体积的测度

Before there is a number

A school formula begins after agreement has already been reached. A rectangle has area $ab$; a rectangular solid has volume $abc$; a sphere has volume $4\pi r^3/3$. The symbols vary and the computations may be difficult, but the word volume is treated as settled. Grothendieck’s Montpellier question began one level earlier: what operation had been performed when a number was called the volume of an object? The problem included the class of admissible objects, invariance under cutting and reassembly, and extension from boxes to irregular bodies without dependence on a chosen approximation. In Récoltes et Semailles, he remembered spending most of his undergraduate energy from 1945 to 1948 on this cluster of questions.11Grothendieck, Récoltes et Semailles (Université des Sciences et Techniques du Languedoc and Centre National de la Recherche Scientifique, 1986); Jackson, “As If Summoned from the Void, I,” 1040–44; Laamri, Alexander Grothendieck: période 1928–1956, 69–79; Scharlau, “Who Is Alexander Grothendieck?” 931–33; Henri Lebesgue, “Intégrale, longueur, aire,” Annali di Matematica Pura ed Applicata, Serie III 7 (1902): 231–359, https://doi.org/10.1007/BF02420592.

The questions were not new. By 1945 mathematicians had inherited centuries of work on quadrature, limits, integration, and the geometry of curves and surfaces. Henri Lebesgue’s 1902 thesis had provided a powerful reorganization of integration, length, and area. The important biographical fact is not that Grothendieck preceded that work, but that his education had not placed it before him in a usable way. Jacques Soula told him of Lebesgue. Grothendieck later said he did not seek out the book, preferring to continue the construction. When he reached Paris, he learned that the territory had a name and a literature.22Grothendieck, Récoltes et Semailles (Université des Sciences et Techniques du Languedoc and Centre National de la Recherche Scientifique, 1986); Jackson, “As If Summoned from the Void, I,” 1040–44; Laamri, Alexander Grothendieck: période 1928–1956, 69–79; Scharlau, “Who Is Alexander Grothendieck?” 931–33; Lebesgue, “Intégrale, longueur, aire,” 231–359.

The word “rediscovery” is retained only as attributed retrospective language. Grothendieck’s later description says that he considered measurable sets and almost-everywhere convergence, and later biographers repeat that he reconstructed a broad form of the Lebesgue integral. No legible, authenticated copy of the whole Montpellier manuscript has been reviewed here. The evidence therefore does not establish that his definitions matched Lebesgue’s, that he proved a particular convergence theorem, or that he handled completion and nonmeasurable sets. A theory of integration is a network of definitions and theorems; sharing its destination does not prove identity of route.33Grothendieck, Récoltes et Semailles (Université des Sciences et Techniques du Languedoc and Centre National de la Recherche Scientifique, 1986); Jackson, “As If Summoned from the Void, I,” 1040–44; Laamri, Alexander Grothendieck: période 1928–1956, 69–79; Scharlau, “Who Is Alexander Grothendieck?” 931–33; Lebesgue, “Intégrale, longueur, aire,” 231–359.

The modern distinctions below clarify what a definition of volume demands and provide criteria for any future comparison with the manuscript. They describe standard measure theory, not necessarily Grothendieck’s student formulation. Historical claims about his work remain separately sourced.

Cutting without losing

Begin with the expectation that size is additive. If two measurable regions $A$ and $B$ do not overlap, the size of their union ought to be the sum of their sizes. For finitely many pieces this is finite additivity. Analysis quickly asks for more. An open interval can be assembled from countably many disjoint intervals; an open region in Euclidean space can be decomposed into countably many boxes; limiting procedures routinely produce countable families. A measure $\mu$ is therefore required to satisfy

$$\mu\!\left(\bigcup_{n=1}^{\infty}A_n\right) =\sum_{n=1}^{\infty}\mu(A_n)$$

for pairwise disjoint measurable sets $A_n$. This is countable, or sigma, additivity.

Countable additivity contains a theory of limits in compact form. If $A_1\subseteq A_2\subseteq\cdots$, write the union as the disjoint union of $A_1$ and successive differences. Then

$$\mu\!\left(\bigcup_n A_n\right)=\lim_n\mu(A_n).$$

If $A_1\supseteq A_2\supseteq\cdots$ and $\mu(A_1)<\infty$, a corresponding continuity from above holds. These conclusions show that measurement respects controlled approximation. They are not obvious consequences of assigning numbers to individual shapes; they follow from the architecture of the definition.

The ordinary intuition is that no size vanishes when an object is cut into infinitely many labeled pieces. That picture illuminates additivity but has limits. Infinite series may diverge; sets need not have visually distinct pieces; and a countable union is not a physical cutting performed in time. Countable additivity is an axiom about an abstract operation on sets. Its power comes precisely from applying beyond literal dissection.

One must next specify the admissible sets. A sigma-algebra $\Sigma$ on a space $X$ is a collection of subsets containing $X$, closed under complements, and closed under countable unions. These conditions also imply closure under countable intersections and differences. A measure space is a triple $(X,\Sigma,\mu)$, where $\mu:\Sigma\to[0,\infty]$ is countably additive and $\mu(\varnothing)=0$. The set $X$ says where the theory lives, $\Sigma$ says which questions about subsets the theory can answer, and $\mu$ gives the answers.

On a finite set, every subset can be measured. On the real line, however, translation invariance, countable additivity, and the usual length of intervals conflict with assigning a measure to every subset under the ordinary set-theoretic framework with choice. A Vitali construction makes the conflict explicit. Measurability records this real limit on how familiar properties can coexist.

This boundary is important for evaluating Grothendieck’s recollection. He wrote that he had not encountered a set that was not measurable. That does not make his work trivial; most analysis uses highly structured measurable sets. It does mean that the theory, as he remembered it, had not been forced against one of its foundational obstructions. Descriptions of a very general version therefore require a qualification: no audit against all modern boundaries has been possible.44Grothendieck, Récoltes et Semailles (Université des Sciences et Techniques du Languedoc and Centre National de la Recherche Scientifique, 1986); Jackson, “As If Summoned from the Void, I,” 1040–44; Laamri, Alexander Grothendieck: période 1928–1956, 69–79; Scharlau, “Who Is Alexander Grothendieck?” 931–33; Lebesgue, “Intégrale, longueur, aire,” 231–359.

From boxes to outer measure

Rectangular boxes offer a starting stock. In $\mathbb{R}^n$, a half-open box

$$Q=\prod_{i=1}^{n}[a_i,b_i)$$

receives volume $\prod_i(b_i-a_i)$. Half-open endpoints are convenient because adjacent boxes can be disjoint without leaving gaps. Finite unions of such boxes form an algebra of sets, and volume can be defined consistently there after subdivision. But an algebra is closed only under finite unions. Analysis wants countable operations.

The standard extension strategy first relaxes the problem. For any subset $A\subseteq\mathbb{R}^n$, cover $A$ by countably many boxes and take the cheapest total volume:

$$m^*(A)=\inf\left\{\sum_{k=1}^{\infty}\operatorname{vol}(Q_k): A\subseteq\bigcup_{k=1}^{\infty}Q_k\right\}.$$

This outer measure is defined for every subset. It is monotone and countably subadditive, meaning that the outer size of a union is no greater than the sum of the outer sizes. Yet outer measure is not countably additive on all subsets. It is an envelope, not the finished assignment.

Carathéodory’s criterion selects a set $A$ as measurable when it splits every test set $E$ exactly:

$$m^*(E)=m^*(E\cap A)+m^*(E\setminus A).$$

The resulting measurable sets form a sigma-algebra, and $m^*$ restricted to them is Lebesgue measure. The definition has a striking logic. Rather than asking whether $A$ itself looks regular, it asks whether inserting the boundary of $A$ distorts the outer measurement of any $E$. Measurability is tested relationally.

For an ordinary reader, one may imagine covering a spill with smaller and smaller square tiles and taking the lowest possible total area. This helps explain the infimum over coverings. It becomes misleading if taken literally. The boxes may overlap, there need not be an optimal covering, and the criterion concerns all scales and countable families rather than an actual tiling. The mathematical construction succeeds because infima and countable subadditivity are proved to interact, not because an ideal physical mosaic exists.

Other extension theorems begin with a premeasure on an algebra and extend it to the generated sigma-algebra. The Carathéodory extension theorem gives existence, and under sigma-finiteness a suitable uniqueness statement. This reveals a general pattern: define an object on simple pieces, identify the operations under which it must be stable, and prove an extension whose ambiguity is controlled. That pattern would later be congenial to Grothendieck, but the evidence does not show that this theorem was his Montpellier route.

Normalization and invariance matter. Lebesgue measure assigns an interval its length and is invariant under translations. On $\mathbb{R}^n$, it assigns boxes their ordinary volume and transforms under a linear map $T$ by

$$m(TA)=|\det T|\,m(A)$$

for measurable $A$. The determinant is therefore more than a formula from linear algebra; it is the factor by which $n$-dimensional volume scales. Rotations and reflections have determinant of absolute value one, so they preserve volume. This links the geometric expectations of school mathematics to the abstract measure.

The uniqueness has hypotheses. A translation-invariant measure on the Borel sets of $\mathbb{R}^n$, finite on compact sets and normalized on a unit cube, is Lebesgue measure. In modern language this is a special case of Haar measure on a locally compact group. The general statement lies beyond the student story, but it clarifies what “intrinsic” can mean: the measure is characterized not by a coordinate recipe alone but by invariance and normalization. Again, no claim is made that Grothendieck formulated Haar measure in Montpellier.

Length is not just one-dimensional volume

If the ambition begins with length, area, and volume together, dimensional distinctions soon appear. Lebesgue measure on $\mathbb{R}^2$ gives a line segment area zero; measure on $\mathbb{R}^3$ gives an ordinary surface volume zero. To measure a lower-dimensional object inside a higher-dimensional space, one needs a measure suited to its dimension or a parametrized definition. The word size is incomplete until its dimension and multiplicity conventions are stated.

Hausdorff measure provides one systematic answer. For $s\ge0$, cover a set $A$ by sets $U_i$ of diameter at most $\delta$, minimize $\sum_i(\operatorname{diam}U_i)^s$, and then let $\delta\downarrow0$, with a conventional normalization factor. The critical exponent where the result changes from infinity to zero leads to Hausdorff dimension. For sufficiently regular curves, one-dimensional Hausdorff measure agrees with ordinary arc length of the image; for regular surfaces, the appropriate Hausdorff measure gives area.

This construction postdates the basic problem in relevant formulations and belongs here as a clarification, not a student attribution. It shows that the demand for an intrinsic measurement can lead beyond the Lebesgue measure of ambient space. A curve can be continuous and so irregular that its length is infinite. A set can have fractional Hausdorff dimension. No finite list of elementary formulas anticipates all these cases.

Parametrization adds another choice. Let $\gamma:[a,b]\to\mathbb{R}^n$ be absolutely continuous. Then

$$L(\gamma)=\int_a^b\|\gamma'(t)\|\,dt.$$

If $\gamma$ traverses the same arc twice, this integral counts both traversals. The Hausdorff measure of the image does not automatically do so. The first measures a path with multiplicity; the second measures a set. A textbook that says only “length of the curve” may conceal this difference because ordinary examples are injectively parametrized.

For surfaces, the corresponding area formula involves the Jacobian of a parametrization. If $f:U\subseteq\mathbb{R}^2\to\mathbb{R}^3$ is regular and injective under appropriate hypotheses, then

$$\operatorname{Area}(f(U)) =\int_U\left\|\frac{\partial f}{\partial u}\times \frac{\partial f}{\partial v}\right\|\,du\,dv.$$

This is a powerful formula, but its proof and scope depend on definitions of area and regularity. It does not itself settle the area of every wild surface. The distinction between a formula under hypotheses and a definition across a class is the exact distinction that animated the Montpellier inquiry.

Integration by range rather than domain

Riemann integration partitions the domain into intervals or boxes and controls oscillation there. Lebesgue integration can be introduced by partitioning the range of a function and measuring the sets on which its values fall. The slogan is useful: Riemann asks where the graph lies over small pieces of the domain; Lebesgue asks how much of the domain carries values of a given size. But slogans are not definitions, and many modern treatments construct the integral through simple functions.

A nonnegative simple function has the form

$$s=\sum_{k=1}^{m}a_k\mathbf{1}_{A_k},$$

where the $A_k$ are measurable and the $a_k$ are nonnegative. Its integral is

$$\int_X s\,d\mu=\sum_{k=1}^{m}a_k\mu(A_k).$$

For a nonnegative measurable $f$, define

$$\int_X f\,d\mu= \sup\left\{\int_Xs\,d\mu:0\le s\le f,\ s\text{ simple}\right\}.$$

For a signed or complex function, positive and negative parts or real and imaginary parts are handled under integrability conditions.

This definition makes the integral an organized answer to the volume question. For the indicator $\mathbf{1}_A$, the integral is $\mu(A)$. For a nonnegative function on a region, the integral may be interpreted as the measure of the set under its graph in one higher dimension, provided the relevant product measure has been constructed. Area, volume, and total quantity enter one language.

The under-graph picture illuminates why integration generalizes volume. Its limit is that integration works on abstract measure spaces where there is no graph sitting in Euclidean space. Probability spaces, sequence spaces, and function spaces may carry measures. The definition through simple functions survives because it depends on measurable sets and values, not on a picture. This is the kind of generality that removes accidental geometry while preserving operations.

Lebesgue integrability is insensitive to changes on a null set. If $f=g$ almost everywhere, then their integrals agree whenever defined in the usual integrable sense. “Almost everywhere” means that the exceptional set has measure zero, not that exceptions are merely rare in colloquial terms. A dense set can have measure zero, as the rationals do in the real line. Conversely, a set can be nowhere dense and still have positive measure. Measure and topology answer different questions.

Grothendieck later remembered using almost-everywhere convergence before he knew the general notion of topological space. Mathematically, that is possible. Almost-everywhere convergence asks that for all points outside one null set, the numerical sequence $f_n(x)$ converge to $f(x)$. It uses the topology of the scalar values implicitly, but it does not require an abstract topology on the domain. His recollection thus conveys an uneven education rather than a contradiction.55Grothendieck, Récoltes et Semailles (Université des Sciences et Techniques du Languedoc and Centre National de la Recherche Scientifique, 1986); Jackson, “As If Summoned from the Void, I,” 1040–44; Laamri, Alexander Grothendieck: période 1928–1956, 69–79; Scharlau, “Who Is Alexander Grothendieck?” 931–33; Lebesgue, “Intégrale, longueur, aire,” 231–359.

Measuring a function by its level sets

For a nonnegative measurable function $f$, each value can be reconstructed from the levels below it:

$$f(x)=\int_0^\infty \mathbf 1_{\{f(x)>t\}}\,dt.$$

Tonelli’s theorem then gives the layer-cake formula

$$\int_X f\,d\mu =\int_0^\infty \mu\bigl(\{x:f(x)>t\}\bigr)\,dt.$$

The integral of a function is expressed through the measures of its superlevel sets.

For $f(x)=x$ on $[0,1]$, the set where $f(x)>t$ has length $1-t$ for $0\le t\le1$ and is empty above one. The formula yields

$$\int_0^1x\,dx=\int_0^1(1-t)\,dt=\frac12.$$

This is the familiar triangular area computed by horizontal slices rather than vertical ones.

The picture of stacked layers is exact for the displayed nonnegative case, but it has limits. A signed function requires separate treatment of positive and negative parts, and an infinite value is permitted before integrability is imposed. The formula also depends on measurability of the superlevel sets and on a theorem that justifies exchanging the two integrations.

This identity connects the Montpellier words area, volume, and integral without making them synonyms. A function determines a family of measurable sets; their sizes determine the total under the graph. It is a modern example of the structural relation Grothendieck sought, not evidence that his student manuscript contained the formula.

The formula also extends Cavalieri’s geometric principle. A body can be measured through the areas of parallel slices, while a function can be integrated through the measures of its level sets. In both cases, the lower-dimensional quantities vary with a parameter and are integrated to recover the total. The precise theorem replaces the informal claim that equal slices imply equal volumes with measurability and integration hypotheses.

The indicator function gives the limiting elementary case. If $f=\mathbf 1_A$, then $\{f>t\}=A$ for $0\le t<1$ and is empty for $t\ge1$. The layer-cake formula reduces to

$$\int_X\mathbf 1_A\,d\mu =\int_0^1\mu(A)\,dt =\mu(A).$$

Integration therefore contains measurement of sets as a special case rather than as a separate operation.

Theorems that govern passage to a limit

An integral becomes useful when it can pass through limits under intelligible conditions. The monotone convergence theorem states that if $0\le f_1\le f_2\le\cdots$ and $f_n(x)\to f(x)$, then

$$\int f\,d\mu=\lim_{n\to\infty}\int f_n\,d\mu.$$

This theorem is almost built into the simple-function construction, but “almost” still requires proof. It lets increasingly accurate nonnegative approximations carry their totals to the limit.

Fatou’s lemma gives a one-sided inequality:

$$\int\liminf_{n\to\infty}f_n\,d\mu \le \liminf_{n\to\infty}\int f_n\,d\mu$$

for nonnegative measurable functions. The dominated convergence theorem says that if $f_n\to f$ almost everywhere and $|f_n|\le g$ for an integrable $g$, then $f$ is integrable and

$$\int f\,d\mu=\lim_n\int f_n\,d\mu.$$

The dominating function prevents mass from escaping into increasingly narrow, increasingly tall spikes.

For an ordinary reader, domination is a uniform budget: every $f_n$ must fit beneath one integrable envelope. This conveys the control. It can also mislead, because $|f_n|\le g$ is pointwise almost everywhere and the proof uses measure-theoretic tools, not financial conservation. Moreover, pointwise convergence alone is insufficient. A sequence can converge at every point while its integrals fail to converge to the integral of the limit. The theorem’s hypotheses are the mathematical content, not technical clutter.

Consider $f_n=n\mathbf{1}_{(0,1/n)}$ on $(0,1)$. For every $x>0$, eventually $x\notin(0,1/n)$, so $f_n(x)\to0$. Yet

$$\int_0^1 f_n(x)\,dx=1$$

for every $n$. Height compensates for shrinking support. No single integrable function dominates all $f_n$. This example shows why a theory of total quantity needs rules for limits; intuition focused on each fixed point misses the moving concentration.

The record does not establish which convergence results appeared in the Montpellier pages. Grothendieck’s later reference to almost-everywhere convergence does not show that he proved dominated convergence in its modern form. A manuscript audit would need to identify the mode of convergence, the bounds imposed, and the conclusion established about integrals.66Grothendieck, Récoltes et Semailles (Université des Sciences et Techniques du Languedoc and Centre National de la Recherche Scientifique, 1986); Jackson, “As If Summoned from the Void, I,” 1040–44; Laamri, Alexander Grothendieck: période 1928–1956, 69–79; Scharlau, “Who Is Alexander Grothendieck?” 931–33; Lebesgue, “Intégrale, longueur, aire,” 231–359.

Null sets and completion

A set $N$ is null if $\mu(N)=0$. If every subset of every null measurable set is also measurable, the measure space is complete. Lebesgue measure is usually taken in its completed form. Borel measure on the sigma-algebra generated by open sets need not itself include every subset of a Borel null set; completion adds them. Thus even after “the measurable sets” have been introduced, there may be more than one natural sigma-algebra in play.

This matters when functions are identified almost everywhere. If a function differs from a measurable function only on a subset of a null set, completeness ensures the altered function remains measurable. The convenience is structural: the equivalence relation “equal almost everywhere” behaves smoothly. But completion can have costs under maps and product constructions, where one must track which sigma-algebra is intended. Measure theory replaces one vague word with a collection of exact choices, not with a single frictionless object.

The young Grothendieck’s later admission about not meeting nonmeasurable sets may include this boundary as well as Vitali-type pathologies. Without the manuscript, one cannot tell whether his collection was a sigma-algebra, a completed sigma-algebra, or a more informal class closed under the operations he needed. Historical accuracy requires leaving those options open.

Definitions often reveal new questions at their edges. Countable additivity raises the problem of which sets are measurable; almost-everywhere equality raises completeness; product measures raise agreement of iterated integrals. A successful foundation does not terminate inquiry but organizes the next failures. The Montpellier record documents an encounter with the initial measure problem, not the full later pattern.

Reversing the order: Tonelli and Fubini

Volume in a product space can be computed by slicing. For a nonnegative measurable function $f$ on $X\times Y$, Tonelli’s theorem, under standard sigma-finiteness or appropriate measure-space hypotheses, permits

$$\int_{X\times Y} f\,d(\mu\times\nu) =\int_X\left(\int_Y f(x,y)\,d\nu(y)\right)d\mu(x)$$

with values possibly infinite. Fubini’s theorem gives the corresponding equality for integrable signed or complex functions and allows the order of integration to be exchanged.

The ordinary picture slices a loaf and adds the areas of the slices. It is faithful for suitable Euclidean bodies and motivates Cavalieri’s principle. It becomes unreliable when sections are not measurable or when positive and negative contributions are not absolutely integrable. Conditional double series already warn that changing order can change an answer. Fubini’s hypotheses ensure that “sum the slices” is not an arbitrary procedure.

This pair of theorems also shows why double and triple integrals in a curriculum can conceal foundational work. Writing $\int\!\int f(x,y)\,dx\,dy$ assumes that the measures and measurable function have been defined and that an iterated integral represents the product integral. A calculational course may properly postpone those issues. A student asking why the notation is legitimate has moved from calculus into analysis.

No evidence says that Grothendieck formulated product measures in this language during 1945–1948. But the question of volume naturally leads there. If the manuscript is recovered, its treatment of sections, repeated integration, and countable operations would help locate it relative to contemporary theory. Until then, the mathematical landscape can be mapped without planting his flag on every theorem.

From measurable functions to spaces

Measure theory does more than assign integrals. It creates spaces of functions. For $1\le p<\infty$, define

$$\|f\|_p=\left(\int_X|f|^p\,d\mu\right)^{1/p},$$

and let $L^p(X,\mu)$ consist of measurable functions with finite $p$-norm, identified when they agree almost everywhere. For $p=\infty$, the norm is the essential supremum. These are Banach spaces; $L^2$ is a Hilbert space with inner product $\langle f,g\rangle=\int f\overline g\,d\mu$.

The passage from individual functions to a function space would become central to Grothendieck’s first research field. A function is now a point in an infinite-dimensional vector space. Integration defines linear functionals; kernels define operators; convergence can be studied through norms or weaker topologies. Montpellier’s problem did not yet amount to topological vector space theory, but measure and integration provided some of its most important examples.

There is a conceptual reversal here. At first, a function helps measure a region by integration. Later, all integrable functions become an object whose geometry can itself be studied. Distance between functions is defined by an integral. Duality asks which continuous linear functionals act on the space. Compactness asks when families of functions possess convergent subsequences in a chosen topology. These were among the questions alive in Nancy.

The familiar formula

$$\|f+g\|_p\le\|f\|_p+\|g\|_p$$

is Minkowski’s inequality, and Hölder’s inequality controls $\int|fg|$. Such inequalities are not decorative estimates. They prove that the definitions create a stable linear and metric structure. A general theory must repeatedly show that its operations remain inside the declared space.

One must resist a neat causal line from the Montpellier integral to the Nancy thesis. Grothendieck’s interest in functional analysis was shaped in Paris by Leray’s course, conversations, and the research program of Dieudonné and Schwartz. The $L^p$ example shows mathematical continuity, not documented personal causation. It explains why his first problem and first professional field were adjacent enough to speak to one another.77Grothendieck, Récoltes et Semailles (Université des Sciences et Techniques du Languedoc and Centre National de la Recherche Scientifique, 1986); Jackson, “As If Summoned from the Void, I,” 1040–44; Laamri, Alexander Grothendieck: période 1928–1956, 69–79; Scharlau, “Who Is Alexander Grothendieck?” 931–33; Lebesgue, “Intégrale, longueur, aire,” 231–359.

The measure of an independent construction

Later retrospective accounts place the Montpellier work in territory already associated with Lebesgue’s measure and integration theory. Without an authenticated manuscript, its theorem-level scope and any historically new result cannot be determined. The bounded biographical claim is that Grothendieck described an independent route through questions of measure and integration; Magnier’s later recollection says that this work influenced his assessment of the scholarship application.88Grothendieck, Récoltes et Semailles (Université des Sciences et Techniques du Languedoc and Centre National de la Recherche Scientifique, 1986); Jackson, “As If Summoned from the Void, I,” 1040–44; Laamri, Alexander Grothendieck: période 1928–1956, 69–79; Scharlau, “Who Is Alexander Grothendieck?” 931–33; Lebesgue, “Intégrale, longueur, aire,” 231–359.

Later accounts emphasize that Grothendieck reconstructed the theory with limited resources, but this does not mean that he had no tools. He had secondary schooling, university courses, books for the program, calculus, and a small text on abstract spaces mentioned in his later recollection. The relevant assessment includes both resources and absences.

Nor should independent discovery be made a moral superiority. Mathematics is cumulative because people read one another. Had Grothendieck consulted Lebesgue early, he might have reached new questions sooner. His refusal was productive for his formation and inefficient for the literature. Those two statements coexist. The later research mathematician would need both capacities: the power to rebuild a subject and the discipline to discover what the community already knew.

Grothendieck wrote in Récoltes et Semailles that he learned in solitude something no master can truly teach. The statement concerns the act of doing mathematics, not the dispensability of inherited knowledge. A proof becomes one’s own understanding only through active reconstruction, even when its idea comes from a teacher. Conversely, active reconstruction becomes public mathematics only when it enters a shared record. Montpellier supplied the first half of that relation. Paris would insist on the second.99Grothendieck, Récoltes et Semailles (Université des Sciences et Techniques du Languedoc and Centre National de la Recherche Scientifique, 1986); Jackson, “As If Summoned from the Void, I,” 1040–44; Laamri, Alexander Grothendieck: période 1928–1956, 69–79; Scharlau, “Who Is Alexander Grothendieck?” 931–33; Lebesgue, “Intégrale, longueur, aire,” 231–359.

Where analogy stops

Volume invites metaphors of filling: cover an object with tiny cubes, pour water into it, weigh a uniform material. Each image captures a property and conceals another. Cubes suggest approximation but privilege coordinates. Water suggests additivity but presupposes a physical container and three dimensions. Weight suggests a measure with density but introduces material variation. None decides measurability or proves countable additivity.

The most faithful general intuition is not “measure is the amount of stuff.” Abstract measure may represent probability, mass, time, counting, or a spectral distribution. It need not be finite, and points can have positive measure or zero measure depending on the space. The invariant content is an assignment to admissible sets compatible with countable disjoint union. Even this sentence leaves out signed and complex measures, which extend the idea while sacrificing nonnegativity.

Probability is a measure normalized by $\mu(X)=1$. Counting measure assigns to a set its number of points, possibly infinity. A Dirac measure $\delta_x$ concentrates unit mass at one point. Lebesgue measure assigns a point mass zero. These examples show that a measure is not determined by the underlying set alone. The same $X$ can carry many measures, each encoding a different question.

This multiplicity helps answer the young question in a mature form. There is no context-free “real volume” waiting behind every object. There are characterizations whose hypotheses make one measure canonical for a task. Euclidean invariance and normalization single out Lebesgue measure in its domain; a density $w(x)$ produces a weighted measure $w(x)\,dx$; a manifold carries volume forms or Riemannian volume after additional structure is chosen. Intrinsic does not mean structureless. It means that the dependence on structure has been stated and behaves naturally under the relevant transformations.

A definition that opens rather than closes

Soula’s reported remark that Lebesgue had brought mathematics to an endpoint was preserved by Grothendieck for irony. Measure theory did the opposite. It opened probability, harmonic analysis, ergodic theory, functional analysis, and modern partial differential equations. Even within foundations, the existence of nonmeasurable sets, the choice of sigma-algebra, and relations between measures remained active. A definition can resolve a confusion without exhausting the subject.

Grothendieck’s later career would be marked by definitions whose importance lay in the problems they made visible. It is tempting to read that style directly back into 1945. The evidence supports a smaller statement. At Montpellier he learned through practice that asking for a satisfactory definition could occupy years and produce an interconnected theory. He also learned, on reaching Paris, that a private solution might be both mathematically serious and historically belated.1010Grothendieck, Récoltes et Semailles (Université des Sciences et Techniques du Languedoc and Centre National de la Recherche Scientifique, 1986); Jackson, “As If Summoned from the Void, I,” 1040–44; Laamri, Alexander Grothendieck: période 1928–1956, 69–79; Scharlau, “Who Is Alexander Grothendieck?” 931–33; Lebesgue, “Intégrale, longueur, aire,” 231–359.

This double lesson guarded against two errors. One is accepting a formula because authority has printed it. The other is believing that personal conviction establishes novelty. First principles challenge the first error; entry into a research community corrects the second. Neither can replace the other.

The manuscript has become emblematic of this junction, but the evidence remains limited. Grothendieck’s mature recollection describes densely written pages. Their exact sequence, definitions, and proofs have not been verified against an authenticated complete manuscript. No facsimile or modernized table of contents can stand in for that object.

The remaining gap

At the end of the Montpellier years, the gap in the textbooks had changed form. In his later account, Grothendieck presented himself as having found a framework for the quantities that troubled him. His separation from the existing literature remained until Parisian mathematicians identified the connection with Lebesgue’s theory. Later recollections also describe the reconstruction as evidence of unusual capacity.1111Grothendieck, Récoltes et Semailles (Université des Sciences et Techniques du Languedoc and Centre National de la Recherche Scientifique, 1986); Jackson, “As If Summoned from the Void, I,” 1040–44; Laamri, Alexander Grothendieck: période 1928–1956, 69–79; Scharlau, “Who Is Alexander Grothendieck?” 931–33; Lebesgue, “Intégrale, longueur, aire,” 231–359.

Another gap then became visible: the algebraic and topological vocabulary common to the seminar world. Measure theory had given him a domain in which he could work deeply, but not the broad preparation his new peers assumed. That asymmetry would define his first Paris year. He could pursue a foundational question for three years and still be baffled by basic terms in another field. Expertise is local before it becomes broad.

The measure problem ended in a change of scale rather than a new public theorem. An isolated inquiry had reached the edge of a collective discipline. Grothendieck next had to learn what others had named, proved, and left open while retaining the habit of reconstructing reasons for himself.

The problem before Lebesgue

To understand what was already known by 1945, one must avoid a fable in which calculus had possessed only crude formulas until Lebesgue supplied rigor at a stroke. Cauchy and Riemann had given precise forms of integration for broad classes of functions. Jordan developed a content for bounded sets whose boundaries were sufficiently tame. Borel’s work on sets and measure established the importance of countable operations. Lebesgue reorganized these developments and extended their reach, but he worked inside an active history. The student in Montpellier was separated from that history, not standing before an untouched problem.1212Grothendieck, Récoltes et Semailles (Université des Sciences et Techniques du Languedoc and Centre National de la Recherche Scientifique, 1986); Jackson, “As If Summoned from the Void, I,” 1040–44; Laamri, Alexander Grothendieck: période 1928–1956, 69–79; Scharlau, “Who Is Alexander Grothendieck?” 931–33; Lebesgue, “Intégrale, longueur, aire,” 231–359.

For a bounded function $f:[a,b]\to\mathbb{R}$, the Riemann integral can be defined through upper and lower sums. Given a partition $P:a=x_0<\cdots<x_n=b$, set

$$U(f,P)=\sum_{i=1}^{n} \left(\sup_{x\in[x_{i-1},x_i]}f(x)\right)(x_i-x_{i-1})$$

and define $L(f,P)$ similarly with infima. The function is Riemann integrable when the infimum of all upper sums equals the supremum of all lower sums. This is a definition, not just a recipe, and it already turns area into a problem of approximation.

Lebesgue’s advance should therefore not be described as replacing “no definition” with definition. One key difference is the class of functions and the mode of convergence the theory can manage. A bounded function on a compact interval is Riemann integrable exactly when its set of discontinuities has Lebesgue measure zero. The indicator of the rationals is discontinuous everywhere and not Riemann integrable, but it is Lebesgue measurable and has Lebesgue integral zero because the rationals form a countable null set. The indicator of the irrationals then has integral equal to the interval’s length.

This example can sound paradoxical only if the integral is expected to average point labels by counting points. Both rationals and irrationals are dense; every interval contains infinitely many of each. Lebesgue measure does not count them equally. A countable set has measure zero, while the complement of a countable set in an interval has full measure. Density is topological; measure is measure-theoretic. The example dramatizes how a new definition separates notions that ordinary language merges.

Jordan content offers another boundary. A bounded set is Jordan measurable when its inner and outer elementary contents agree; equivalently, for bounded sets under standard conditions, its boundary has Lebesgue measure zero. Jordan measurable sets form an algebra but not a sigma-algebra. They work well for many classical regions and poorly under countable limiting operations. Lebesgue measure’s closure under countable unions is not abstraction for its own sake. It is what lets analysis take limits without leaving the domain of measurement.

The historical phrase “the Lebesgue integral” can refer to several equivalent constructions presented in different orders. One may begin with measurable sets, simple functions, monotone limits, or distribution functions. An independent student could reach an equivalent integration procedure without duplicating Lebesgue’s exposition. Conversely, a procedure agreeing on many examples might fail an extension or convergence theorem. That is why an archival comparison must inspect definitions and proofs rather than hunt only for familiar formulas.

Lebesgue’s thesis title—Intégrale, longueur, aire—is especially relevant because it names almost the same triad Grothendieck later recalled. The coincidence does not show that Soula handed him the thesis or that he saw its title; his testimony says he did not seek out the book. It shows how naturally the elementary dissatisfaction leads to the historical center of measure theory. A verified image of the thesis title page could make this relation visible without pretending that the two projects had equal historical standing.1313Grothendieck, Récoltes et Semailles (Université des Sciences et Techniques du Languedoc and Centre National de la Recherche Scientifique, 1986); Jackson, “As If Summoned from the Void, I,” 1040–44; Laamri, Alexander Grothendieck: période 1928–1956, 69–79; Scharlau, “Who Is Alexander Grothendieck?” 931–33; Lebesgue, “Intégrale, longueur, aire,” 231–359.

When boundaries carry the difficulty

For ordinary geometric regions, measure is often controlled by the boundary. A rectangle’s boundary has $n$-dimensional Lebesgue measure zero in $\mathbb{R}^n$, so choices about including or excluding edges do not change its volume. This is why half-open, open, and closed boxes all receive the same measure. The technical convenience of half-open boxes should not be confused with a physical distinction among them.

More irregular boundaries can carry positive measure. A “fat Cantor set” has empty interior yet positive Lebesgue measure. Removing open intervals in a carefully chosen construction leaves a closed set with no interval inside but nonzero length. The classical middle-thirds Cantor set, by contrast, is uncountable and has measure zero. Cardinality, interior, density, and measure therefore diverge.

Such examples explain why definitions cannot rely on visual regularity. A finite drawing of a Cantor construction looks like rows of intervals. It cannot display the limiting set’s uncountability or measure by inspection. The proof tracks the total length removed and uses continuity of measure. Mathematical pictures suggest a process; countable additivity determines the result.

Boundary behavior also governs Riemann integrability of indicators. The function $\mathbf{1}_A$ is discontinuous precisely at the boundary of $A$. If that boundary has measure zero, the indicator is Riemann integrable and its integral gives the Jordan content. Lebesgue integration admits many more measurable boundaries. Thus a geometric question about a region and an analytic question about a function are connected by the indicator operation. This bridge between objects and the function spaces built from them is one reason measure theory feeds functional analysis.

These examples establish distinct levels of possible achievement: measuring elementary regions, closing the domain under countable operations, controlling irregular boundaries, and proving convergence of integrals. Without the manuscript, the phrase “a theory of measure” does not establish which levels were reached.

Signed quantity and density

Physical and analytic quantities are not always nonnegative. Charge can have signs; a derivative may define a signed distribution of change. A signed measure $\nu$ is countably additive with values in the extended real line under conditions preventing the undefined expression $+\infty-\infty$. The Hahn decomposition theorem separates the space into a positive part and a negative part, and the Jordan decomposition writes

$$\nu=\nu^+-\nu^-$$

with mutually singular positive measures. Its total variation is $|\nu|=\nu^++\nu^-$.

This language makes “total” more delicate. Cancellation can make $\nu(X)$ small while $|\nu|(X)$ is large. The distinction resembles the difference between net displacement and total distance traveled. That analogy is faithful to cancellation but not to the full theory: positive and negative sets in a Hahn decomposition are measure-theoretic, and uniqueness is only up to null sets.

If a measure $\nu$ is absolutely continuous with respect to $\mu$, written $\nu\ll\mu$, then $\mu(A)=0$ implies $\nu(A)=0$. Under sigma-finite hypotheses, the Radon–Nikodym theorem gives a measurable density $f$ such that

$$\nu(A)=\int_A f\,d\mu.$$

The function $f=d\nu/d\mu$ is determined almost everywhere. This theorem formalizes the idea that one distribution of quantity has a density relative to another.

Density is a powerful ordinary intuition. Mass density turns spatial volume into mass; probability density turns Lebesgue measure into probability. Its limitation is that not every measure has a density relative to Lebesgue measure. A Dirac mass at one point is singular: the point has Lebesgue measure zero but Dirac measure one. Mixed measures may have continuous and atomic parts. The background measure must be named before “the density” has meaning.

The Radon–Nikodym theorem was not a Montpellier novelty, and no claim is made that Grothendieck reconstructed it. It belongs in the explanatory map because a theory advertised as measure and integration is often judged by how it relates measures to linear functionals. Later functional analysis turns integration against $f$ into a duality pairing. The theorem is one of the routes by which geometric quantity becomes operator language.

Measurable does not mean integrable

A measurable function may have infinite integral. On $(0,1)$, the function $f(x)=1/x$ is measurable and nonnegative, but

$$\int_0^1\frac{1}{x}\,dx=\infty.$$

The function $g(x)=x^{-1/2}$ is integrable there, with finite integral, even though it is unbounded. Boundedness and integrability are different properties; the size of the region on which a function is large matters.

For a signed function, the expression $\int f$ is well defined as a finite Lebesgue integral when $\int|f|<\infty$. If both $\int f^+$ and $\int f^-$ are infinite, subtracting them is meaningless. Principal values sometimes assign symmetric limits to such expressions, but those are different constructions with additional structure. The definition refuses an answer where cancellation would depend on order.

This refusal is philosophically important. A rigorous extension does not maximize the number of expressions bearing a value at any cost. It marks when the desired operations fail to determine one. The same discipline appears in measurability: not every subset is admitted merely because one wants a universal notion of size. A good general language expands a domain while keeping failures explicit.

Improper Riemann integrals and Lebesgue integrals can also diverge in meaning. The integral of $\sin x/x$ over $(0,\infty)$ converges as an improper integral but not absolutely, so it is not Lebesgue integrable in the finite $\int|f|$ sense over that domain. Rearrangement and multidimensional order issues then become delicate. “More general” is not a single linear ranking; different extensions preserve different operations.

If Grothendieck’s manuscript used a notion of integral broader or narrower than the standard Lebesgue one, that would need to be described on its own terms. Historical comparison should ask which properties were preserved, not assume every extension lies on one ladder culminating in the modern textbook.1414Grothendieck, Récoltes et Semailles (Université des Sciences et Techniques du Languedoc and Centre National de la Recherche Scientifique, 1986); Jackson, “As If Summoned from the Void, I,” 1040–44; Laamri, Alexander Grothendieck: période 1928–1956, 69–79; Scharlau, “Who Is Alexander Grothendieck?” 931–33; Lebesgue, “Intégrale, longueur, aire,” 231–359.

Modes of convergence

A sequence of functions can converge pointwise, uniformly, almost everywhere, in measure, or in an $L^p$ norm. These modes are related under hypotheses and distinct in general. Pointwise convergence fixes $x$ and lets $n$ vary. Uniform convergence requires one index after which all points are controlled:

$$\sup_{x\in X}|f_n(x)-f(x)|\to0.$$

Almost-everywhere convergence permits one null exceptional set. Convergence in measure requires the measure of the set where the error exceeds any fixed $\varepsilon$ to tend to zero.

On a finite measure space, almost-everywhere convergence implies convergence in measure, but the converse may hold only after taking a subsequence. $L^p$ convergence implies convergence in measure under common finite-measure or suitable general hypotheses, yet it need not imply pointwise convergence of the whole sequence. These relations form a map, not a chain in which each stronger word automatically contains the next without conditions.

The “typewriter sequence” on $[0,1]$ illustrates movement without pointwise rest. Enumerate indicators of dyadic intervals, moving across each scale before halving the interval length. The sequence converges to zero in measure because each support’s length tends to zero, but at every point the values return to one infinitely often, so pointwise convergence fails. Concentration can move rather than remain at one point.

Egorov’s theorem upgrades almost-everywhere convergence on a finite measure space to uniform convergence outside a set of arbitrarily small measure. Lusin’s theorem says, in a standard finite-measure setting, that a measurable function is continuous outside a set of arbitrarily small measure. These theorems make measurable behavior resemble classical behavior after discarding a small set, but “small” means small measure, not empty, finite, or topologically negligible.

The distinctions matter because the phrase “converges” is incomplete in a space of functions. Later, locally convex spaces would let several seminorms encode different kinds of control. Montpellier’s almost-everywhere language already contained the seed of that pluralism at the level of examples. This is a mathematical relationship, not evidence that he had a theory of locally convex topology in 1948.

An archival audit should record every convergence sign in the manuscript and determine its definition. It must establish whether exceptional sets depended on $n$, whether convergence was pointwise outside one null set, which bounds permitted exchange of limit and integral, and whether “uniform” referred to the domain or to a family of sets. Modern readers can otherwise import current conventions into historical notation.

The geometry of $L^p$

The spaces $L^p$ turn integration into geometry, but their points are equivalence classes rather than literal functions. If two representatives differ only on a null set, their $L^p$ distance is zero, so they represent the same point. This quotient is essential: without it, $\|\cdot\|_p$ would be only a seminorm. A reader who imagines the space as a bag of graphs misses the identification on which its normed structure depends.

For $1<p<\infty$, if $q$ satisfies $1/p+1/q=1$, Hölder’s inequality gives

$$\left|\int fg\,d\mu\right| \le \|f\|_p\|g\|_q.$$

Thus $g\in L^q$ defines a continuous linear functional on $L^p$. Under standard hypotheses, every continuous linear functional arises this way. The dual of $L^p$ is $L^q$. Endpoints and general measure spaces require care: the dual of $L^1$ is represented by $L^\infty$ under broad standard conditions, while the dual of $L^\infty$ is generally larger than $L^1$.

This is the entrance to duality. Instead of asking only what a function is, analysis asks how all continuous linear measurements of functions can be represented. Integration supplies the pairing. Schwartz and Dieudonné’s Nancy program would investigate duality far beyond normed $L^p$ spaces, where the topology is generated by families of seminorms and several dual topologies compete.

The $L^2$ case adds orthogonality. Cauchy–Schwarz makes $\langle f,g\rangle=\int f\overline g$ an inner product, and completeness makes $L^2$ a Hilbert space. Fourier analysis can then treat functions as vectors expanded along orthogonal directions. Volume has traveled a long way: the integral first assigns a total under a graph and then defines angles in an infinite-dimensional space.

This trajectory helps explain why the student’s interest could lead naturally toward functional analysis without making the transition inevitable. The measure problem creates examples whose structural questions—duality, completeness, compactness, operators—were central after the war. Paris gave Grothendieck names and teachers for those questions. Nancy gave him unresolved versions.1515Grothendieck, Récoltes et Semailles (Université des Sciences et Techniques du Languedoc and Centre National de la Recherche Scientifique, 1986); Jackson, “As If Summoned from the Void, I,” 1040–44; Laamri, Alexander Grothendieck: période 1928–1956, 69–79; Scharlau, “Who Is Alexander Grothendieck?” 931–33; Lebesgue, “Intégrale, longueur, aire,” 231–359.

Proof as compression

A list of correct formulas can be extensive and still lack a theory. A theory compresses them by showing that many conclusions follow from a small number of definitions and general theorems. Countable additivity yields continuity under monotone set limits. The simple-function construction yields monotone convergence. Hölder’s inequality yields continuity of dual pairings. Each compression makes the reason for many special cases visible.

Generality earns its place when it performs this compression without discarding the examples that motivated it. An abstraction that merely renames each case has achieved little. One that proves one theorem from which many calculations follow has changed the economy of reasoning. Grothendieck’s mature work is often praised in these terms. At Montpellier, the securely documented fact is that he sought something more compressive than separate recipes for lengths, areas, and volumes.1616Grothendieck, Récoltes et Semailles (Université des Sciences et Techniques du Languedoc and Centre National de la Recherche Scientifique, 1986); Jackson, “As If Summoned from the Void, I,” 1040–44; Laamri, Alexander Grothendieck: période 1928–1956, 69–79; Scharlau, “Who Is Alexander Grothendieck?” 931–33; Lebesgue, “Intégrale, longueur, aire,” 231–359.

There is a danger in calling every unification “simple.” The definitions above require sigma-algebras, extended real values, equivalence almost everywhere, and limiting theorems. They are simpler structurally, not necessarily shorter for a beginner. Simplicity can mean fewer arbitrary exceptions, clearer functorial behavior, or one proof replacing many. It does not mean the absence of technical work.

Grothendieck’s later prose contrasted a nourishing, childlike engagement with mathematics against professional routine. The Montpellier account supports the narrower fact that he followed a basic word until its structure unfolded. It does not establish independent reconstruction as the only authentic path to understanding measure theory.

What a located manuscript would have to show

If pages from the Montpellier work are identified, authentication comes first. Paper, handwriting, provenance, pagination, and relation to the manuscript Grothendieck said he carried to Paris must be established. A later set of notes cannot silently substitute for the 1945–1948 document. Images should preserve margins and recto-verso relations; transcription should record deletions and uncertain symbols rather than regularize them immediately.1717Grothendieck, Récoltes et Semailles (Université des Sciences et Techniques du Languedoc and Centre National de la Recherche Scientifique, 1986); Jackson, “As If Summoned from the Void, I,” 1040–44; Laamri, Alexander Grothendieck: période 1928–1956, 69–79; Scharlau, “Who Is Alexander Grothendieck?” 931–33; Lebesgue, “Intégrale, longueur, aire,” 231–359.

The mathematical audit should then inventory the primitives taken as given: real numbers, sets, intervals, topology, and sequences. It must identify the defined collection of measurable sets, its closure under finite or countable operations, the range and finiteness properties of the measure, and whether the construction begins with outer content, inner approximation, or functions. Every later theorem depends on these choices.

Invariance and extension form the next test. The audit must determine agreement with interval length and box volume, translation invariance, treatment of open and closed sets, the status of any extension theorem, measurability of subsets of null sets, and the normalization used in any uniqueness claim. Equivalence to Lebesgue theory depends on these answers.

For integration, the audit should distinguish bounded from unbounded functions, nonnegative from signed functions, and pointwise from almost-everywhere identities. It should record whether simple functions are used, how infinite integrals are handled, and which interchange-of-limit results are proved. A familiar notation such as $\int f$ cannot be allowed to carry modern properties that the text has not established.

Finally, the manuscript requires historical comparison with Lebesgue, Daniell, Carathéodory, Riesz, and textbooks available in Montpellier. Independent work can converge on known structures without textual borrowing, while a phrase or notation may reveal an unremembered source. The aim is an exact account of what the manuscript contains, not confirmation or disproof of a later legend.

Until that audit is possible, the most that can be said is deliberately bounded: between 1945 and 1948, dissatisfied with textbook treatments of length, area, and volume, Grothendieck independently developed a substantial approach to measure and integration; in Paris he learned that its central territory was already known through Lebesgue and others. This wording preserves achievement, priority, and uncertainty at once.1818Grothendieck, Récoltes et Semailles (Université des Sciences et Techniques du Languedoc and Centre National de la Recherche Scientifique, 1986); Jackson, “As If Summoned from the Void, I,” 1040–44; Laamri, Alexander Grothendieck: période 1928–1956, 69–79; Scharlau, “Who Is Alexander Grothendieck?” 931–33; Lebesgue, “Intégrale, longueur, aire,” 231–359.

Coordinates change, measure transforms

An intrinsic theory must explain what happens when coordinates change. If $\Phi:U\to V$ is a sufficiently regular one-to-one map between open subsets of $\mathbb R^n$, the change-of-variables theorem states, under standard hypotheses,

$$\int_V f(y)\,dy =\int_U f(\Phi(x))\,|\det D\Phi(x)|\,dx.$$

The Jacobian determinant is the local volume-scaling factor. It converts a coordinate description without pretending that volume is independent of all structure.

For a linear map $A$, the factor is constant: $|\det A|$. A singular map with determinant zero collapses dimension and cannot support the same substitution formula as a diffeomorphism. For nonlinear maps, the derivative supplies a linear approximation at each point, and integration assembles the local factors.

The familiar image is a grid distorted by $\Phi$, with each tiny box changing volume by the Jacobian. It is a faithful heuristic for smooth maps. Its limit is that a proof cannot rely on infinitesimal boxes as actual sets of nonzero size. Approximation, differentiability, measurability, and control of overlaps turn the image into a theorem.

This theorem clarifies the student’s word intrinsic. A coordinate formula is not automatically nonintrinsic. It can describe an intrinsic quantity when its transformation law compensates exactly for changing coordinates. The determinant is the compensation.

On a smooth manifold, a volume form or density gives compatible local expressions. Integration is defined in coordinate charts and shown independent of the chosen atlas through change of variables. Additional structure, such as a Riemannian metric and orientation or a density avoiding orientation choices, determines the volume element.

Thus an intrinsic definition does not remove local coordinates from every calculation. It proves that different local calculations agree. This local-to-global pattern would later matter profoundly in geometry, but no claim is made that the Montpellier manuscript contained manifold integration.

Measurement and choice

Measure theory repeatedly separates arbitrary choices from invariant outcomes. One chooses box coverings to define outer measure, simple functions to approximate an integrand, representatives of almost-everywhere classes, and coordinates for substitution. The final value must not depend on the choice within stated conditions.

Independence is established by a theorem, not by suppressing the choice in notation. Writing $\int f$ hides partitions, approximants, and representatives only after their compatibility has been proved. A first-principles question often begins by noticing such hidden choices.

This gives a precise philosophical content to the Montpellier episode. Grothendieck was not seeking only a more abstract word for volume. He was seeking a rule whose outcome survived the permissible ways of calculating it. The mature theory answers by defining the permissible class and proving invariance.

The answer is conditional rather than absolute. Change the sigma-algebra, measure, multiplicity convention, dimension, or geometric structure, and “volume” may change. Rigor does not reveal one number outside every context; it makes context and invariance explicit.