eilenberg’s homological algebra, Eilenberg–Mac Lane’s functors, Buchsbaum’s exact categories, Hirzebruch’s theorem, Chow’s cycles, Chern’s classes, Weil’s conjectures, and Zariski’s local algebra all entered the construction.3838Cartan and Eilenberg, Homological Algebra; Dieudonné, “A. Grothendieck’s Early Work (1950–1960)”.

The work moved through a hundred-page paper, seminars, correspondence, the Borel–Serre exposition, and collaboration with Dieudonné. Each form has a separate documentary record.3939Jackson, “As If Summoned from the Void, I”; Grothendieck and Serre, Correspondance Grothendieck–Serre; Grothendieck, “Sur les faisceaux algébriques et les faisceaux analytiques cohérents”.

At the end of 1958 none of the hard mathematical gates was closed. The Weil conjectures remained open. Étale cohomology remained to be built. EGA had barely begun. The inspected record does not establish which IHÉS employment clock, if any, had taken contractual or practical effect by year’s end. The incompleteness is the ending’s factual tension, not a defect requiring a triumphant forecast.

The boundary is also terminological. Contemporary records in this chapter speak of varieties, sheaves, cohomology, fiber spaces, and proper maps. Modern scheme, site, topos, and $\ell$-adic language appears only in marked explanatory passages. This separation allows later mathematics to clarify the 1958 problem without being assigned to the 1958 speaker. It also keeps Grothendieck’s retrospective vocabulary from replacing the words and constructions then available.

The same discipline applies to institutions. The foundation of IHÉS in 1958, the beginning of EGA writing that autumn, Jackson’s reported March 1959 appointment transition, and the profile’s 1958–1970 service range are related but non-equivalent records. Recruitment, acceptance, contract signature, effective start, and retrospective service range cannot be collapsed into one moment without a personnel record.

No month is treated here as a verified employment boundary. Part IV stops at the end of its 1958 narrative scope, not at a personnel finding.

Part IV ends with two well-supported dated facts: the Edinburgh lecture had been delivered, and EGA writing had begun. The IHÉS employment transition remains a source conflict: Jackson reports March 1959, the current profile lists 1958–1970, and no personnel record opened for this account reconciles the clocks.4040Jackson, “As If Summoned from the Void, I,” 1049–50; Institut des Hautes Études Scientifiques, “Alexander Grothendieck, Permanent Professor from 1958 to 1970”.