The Scheme

The polemic on page 193

The first page of Alexander Grothendieck’s May 1959 Bourbaki lecture begins with a familiar dictionary. An affine algebraic space over a field $k$ is determined by its affine algebra, and a map of spaces corresponds contravariantly to a homomorphism of algebras. Then the text turns. Commutative algebra had long produced statements more general than the classical geometric setting required. There was usually no need to restrict every ring to a finitely generated algebra over a field, and no reason, with or without a ground field, to exclude nilpotent elements. Geometers, the speaker charged, had refused these indications. Their insistence on reduced spaces had obstructed natural methods.11Grothendieck, “Géométrie formelle et géométrie algébrique,” 193.

The accusation sounds sweeping because it was meant to clear space. It should not be read as a neutral history of every algebraic geometer before 1959. Zariski’s local methods, Weil’s abstract varieties, Chevalley’s local-ring schemes, Nagata’s geometry over Dedekind domains, and Serre’s sheaf-theoretic work had already strained the classical category in different directions. Grothendieck’s own exposition depended on them. The polemic identified a remaining refusal: algebraic geometers often discarded nilpotents and privileged varieties over fields even when their algebra suggested a broader geometric object.22Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

On that page, the scheme enters without metaphysics. Let $A$ be a commutative ring. Take its prime ideals as points, give them the Zariski topology, and attach a sheaf of rings whose stalk at a prime $\mathfrak p$ is the localization $A_{\mathfrak p}$. The resulting ringed space is the prime spectrum of $A$. Schemes are obtained by gluing such spectra. This compact procedure gave geometry to arbitrary commutative rings and made local algebra part of the object rather than auxiliary data.33Grothendieck, “Géométrie formelle et géométrie algébrique,” 193–96; Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas,” I, §§1–2.

The definition has since become elementary in the technical sense: it appears near the beginning of standard courses. Historically, it was not elementary in the sense of obvious. Each component asked readers to revise an intuition. Points would include prime ideals that were not maximal. A one-point space might carry nontrivial infinitesimal structure. The same topological space could underlie different schemes. A number ring could become a geometric base. Maps would be understood through pullback of functions. Above all, the enlarged objects were not introduced as exceptions around familiar varieties. They were to be the ordinary objects of the new language.

What an equation forgets

Begin with the human problem. Consider the equations $x=0$ and $x^2=0$ over a field. As sets of solutions, both define the same point. If geometry records only that set, the exponent disappears. Yet the exponent matters. It records multiplicity. It affects how the equation meets another one. In a family, it can mark the collision of distinct points. In deformation theory, it participates in first-order information. The ring $k[x]/(x)$ differs from $k[x]/(x^2)$, even though their prime spectra have the same single underlying point.44Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

Classical algebraic geometry possessed ways to handle multiplicity, local intersection, and infinitesimal calculation. Scheme theory did not discover that exponents matter. Its change was to make the relevant ring—with its nilpotents—part of the geometric object from the start. One no longer had a reduced point plus a later annotation saying “double.” One had a nonreduced scheme whose structure sheaf retained the doubling.55Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

A second loss occurs when one takes only ordinary solutions. The polynomial ring $k[x]$, for $k$ algebraically closed, has maximal ideals $(x-a)$ corresponding to familiar points $a\in k$. But it also has the prime ideal $(0)$. This is the generic point of the affine line: its closure is the whole line. It is not an additional coordinate value hiding beyond all $a$. It is a point whose local information reflects the function field and whose specializations include every closed point.

The generic point turns an irreducible closed subset into something that can itself be the closure of one point. This makes irreducible components and specialization relations internal to the topology. Again, algebraic geometers had long reasoned generically. Scheme theory built genericity into the space. The gain is formal economy and compatibility, not the invention of generic reasoning.

A third loss occurs when geometry is tied to a single field. The ring $\mathbb Z$ is not an algebra over one preferred field. Its spectrum nevertheless has prime ideals $(p)$ for every prime number $p$, together with the generic prime $(0)$. An equation with integer coefficients can define a scheme over this base. Its fiber over $(p)$ records reduction modulo $p$; its generic fiber records characteristic zero. The object makes comparison possible without pretending that all fibers are alike.

These three losses—multiplicity, genericity, and arithmetic variation—explain much of the scheme’s initial force. The scheme keeps information routinely produced by algebra, admits new cases, and organizes changes of base. What seemed at first like greater abstraction was also a refusal to throw data away.

The human object

For a general reader, the faithful first statement is this: a scheme is a geometric object assembled from commutative rings in such a way that every point has a ring of local functions. The points tell only part of the story. The topology tells which points specialize to which others. The local rings tell how functions behave nearby, including infinitesimal and multiplicity information. Maps of schemes respect both space and local algebra.66Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

“A space with functions” is a useful phrase, provided one does not imagine that the space exists first and functions are painted onto it. For an affine scheme, the ring generates both the point set and the sheaf. Change the ring and one may change the topology, the local rings, or both. The relation is constitutive.77Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

A city-map analogy can help. A conventional map gives locations and neighborhoods; an augmented map might attach to each region its regulations, utilities, and available operations. Two maps with identical streets could differ in what can legally or physically be done there. The analogy illuminates why the same point set can carry different structure. It fails because a scheme’s local rings are not external metadata supplied after the topology. They arise coherently from the ring and determine algebraic possibilities with exact gluing laws.

Another analogy describes generic points. A folder representing an entire case is not one more page alongside the pages it contains. Similarly, a generic point is not one additional classical location. Its closure represents an irreducible region, and its specializations express how generic information can become special. This analogy fails if it makes the topology bureaucratic rather than mathematical; closure and localization are precise operations, not labels chosen for convenience.

The language is difficult partly because ordinary spatial intuition privileges closed points. In Hausdorff spaces, points are closed, and distinct points can be separated. Zariski spaces behave differently. Generic points are not closed; open sets are large; limits and specializations follow algebra. This coarseness is not a defect accidentally tolerated. The sheaf carries fine algebraic information that the topology alone does not.

Reading a spectrum

Take $A=k[x]$, with $k$ algebraically closed. The points of $\operatorname{Spec}A$ are $(0)$ and the maximal ideals $(x-a)$. A basic open set $D(f)$ consists of primes not containing $f$. If $f=x-a$, then $D(x-a)$ contains the generic point and every closed point except $(x-a)$. The only open set containing $(x-a)$ may contain many other points; the topology does not isolate it as Euclidean topology would.88Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

On $D(f)$, the regular functions form the localization $A_f$, allowing denominators that are powers of $f$. At a point $\mathfrak p$, the stalk $A_{\mathfrak p}$ allows every denominator outside $\mathfrak p$. This is the ring of functions defined near that point. Its unique maximal ideal consists of functions vanishing there in the algebraic sense.99Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

The construction turns familiar algebraic operations into geometric ones. A quotient $A/I$ gives a closed subscheme of $\operatorname{Spec}A$. Localization gives an open subscheme. A tensor product gives a fiber product of affine schemes:

$$\operatorname{Spec}B\times_{\operatorname{Spec}A}\operatorname{Spec}C \cong \operatorname{Spec}(B\otimes_A C).$$

The formula is a central reason for admitting arbitrary rings. Tensor products can produce nilpotents even if the original rings are reduced. A category that automatically reduces after every operation loses the natural output of its own products.

For example, let $k$ have characteristic different from two and consider a map that brings two points together. At the special parameter, the fiber product or scheme-theoretic fiber may acquire a ring like $k[x]/(x^2)$. The nonreduced fiber records the collision. If one immediately replaces it by its reduced point, one cannot expect intersection multiplicity or infinitesimal variation to remain visible.

This does not mean that every nilpotent has one intuitive geometric picture. “Infinitesimal thickness” is a serviceable phrase for square-zero ideals and thickenings. It becomes unreliable if it suggests a literal tiny Euclidean width. Nilpotents are algebraic elements killed by a power. Their geometric significance depends on the morphism, module, or deformation problem in which they occur.

The spectrum also reverses arrows. A ring map $\varphi:A\to B$ sends a prime $\mathfrak q\subset B$ to $\varphi^{-1}(\mathfrak q)\subset A$, producing $\operatorname{Spec}B\to\operatorname{Spec}A$. This reversal is not an inconvenience to hide. It expresses pullback: a function on the target becomes a function on the source. Once accepted, it makes algebraic constructions line up with geometric ones.

The precise object

For a commutative ring $A$, define

$$X=\operatorname{Spec}A.$$

The underlying set consists of prime ideals. Closed subsets are1010Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

$$V(I)=\{\mathfrak p\in\operatorname{Spec}A\mid I\subseteq\mathfrak p\}$$

for ideals $I\subseteq A$. Basic opens are $D(f)=X\setminus V((f))$. The assignment $D(f)\mapsto A_f$, with the natural restriction maps, determines a sheaf $\mathcal O_X$. The stalk at $\mathfrak p$ is $A_{\mathfrak p}$. The pair $(X,\mathcal O_X)$ is an affine scheme.

A locally ringed space is a topological space equipped with a sheaf of rings whose stalks are local rings. A scheme in modern terminology is a locally ringed space covered by open subsets isomorphic to affine schemes. A morphism consists of a continuous map and a compatible morphism of sheaves that induces local homomorphisms on stalks.

Historical terminology must now interrupt the definition. EGA I used préschéma for the general locally affine object and schéma when an additional separation condition held. In modern algebraic geometry, “scheme” usually means EGA’s prescheme, and “separated scheme” names the extra property. The condition is that the diagonal $X\to X\times_S X$ be a closed immersion for a morphism $X\to S$. This Part follows modern usage unless quoting the 1959–1961 sources.1111Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas,” I, §§2, 5.

The scheme is local in a strong but controlled sense. Affine opens may be glued along compatible open subschemes. Conversely, many properties can be checked on affine covers. Yet not every global question reduces mechanically to separate affine calculations. Cohomology measures failures of local data to fit globally; properness and projectivity concern the morphism as a whole; moduli and descent ask whether locally given families come from global ones.

Quasi-coherent sheaves give the bridge between modules and geometry. If $M$ is an $A$-module, there is an associated sheaf $\widetilde M$ on $\operatorname{Spec}A$, with $\widetilde M(D(f))\cong M_f$. EGA I systematized these sheaves and their coherent finite-type versions. Serre’s FAC had already demonstrated the power of coherent algebraic sheaves on varieties; the scheme setting extended the module–sheaf dictionary over more general rings and bases.1212Serre, “Faisceaux algébriques cohérents”; Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas,” I, §§9–10.

One point, two geometries

Let

$$D=\operatorname{Spec}k[\varepsilon]/(\varepsilon^2).$$

The ring has one prime ideal, $(\varepsilon)$, so $D$ has one topological point. Its reduction is $\operatorname{Spec}k$. But the global sections differ: $k[\varepsilon]/(\varepsilon^2)$ contains a nonzero nilpotent. A map $D\to X$ whose closed point lands at $x\in X(k)$ corresponds, under suitable hypotheses, to a tangent vector at $x$.1313Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

For an affine scheme $X=\operatorname{Spec}A$, such a map is a $k$-algebra homomorphism

$$A\longrightarrow k[\varepsilon]/(\varepsilon^2).$$

If its reduction modulo $\varepsilon$ is the point $x:A\to k$, the homomorphism has the form

$$a\longmapsto x(a)+\varepsilon\,d(a),$$

where $d:A\to k$ is a derivation relative to $x$. The product rule for the homomorphism is exactly the Leibniz rule

$$d(ab)=x(a)d(b)+x(b)d(a).$$

Thus a tangent vector becomes a map from an infinitesimal test scheme. The calculation is small, but its methodological consequence is large: geometry can be probed by nonreduced schemes.

Credit should remain exact. Differential and tangent constructions long predated scheme theory. The dual-number formulation is an instance of a broader functorial practice, not proof that Grothendieck invented tangent vectors. What the scheme framework did was make this probe available uniformly, over general bases and in categories of families.

The example also shows why points do not determine a scheme. If one keeps only the unique underlying point of $D$, the derivation disappears. Even residue fields at points do not recover the nilpotent global sections. The structure sheaf is essential.

There is a limit to the example. Not all deformation information is first order, and not every deformation problem is represented by an ordinary scheme. Higher-order Artin rings, obstruction groups, formal schemes, stacks, and later derived methods enter. The dual numbers reveal the doorway; they are not the whole house.

A line with more than its visible points

Return to $\operatorname{Spec}k[x]$. The generic point $\eta=(0)$ has residue field $k(x)$, the rational function field. Every nonempty open set contains $\eta$. The closed points $(x-a)$ are specializations of $\eta$. Algebraically, a statement true at the generic point may extend to an open subset but fail at special points.1414Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

This topology turns phrases such as “generically smooth” or “true on a dense open set” into assertions about a morphism near a point. It does not make generic truth equivalent to universal truth. Indeed, one of relative geometry’s disciplines is to track exactly where a property spreads and where it fails.1515Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

Over a non-algebraically closed field, closed points need not have residue field equal to the base field. For $k[x]$, irreducible polynomials give closed points with finite field extensions as residue fields. A point therefore includes its residue field. The naive picture of points as coordinate tuples over $k$ is insufficient even before nilpotents enter.

The scheme-theoretic point is economical: an irreducible closed subset has a unique generic point. This “sobriety” lets topology reflect prime ideals and irreducibility. But it can also tempt metaphysical exaggeration. A generic point is not an omniscient observer or an infinitely vague location. It is a prime ideal with a precisely defined closure and local ring.

In biography, generic points pose a narrative problem of their own. They are easy to dramatize as evidence that Grothendieck saw invisible worlds. The mathematics needs no such mythology. He adopted an algebraically forced topology because it made localization, specialization, and irreducible components work together. The conceptual boldness lies in granting the construction full geometric status and following its consequences.

The arithmetic spectrum

$\operatorname{Spec}\mathbb Z$ contains a generic point $(0)$ and closed points $(p)$. The local ring at $(p)$ is $\mathbb Z_{(p)}$; the residue field is $\mathbb F_p$. A scheme $X\to\operatorname{Spec}\mathbb Z$ may therefore have fibers $X_{\mathbb F_p}$ in every positive characteristic and a generic fiber $X_{\mathbb Q}$.1616Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

This picture contributed to the slogan that prime ideals are points. The slogan is exact within the definition and dangerous outside it. Prime numbers do not become positions in physical space. Rather, constructions traditionally associated with geometric families—fibers, local rings, specialization, properness—can be applied over the arithmetic base.1717Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

Take $X=\operatorname{Spec}\mathbb Z[x]/(f)$. Its fiber over $(p)$ is

$$\operatorname{Spec}\left(\mathbb F_p[x]/(\bar f)\right),$$

where $\bar f$ is the reduction of $f$ modulo $p$. The generic fiber is

$$\operatorname{Spec}\left(\mathbb Q[x]/(f)\right).$$

The formulas make reduction a fiber product. Questions about good and bad reduction become questions about a single morphism at different points of its base.

Again, the scheme does not solve those questions automatically. A fiber may be singular, nonreduced, disconnected, or dimensionally surprising. Flatness, smoothness, properness, and other relative properties are needed to control behavior. Chapter 27 will examine that grammar. Here the arithmetic example shows why the scheme could no longer be thought of merely as a broadened complex variety.

The move also changed the relation between local and global number theory. Spectra of rings of integers, localizations at primes, completions, and residue fields could be placed in geometric diagrams. Weil’s analogies between number fields and function fields gained a categorical setting of exceptional reach. Grothendieck did not originate those analogies. Scheme theory gave them a common operational language.

Gluing without pictures

Affine spectra are the local pieces, not all schemes. The projective line, for example, can be obtained by gluing two copies of $\operatorname{Spec}k[x]$ along $D(x)\cong\operatorname{Spec}k[x,x^{-1}]$, with the coordinate change $y=x^{-1}$. This recovers a familiar object through rings and localizations.1818Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

The gluing construction is rigorous. One specifies open subschemes $U_{ij}\subseteq U_i$ and isomorphisms $U_{ij}\to U_{ji}$ satisfying identity, symmetry, and cocycle conditions. The resulting locally ringed space is unique up to unique isomorphism in the appropriate sense. The phrase “patch the pieces” conceals these compatibility requirements.1919Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

Schemes can also be glued in ways that produce objects classical conventions would reject. Glue two affine lines along the complement of the origin while leaving the two origins distinct. The result is the affine line with doubled origin. It is a scheme in modern terminology but is not separated over $k$. In EGA’s terminology it is a prescheme but not a scheme.

This example explains why separation entered early. Algebraic analogues of Hausdorffness prevent certain doubled behaviors, though the analogy is not exact. The diagonal criterion is the reliable definition. Grothendieck’s theory did not indiscriminately celebrate every gluing; it provided a category broad enough to contain the result and properties precise enough to distinguish it.

The doubled origin also reveals a methodological choice. One could define “geometric object” so narrowly that bad gluings never count. Or one could admit the local object and mark the global defect by a morphism property. The second strategy often behaves better under constructions and makes hypotheses visible. It is characteristic of the scheme-theoretic foundation.

The genealogy of the object

Assigning the scheme to one isolated precursor would misstate EGA’s own genealogy. Its introduction traces prime spectra and local algebra through existing commutative algebra, names Zariski’s local methods, identifies Serre’s coherent-sheaf work as fundamental, discusses Chevalley’s schemes of local rings, and cites Nagata’s work over Dedekind domains. Grothendieck’s framework combined and generalized those paths, admitted spectra of arbitrary commutative rings, kept nilpotents, and made the resulting category the foundation of a large research program.2020Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”, introduction, pp. 5–8.

Dieudonné’s relation to the object differs from Grothendieck’s. The May 1959 Bourbaki lecture is Grothendieck’s exposition. EGA I attributes the work to Grothendieck and identifies Dieudonné as collaborator in its redaction. The correspondence shows Dieudonné drafting the treatise from Grothendieck’s materials and the continuing exchange. Dieudonné helped stabilize, organize, and disseminate the scheme language. The sources do not support relabeling the central conceptual generalization as Dieudonné’s invention, and they provide no justification for erasing his authorship-level labor from EGA.2121Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck and Serre, Correspondance Grothendieck–Serre, 73 and pp. 101–117.

Serre’s contribution is different again. FAC supplied the decisive demonstration that the Zariski topology and coherent sheaves could sustain cohomology. EGA’s introduction explicitly calls it fundamental and a natural preparation for schemes. Serre did not define the general scheme of EGA, and he was not its coauthor. His theorems, taste, questions, and criticisms were formative conditions of the work.

Chevalley deserves more than the status of precursor-name. EGA I notes that his local-ring schemes lent themselves naturally to an extension of algebraic geometry, though not with the flexibility sought there. Grothendieck’s October 1958 letter refers to a short Bourbaki paper inspired by conversations with Chevalley. The evidence establishes intellectual contact and a relevant earlier construction. It does not justify assigning every aspect of $\operatorname{Spec}A$ or scheme gluing to one conversation.

EGA I also names Nagata’s work over Dedekind domains as one route beyond geometry over a single field. Because Nagata’s papers have not been examined for this account, the chapter does not characterize his examples or infer a broader intellectual role from them. The primary claim is limited to EGA’s explicit acknowledgment.

The genealogy is therefore a network with a decisive organizer. Saying so does not weaken the originality of the scheme-theoretic program. Synthesis can be original when it changes which questions are natural, which objects are admitted, and which theorems become instances of one method.

Nilpotents: neither nuisance nor mysticism

The 1959 lecture makes nilpotents a point of controversy because they reveal the difference between classical points and scheme structure most sharply. If $n\in A$ is nilpotent, then $n$ belongs to every prime ideal. Consequently,2222Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

$$\operatorname{Spec}A \quad\text{and}\quad \operatorname{Spec}(A/\sqrt{0})$$

have the same underlying topological space. The difference is wholly in the structure sheaf.2323Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

Information that no prime point detects remains necessary because algebraic operations preserve or create it. Quotients can create it. Fiber products can create it. Infinitesimal neighborhoods are defined by it. Tangent and obstruction calculations use square-zero ideals. Intersection multiplicities are lost by premature reduction.

One can see a collision algebraically. Consider the family

$$X=\operatorname{Spec} k[t,x]/(x^2-t) \longrightarrow \operatorname{Spec}k[t].$$

For a nonzero parameter $t=a$ over a suitable field, the fiber may have two distinct geometric points after adjoining square roots. At $t=0$, the fiber is

$$\operatorname{Spec} k[x]/(x^2),$$

a doubled point. The nonreduced fiber records that the two solutions meet with multiplicity. Replacing it by $\operatorname{Spec}k$ erases how the family arrived.

The example has limits. In characteristic two the behavior requires separate care, and over a field where $a$ has no square root the visible point count changes. The scheme formulation accommodates these distinctions through residue fields and geometric fibers. A loose story about “two points merging” must not replace the actual base and characteristic.

Grothendieck’s insistence on nilpotents was thus methodological. It allowed the category to be closed under constructions and made infinitesimal data intrinsic. Later writers sometimes turn this insistence into a claim that unseen or spiritual dimensions are more real than visible ones. Nothing in the mathematics warrants that move. Nilpotents are exact ring elements with exact powers equal to zero.

Schemes and varieties

A classical variety can be made into a scheme, and under standard hypotheses much of its geometry is preserved. This inclusion explains both continuity and displacement. The old objects did not vanish. They became a distinguished class among more general ones.2424Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

The word “variety” itself has never had one universal definition. It may require irreducibility, reduction, finite type, separation, or an algebraically closed base, depending on author and period. Scheme theory did not replace one perfectly fixed category. It supplied terms in which the differences could be stated: reduced, irreducible, integral, finite type over a field, separated.2525Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

This adjectival strategy is powerful. Rather than hide assumptions inside the noun, it places them where theorems can use them. A scheme may be reduced but reducible, irreducible but nonreduced, integral, Noetherian, normal, regular, or smooth over a base. These properties are not synonyms. Scheme theory’s precision partly consists in refusing to make “nonsingular variety” carry all of them at once.

Regular and smooth provide a revealing example. Regularity is an absolute condition on local rings. Smoothness is relative to a base. Over a perfect field and under finite-type hypotheses, the notions align in familiar situations; over imperfect fields or general bases, they can diverge. The scheme language separates concepts that classical context allowed one to conflate while admitting more general cases.

The cost is vocabulary. A reader meets a thicket of modifiers. Yet the modifiers encode where arguments depend on field, finiteness, local algebra, or a morphism. EGA’s numbered architecture made that cost explicit rather than relying on tacit convention.

The limits of pictures

Every explanation of schemes oscillates between pictures and warnings. Draw $\operatorname{Spec}\mathbb Z$ as a line with primes; then say it is not a real or complex curve. Draw a doubled point as a thick dot; then say nilpotence has no Euclidean width. Draw a generic point above all closed points; then say the topology has no such vertical metric. The oscillation is not pedagogical failure. The object joins spatial language to algebra in a way no single familiar picture captures.2626Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

Grothendieck himself did not dispense with intuition. His choices were guided by what he regarded as natural methods. But “natural” here means compatible with structures and operations, not immediately visual. A scheme can become intuitive through use: localization becomes neighborhood, tensor product becomes fiber product, quotient becomes closed immersion, derivation becomes an infinitesimal point.2727Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

The biography should therefore avoid a familiar cinematic substitute: one exceptional mind sees spaces unavailable to everyone else. The sources show arguments, definitions, lectures, drafts, and polemic. They do not give access to an internal visual theater. What can be said is stronger because it is checkable: Grothendieck was willing to call general ring spectra geometric and to reorganize a major field around the consequences.

How points change with the observer

There is another way to approach a scheme that became increasingly important in Grothendieck’s work: ask what its points are over every test ring, not only over one field. For a scheme $X$ and a ring $R$, an $R$-valued point is a morphism2828Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

$$\operatorname{Spec}R\longrightarrow X.$$

If $X=\operatorname{Spec}A$ is affine, this is equivalent to a ring homomorphism $A\to R$. Ordinary $k$-rational points are the case $R=k$. Points over field extensions, local rings, dual numbers, and polynomial rings probe different aspects of the same object.2929Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

Take $X=\operatorname{Spec}k[x,y]/(y-x^2)$. A $k$-point is a pair $(a,b)$ with $b=a^2$. An $R$-point is a pair $(r,s)\in R^2$ with $s=r^2$. If $R$ has nilpotents, these points include infinitesimal variations invisible over fields. The equation stays the same; the observer changes.

The word “observer” is only an analogy. A test scheme is not a conscious viewpoint, and $R$-points are not subjective appearances. The benefit of the analogy is to prevent one chosen field from exhausting the object. The mathematical content is the functor

$$R\longmapsto X(R)$$

with its compatibility under ring homomorphisms, or equivalently the functor $T\mapsto\operatorname{Hom}(T,X)$ on schemes.

This perspective will become the subject of Chapter 28. Here it clarifies why the scheme cannot be reduced to its topological points. Two affine schemes are isomorphic if their functors of points are naturally isomorphic, by the Yoneda principle. Nilpotent test rings can distinguish schemes that all field-valued points fail to distinguish. The one-point dual-number scheme and the reduced point provide the simplest example.

The functorial point of view also moderates a common phrase: “a point of a scheme is a prime ideal.” That is true for points of its underlying topological space in an affine chart. But geometry routinely uses $T$-valued points, sections, geometric points $\operatorname{Spec}\Omega\to X$ with $\Omega$ separably or algebraically closed, and infinitesimal points. Scheme theory enlarged the noun and then enlarged the practice of probing it.

Grothendieck did not invent the elementary correspondence between solutions of equations and ring homomorphisms. His decisive move was to make such functorial probing structural, combine it with schemes over arbitrary bases, and use representability as an existence criterion. The object and its observers became two descriptions of one geometry.

Closed subschemes are more than closed sets

Let $X=\operatorname{Spec}A$. An ideal $I\subseteq A$ defines a closed subscheme3030Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

$$V(I)=\operatorname{Spec}(A/I)\longrightarrow X.$$

As a set, $V(I)$ depends only on the radical $\sqrt I$. As a closed subscheme, it depends on $I$. Thus $(x)$ and $(x^2)$ define the same closed subset of the affine line but different closed subschemes.3131Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

This distinction turns algebraic multiplicity into geometry. It also makes intersections behave correctly. If closed subschemes $Y$ and $Z$ of $X$ have ideals $I$ and $J$, their scheme-theoretic intersection is defined by $I+J$. Its structure may be nonreduced even when $Y$ and $Z$ are reduced. The nilpotents can measure tangency or excess intersection.

Consider the line $Y:y=0$ and parabola $Z:y=x^2$ in $\mathbb A^2_k$. Their scheme-theoretic intersection has coordinate ring

$$k[x,y]/(y,y-x^2)\cong k[x]/(x^2).$$

Set-theoretically they meet only at the origin. Scheme-theoretically the doubled point records tangency. This is a more stable explanation than saying the curves “really meet twice” at one location; the ring gives the exact meaning.

If instead $Z$ is the line $y=x$, the intersection ring is

$$k[x,y]/(y,y-x)\cong k,$$

a reduced point. The algebra distinguishes transverse and tangent meetings. Classical intersection theory already counted multiplicities. Scheme structure integrates the local data that produce those counts.

Closed immersions also reveal why a scheme morphism cannot be only a continuous map. The maps from $\operatorname{Spec}k$ and $\operatorname{Spec}k[\varepsilon]/(\varepsilon^2)$ into the affine line may have the same image point. Their sheaf maps differ. Geometry resides in the pullback of functions.

Globally, a quasi-coherent ideal sheaf $\mathcal I\subseteq\mathcal O_X$ defines a closed subscheme with structure sheaf $\mathcal O_X/\mathcal I$. This sheaf formulation lets local quotient rings glue. It also prepares formal geometry, where powers $\mathcal I^n$ define successive infinitesimal neighborhoods.

The human intuition is that a closed subobject has both a location and an equation with an order of vanishing. The limit is that different ideals can encode subtleties not captured by a simple notion of “thickness.” Embedded components, associated primes, and non-primary ideals require algebra, not pictures.

Local rings and singularity

At a point $x$ of a scheme $X$, the local ring $\mathcal O_{X,x}$ gathers functions defined near $x$. Its maximal ideal $\mathfrak m_x$ consists of germs vanishing at $x$, and its residue field

$$\kappa(x)=\mathcal O_{X,x}/\mathfrak m_x$$

records the point’s field of definition. The local ring is the algebraic microscope of scheme theory.3232Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

For a Noetherian local ring, regularity compares the Krull dimension with the dimension of the cotangent space:

$$\dim \mathcal O_{X,x} = \dim_{\kappa(x)}\mathfrak m_x/\mathfrak m_x^2.$$

When equality holds, the local ring is regular. The right-hand side measures the number of first-order generators needed at the point; its dual is related to tangent vectors. A singularity appears when the tangent dimension exceeds the local dimension in the finite-type setting.

Take the cusp $y^2=x^3$. At the origin, the maximal ideal is generated by $x$ and $y$, and the linear part of the equation vanishes. The tangent space has dimension two while the curve has dimension one. Scheme theory expresses this through the local ring, without requiring a complex-analytic picture and without restricting the base to $\mathbb C$.

Regularity alone is not the full relative notion of nonsingularity. A morphism $X\to S$ can have regular source and still fail to be smooth over the base. In positive characteristic, field extensions and inseparability create examples where geometric regularity matters. Grothendieck’s language would refine “nonsingular” into properties of local rings and morphisms.

This refinement illustrates a broad habit. The scheme is not valuable because it assigns an impressive object to every ring. It is valuable because it supports local invariants and relates them to maps. Dimension, depth, normality, regularity, flatness, and smoothness become distinct tools with controlled implications.

The chapter’s period saw much of this vocabulary still being stabilized. Serre’s annotation to the October 1958 correspondence says that Grothendieck’s period term was later replaced by smooth. A biography should not silently replace every historical term and then claim that the mature definition was present unchanged.3333Grothendieck and Serre, Correspondance Grothendieck–Serre, letter of October 17, 1958, pp. 73–74.

Products that force the issue

One of the strongest arguments for schemes is closure under fiber products. Suppose $X\to S$ and $Y\to S$ are schemes. Their fiber product $X\times_S Y$ should describe pairs of points with the same image in $S$, together with the appropriate residue-field and infinitesimal information. In affine form, tensor products construct it.

Tensor products do not respect a blanket demand for reduced rings. Even field extensions can produce nilpotents after tensoring when inseparability is present. If a category excludes the result, base change ceases to be an internal operation. One must either leave the category or discard information. Grothendieck’s broader category takes the algebraic output as geometric.3434Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

For a simple illustration in characteristic $p>0$, let $k$ be a nonperfect field and choose $a\in k$ that is not a $p$th power. Set

$$K=k[u]/(u^p-a).$$

Then after base change from $k$ to $K$,

$$K\otimes_k K$$

contains a nilpotent corresponding to $u\otimes1-1\otimes u$, whose $p$th power vanishes. A reduced field extension has produced a nonreduced self-product. The phenomenon is not decorative; it belongs to inseparable geometry.

Pierre Cartier had treated descent for radicial extensions of height one. Grothendieck’s December 1959 lecture explicitly said that, without schemes and especially without nilpotents, the essential identity between that case and Weil’s finite separable descent could not be formulated. This is a precise historical claim in a primary source, and it shows nilpotents doing conceptual work.3535Grothendieck, “Technique de descente et théorèmes d’existence en géométrie algébrique. I. Généralités. Descente par morphismes fidèlement plats,” 299.

The calculation also warns against identifying “reduced” with “well behaved.” Reduction is often useful and many central geometric objects are reduced. But reducedness is not preserved by every base change. The scheme framework lets one state exactly when geometric reducedness is preserved and when it is not.

Projective geometry survives

The affine spectrum can make the new foundation appear hostile to classical projective geometry. EGA II shows otherwise. From a graded ring one constructs $\operatorname{Proj}$, producing projective space, projective bundles, and closed projective subschemes. The construction is local on standard opens $D_+(f)$, whose coordinate rings are degree-zero parts of localizations.3636Alexander Grothendieck and Jean Dieudonné, “Éléments de géométrie algébrique II: Étude globale élémentaire de quelques classes de morphismes,” Publications Mathématiques de l’IHÉS 8 (1961): 5–222, https://www.numdam.org/item/PMIHES_1961__8__5_0/.

For a ring $A$,3737Grothendieck and Dieudonné, “Éléments de géométrie algébrique II: Étude globale élémentaire de quelques classes de morphismes”.

$$\mathbb P^n_A=\operatorname{Proj}A[x_0,\ldots,x_n]$$

is projective $n$-space over $\operatorname{Spec}A$. If $A$ is a field, it recovers the familiar projective scheme. If $A=\mathbb Z$, it becomes one family whose fibers are projective spaces over $\mathbb Q$ and the finite fields.

Homogeneous ideals define closed subschemes. Nilpotents and arbitrary bases are retained. The homogeneous coordinate method is not discarded; it is placed inside the relative category. Grothendieck’s Hilbert-scheme construction of 1961 would depend on this projective setting.

The phrase “schemes replaced varieties” therefore needs qualification. Scheme theory absorbed affine and projective coordinate geometry, broadened the base, and separated properties formerly packed into the word “variety.” Explicit coordinates remained essential in proofs and constructions. The abstraction reorganized computation rather than abolishing it.

There is also a chronological boundary. $\operatorname{Proj}$ and projective morphisms received systematic treatment in EGA II, published in 1961, just after EGA I’s language chapter. Their appearance belongs fully to this Part but points ahead from the scheme as object to the relative classes of maps studied next.

Three recurrent misreadings

The first misreading says a scheme is simply the set of solutions of equations over all fields. Field-valued solutions are important, but they miss nilpotent directions and may miss descent information. Test rings and the structure sheaf are indispensable.3838Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

The second says a scheme is a topological space whose points happen to be primes. The topology alone forgets the ring modulo nilpotents and cannot reconstruct local functions in general. A scheme is a locally ringed space with affine local models.3939Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

The third says scheme theory proves geometry is “really algebra.” Affine schemes exhibit a contravariant equivalence with commutative rings, but general schemes require gluing, and geometric concepts guide which algebraic constructions matter. The duality is productive precisely because neither vocabulary becomes dispensable.

A fourth, historical misreading is equally persistent: Serre invented sheaves, Grothendieck added schemes, and Dieudonné wrote them down. Sheaves predate Serre; Serre transformed their use in algebraic geometry. Spectra, local rings, and related constructions predate Grothendieck; he produced the decisive general framework and program. Dieudonné did far more than mechanical transcription. Accurate attribution is not a division of a trophy. It is a description of different kinds of mathematical work.

Finally, the scheme should not be made responsible for every later abstraction. Algebraic spaces relax representability by schemes in a topology; stacks retain automorphisms; derived schemes encode homological information in structure sheaves. These developments grow from problems visible in the scheme framework, but their inventors and dates remain their own.

Reduction, irreducibility, and integrality

Scheme language separates properties that a classical “variety” often bundled together. A scheme $X$ is reduced if its structure sheaf has no nonzero nilpotent sections locally. It is irreducible if its underlying topological space is not the union of two proper closed subsets. It is integral if it is both reduced and irreducible.4040Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

These distinctions are elementary and consequential. The doubled point is irreducible but not reduced. Two crossing lines form a reduced but reducible scheme. An integral affine scheme $\operatorname{Spec}A$ corresponds to a domain $A$.4141Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

Every scheme has a reduction $X_{\mathrm{red}}$ with the same underlying topological space and nilpotents removed. The map

$$X_{\mathrm{red}}\longrightarrow X$$

is a closed immersion and is universal for maps from reduced schemes into $X$. Reduction is therefore available when a problem genuinely ignores nilpotents. Scheme theory’s point is not to forbid reduction but to make it an explicit operation rather than an automatic reflex.

Irreducible components carry generic points in the Noetherian situations central to EGA. Associated points and embedded components refine the picture for coherent sheaves and nonreduced schemes. A closed subscheme can have an embedded component invisible among maximal-dimensional irreducible pieces.

These examples show why the scheme is a grammar of qualifications. “Take the variety defined by these equations” becomes: take the closed subscheme defined by this ideal; then ask whether it is reduced, irreducible, integral, equidimensional, or normal. The equations alone do not settle which structure an author intends if a radical is silently taken.

The precision prevents false equivalence. Passing from $I$ to $\sqrt I$ preserves the closed set but changes tangent spaces, intersection multiplicities, and deformation behavior. Passing from a reducible scheme to one component changes gluing at the intersections.

This is also where scheme language resists an unqualified structuralism. Relations and functors matter, but exact ideals still matter. The general framework does not dissolve concrete algebra.

Normalization and birational inheritance

Grothendieck’s foundation did not abandon the birational viewpoint it warned against treating exclusively. Normalization, function fields, valuations, and rational maps remain central.

For an integral affine scheme $X=\operatorname{Spec}A$ with fraction field $K$, the normalization is obtained from the integral closure $\widetilde A$ of $A$ in $K$:4242Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

$$\widetilde X=\operatorname{Spec}\widetilde A\longrightarrow X.$$

Under finite-type hypotheses over suitable bases, this morphism may be finite. It separates certain singular branches and produces a normal scheme.

For the cusp $A=k[t^2,t^3]\subset k[t]$, the normalization is the affine line $\operatorname{Spec}k[t]\to\operatorname{Spec}A$. The map is birational and modifies the singular point while preserving the function field.

The example shows continuity with Zariski’s and Weil’s geometry. Scheme theory can express birational maps over arbitrary bases and study exceptional fibers. The new foundation widens the stage; it does not declare generic methods irrelevant.

EGA I’s introduction objects to an “overly exclusive” birational habit, not to birational geometry itself. Its dedication to Zariski and Weil makes the distinction impossible to miss.4343Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”, introduction, pp. 5–8.

Normality also separates from smoothness. A normal scheme can have singularities, and smoothness is relative. The expanded vocabulary lets theorems state exactly which repair a normalization achieves and which defects remain.

Local versus global affineness

Every scheme is locally affine, but most important schemes are not affine. The projective line cannot be recovered as $\operatorname{Spec}$ of its ring of global functions, since over a field those functions are only constants.4444Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

This fact limits the slogan that geometry is opposite to commutative algebra. Affine schemes are contravariantly equivalent to rings. General schemes require gluing, and global sections forget information.4545Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

The projective line’s two affine charts demonstrate the point. Each chart is $\operatorname{Spec}k[x]$, their overlap is $\operatorname{Spec}k[x,x^{-1}]$, and the coordinate transition inverts $x$. The global object carries points and line bundles no single affine coordinate ring captures.

Quasi-coherent sheaves restore local algebra without claiming one global module suffices. On each affine chart they come from modules; compatibility on overlaps glues them. Cohomology measures some failures of global generation or exactness.

This local–global structure connects the scheme to Serre’s work. FAC showed that the coarse Zariski topology, coupled with coherent sheaves, supported powerful global theorems. EGA’s general schemes inherit that architecture.

The difference between local and global also cautions against a popular equation:

$$\text{scheme}=\text{ring}.$$

The correct affine duality is indispensable, but a general scheme is a ringed space locally dual to rings. The topology and gluing are not redundant packaging.

Residue fields and geometric points

A point $x\in X$ has a residue field $\kappa(x)$. The canonical map4646Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

$$\operatorname{Spec}\kappa(x)\longrightarrow X$$

selects the point with its field structure. If $\Omega$ is an algebraically closed extension of $\kappa(x)$, a map $\operatorname{Spec}\Omega\to X$ is a geometric point above $x$.

This distinction becomes crucial over nonclosed fields and in étale theory. A closed point on a real variety may have residue field $\mathbb C$. A closed point of a scheme over a finite field may have a finite extension as residue field. Counting only rational coordinate tuples misses such points.

The topological point and geometric point serve different tasks. Topological points encode closure and specialization. Geometric points provide base points for fiber functors and test geometric fibers. Neither should silently replace the other.

For $\operatorname{Spec}\mathbb Z$, the residue field at $(p)$ is $\mathbb F_p$, while at the generic point it is $\mathbb Q$. This is how one topological base carries different characteristics.

The scheme concept therefore changes “point” from a uniform atom into structured access. Points have neighborhood rings, residue fields, and specializations. Test schemes enrich access further.

This sophistication is not ornamental. The later étale fundamental group will require geometric base points because ordinary topological paths do not exist in the needed form. Part VI will follow that development rather than attribute it prematurely to EGA I.

The Noetherian frontier

Grothendieck’s definition admits spectra of arbitrary commutative rings. Many theorems in early EGA nevertheless impose Noetherian or finite-type hypotheses. The difference between the category’s scope and a theorem’s scope is central.4747Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

A Noetherian topological space satisfies the descending chain condition on closed subsets; a Noetherian ring satisfies the ascending chain condition on ideals. Finiteness makes irreducible decomposition, coherent modules, and induction manageable.

The broad definition remains useful even when proofs return to Noetherian schemes. Constructions and base changes naturally leave narrower categories; universal properties are clean in the broad category; and later generalization can identify exactly where finiteness was used.

But the broad definition does not guarantee all familiar results. On non-Noetherian schemes, finite type and finite presentation diverge, quasi-coherent ideals can behave unexpectedly, and coherence requires care. Modern statements often refine EGA’s Noetherian formulations.

Historical narration should not alternate between “arbitrary schemes” and Noetherian theorems without warning. Grothendieck’s foundational category was remarkably general; his strongest 1959–1961 existence and cohomology results had real hypotheses.

This layered generality was part of the program’s method. Build operations in a capacious category, then prove finiteness under conditions strong enough for the application. Generality and restriction are not inconsistencies.

A morphism through local rings

A morphism of affine schemes4848Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

$$f:\operatorname{Spec}B\longrightarrow\operatorname{Spec}A$$

comes from a ring homomorphism $A\to B$. At a point $\mathfrak q\subset B$, with image $\mathfrak p=\mathfrak q\cap A$, it induces a local homomorphism

$$A_{\mathfrak p}\longrightarrow B_{\mathfrak q}.$$

“Local” means the maximal ideal of $A_{\mathfrak p}$ maps into the maximal ideal of $B_{\mathfrak q}$, and its inverse image is that maximal ideal. The map respects what vanishes near the points.

For general schemes, a morphism consists of a continuous map and a sheaf homomorphism satisfying this local condition. The definition prevents arbitrary maps of ringed spaces from counting as geometric.

The local maps carry ramification, tangent, and fiber information. Tensoring $B_{\mathfrak q}$ with $\kappa(\mathfrak p)$ gives the local algebra of the fiber. Kähler differentials measure relative infinitesimal change.

This is why a scheme morphism is more than a function on prime ideals. Two ring maps can induce the same continuous map and different sheaf maps. Nilpotent information may live entirely in the latter.

The precision also explains contravariance. Geometry follows pullback of local functions. A point maps forward; a function maps back. Scheme theory does not choose between the directions but binds them.

What the 1959 lecture was doing

Exposé 182 is titled “Formal Geometry and Algebraic Geometry,” not “An Introduction to Schemes.” Its first section introduces schemes because the later comparison needs them. The definition is immediately placed in service of completions and existence.4949Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

This context corrects a museum-like presentation in which $\operatorname{Spec}A$ sits as an isolated invention. The lecture develops formal schemes, coherent modules, and algebraization; arbitrary rings and nilpotents are working necessities within that argument.

The lecture’s polemic against excluding nilpotents therefore precedes a theorem program. It is not a philosophical aside. Formal neighborhoods are built from powers of ideals; their structure rings are typically nonreduced. A reduced-only geometry could not even retain the approximations.

The move from section 1’s scheme language to later formal existence also demonstrates the three-layer method. The human problem is recovery of global algebraic objects from infinitesimal data. The intuition is an object known to every order along a closed set. The precision uses adic completions, coherent modules, and properness.

By embedding the scheme definition in a demanding application, Grothendieck provided the strongest answer to skepticism. The new objects were not asked to be admired; they were asked to work.

A dictionary tested in both directions

The affine dictionary is strongest when it translates not only objects but operations. A quotient $A\to A/I$ becomes a closed immersion. A localization $A\to A_f$ becomes an open immersion. A tensor product becomes a fiber product. An integral ring extension becomes an integral morphism.5050Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

Ideals correspond contravariantly to closed subschemes, but inclusion reverses:

$$I\subseteq J \quad\Longrightarrow\quad V(J)\subseteq V(I).$$

Prime ideals correspond to irreducible closed subsets through closure. Maximal ideals give closed points only under appropriate hypotheses; the scheme remembers nonmaximal primes too.

Modules become quasi-coherent sheaves. Finite modules become coherent sheaves in Noetherian settings. Exact sequences of modules become exact sequences of associated sheaves on affine schemes.

This dictionary is not perfect symmetry between every algebraic word and every geometric word. Global schemes need gluing, and global sections can forget them. Properties such as properness have no single-ring formulation outside the relative affine case.

The dictionary’s value lies in controlled translation. A geometric proof can localize to rings; an algebraic construction can globalize through sheaves. Neither side is declared more real.

Grothendieck’s 1959 lecture presents the affine dictionary as long known and argues that algebra’s generality should be admitted geometrically. The originality is not the observation that equations have coordinate rings. It is the refusal to stop the dictionary at reduced finite-type algebras over fields.

That refusal makes $\operatorname{Spec}\mathbb Z$, formal thickenings, inseparable fiber products, and local schemes ordinary. Each later success tests the decision.

It also creates new counterexamples. Nonseparated schemes, non-Noetherian phenomena, and nonreduced fibers require vocabulary. A broader category gives closure under operations at the cost of more ways to misbehave.

EGA answers by properties rather than exclusions. One can require separated, finite presentation, flat, proper, or smooth when a theorem needs them. The scheme remains broad; the hypothesis becomes precise.

This strategy differs from defining geometry to contain only well-behaved objects. It allows degeneration and failure to appear inside families, where they can be studied rather than discarded.

The dictionary therefore carries an ethic of explicit loss. Reduction, restriction to an open, normalization, or passage to a fiber are named operations. The mathematician can simplify, but the simplification leaves a morphism recording what was changed.

At the end of the chapter, that is the scheme’s most faithful intuition: not a mystical space behind equations, but an exact way to keep algebraic information and its transformations available to geometry.

A final limit on intuition

Every standard picture of a scheme privileges one feature. The spectrum picture emphasizes prime ideals. The functor-of-points picture emphasizes probes. The ringed-space picture emphasizes local functions. The family picture emphasizes a map to a base.5151Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

None can replace the others in every argument. Prime points do not detect nilpotents. Field-valued functors miss infinitesimals. A structure sheaf without functorial maps does not explain representability. A family picture can hide non-flatness.

Mathematical maturity with schemes consists partly in moving among these descriptions and knowing which information each preserves. EGA’s length reflects the work required to prove they cooperate.

This also limits biographical metaphor. One cannot infer Grothendieck’s private image of space from the formal definition. The documents show which descriptions he used publicly and which operations he regarded as natural.

The safest admiration is technical. He accepted arbitrary ring spectra as geometry, built relative operations on them, retained infinitesimal structure, and made the category support major existence theorems.

The safest reservation is technical too. Schemes do not represent every quotient or moduli problem, ordinary structure sheaves do not carry every derived intersection, and a general scheme can be poorly behaved.

Those limits were not all visible in 1959, but some already appeared through nonseparated examples, quotient obstacles, and automorphisms. The foundation grew by meeting them.

The scheme’s historical power lies in this double capacity: it preserved more information than classical varieties and made the remaining losses precise enough to motivate the next objects.

Credit for the scheme

The scheme is credited here as Grothendieck’s decisive general framework, not as an object without predecessors. Prime spectra, localization, local rings, sheaves, coherent methods, and arithmetic analogies had separate histories. Serre’s FAC demonstrated the power of coherent sheaves and Zariski cohomology on varieties. Chevalley’s local-ring schemes and Nagata’s work beyond fields are named by EGA itself. Dieudonné made the framework systematic in the treatise with Grothendieck. The general scheme, including arbitrary commutative rings and nilpotents as ordinary geometry, belongs centrally to Grothendieck’s program. Later functorial, stack-theoretic, and derived enlargements retain their later authors. This distribution is not diplomatic compromise; it describes different mathematical acts. That balance preserves the novelty of the synthesis while refusing to make inherited algebra, topology, and geometry look like raw material without authors. It also keeps modern explanatory examples separate from claims about what appeared explicitly in the 1959 lecture itself.5252Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”; Grothendieck, “Géométrie formelle et géométrie algébrique”.

The name scheme eventually became ordinary, but ordinariness is an outcome of adoption. In 1959 the category still needed defense against the instinct to reduce, close the base field, or privilege classical points. Its later familiarity should not be mistaken for inevitability. The definition prevailed because it supported calculations, constructions, and theorems while preserving the older objects as important cases. Its normality today records success, not absence of an intellectual rupture.

This vocabulary lets the chapter describe degeneration without treating it as corruption: a special fiber may be singular or nonreduced and still belong to the correct family. The adjective records no moral ranking.

From lecture to language

The May 1959 exposé did more than announce the object. It connected schemes to formal geometry and an existence theorem. By 1960 EGA I supplied the extended language. The movement from a Bourbaki lecture to 224 journal pages shows the difference between an idea and a foundation.

In the lecture, one can state that $\operatorname{Spec}A$ has primes, a topology, and a structure sheaf, then proceed to applications. In EGA, every transition requires development: functoriality of spectra, affine and open subschemes, products, change of base, separation, quasi-compactness, quasi-separatedness in the period’s terminology, reduced structures, associated sheaves, coherence, and formal completions. A research community can use the short exposition because many details are shared knowledge. A treatise must decide and prove them.

The published EGA I also recorded influences. Its historical remarks are brief, but they contradict the claim that the scheme arrived without ancestry. It names FAC, Chevalley’s local-ring schemes, Nagata’s work, Zariski, and Weil. The acknowledgments do not distribute credit with the detail a modern historian might want, but they provide a primary-source boundary against mythology.5353Grothendieck and Dieudonné, “Éléments de géométrie algébrique I: Le langage des schémas”, introduction, pp. 5–8.

At the same time, the treatise claims a transformation. FAC is described as an intermediate exposition between the classical viewpoint and the scheme viewpoint. Chevalley’s construction is said to lack the full flexibility intended here. Nagata’s extension is specialized to Dedekind bases. Grothendieck and Dieudonné defined the scope within which these paths could meet rather than compiling prior work.

The chapter title, “The Scheme,” uses a singular that the mathematics resists. There is no archetypal scheme behind all examples. There are affine and projective schemes, reduced and nonreduced schemes, schemes over fields and over $\mathbb Z$, local and formal constructions. What unifies them is a method of assembly and a category of morphisms.

That category shifts attention from substance to relation. The affine line is not only $\operatorname{Spec}k[x]$; it is an object that can map to a base, receive infinitesimal points, be pulled back, and sit in families. The full significance of schemes emerges when the arrow is placed beside the object:

$$X\longrightarrow S.$$

The scheme acquired its full force when it was placed over a base in a map $X\to S$.