The Area and Coarea Formulas

Intuition Beginner

Two of the most useful theorems in measure theory answer one question apiece. The first: if a map stretches and folds a piece of space, how much area does the image cover? The second: if you cut a region into the level sets of some function, how do the sizes of those slices add back up to the size of the whole? The area formula answers the first. The coarea formula answers the second. Together they make the words "surface area" and "integrate over the fibres" rigorous even for maps that are merely Lipschitz, not smooth.

Start with the area formula. A map from a smaller-dimensional space into a larger one paints out a surface. At each point the map has a local stretch factor, the Jacobian, which records how much a tiny patch of the source is magnified when it lands in the target. Adding up the Jacobian over the source gives the area of the image, counted with multiplicity: if the map folds two patches of source onto the same target point, that point is counted twice. This is the honest generalisation of the change-of-variables rule from first-year calculus.

Now the coarea formula. Fix a map that collapses a region down to a lower-dimensional space, like the height function on a landscape. Each value picks out a level set, a slice of the region. The coarea formula says the integral of the Jacobian over the region equals the total size of all the slices, summed across every level. When the map is a flat projection, this is exactly the rule that lets you compute a volume by integrating cross-sectional areas. So the coarea formula is the curved, nonlinear cousin of slicing a solid into thin sheets.

The reason both formulas need the Jacobian is that the map distorts. A slice taken where the map changes fast is spread thin; a slice where the map barely moves is bunched up. The Jacobian is exactly the bookkeeping weight that corrects for this distortion, so the slices reassemble into the right total.

The one-line summary: the area formula measures how much surface a map sweeps out, and the coarea formula integrates a region by carving it into level-set slices. Both replace smoothness with the much weaker Lipschitz condition, which is what makes them the workhorses of geometric measure theory.

Visual Beginner

Picture two panels. On the left, a flat square of rubber is stretched and curled into a saddle-shaped sheet floating in space. A tiny dot on the square becomes a tiny ellipse on the sheet; the ratio of the ellipse's area to the dot's area is the Jacobian at that point. The area formula adds up these local magnifications across the whole square to get the area of the sheet. Where the rubber is folded so that two spots land on one, that spot of the sheet is shaded twice, recording the multiplicity.

On the right, a solid blob is sliced by a height function into a stack of horizontal level sets, drawn as nested curves. Each level set is a fibre of the map. The coarea formula says: weight each point of the blob by its Jacobian, integrate, and you get the same number as taking the size of every slice and integrating those sizes over all the heights. The picture makes the slogan visible: area sweeps a map forward, coarea cuts a region into fibres.

Worked example Beginner

We use the coarea formula in its most familiar disguise: polar-style integration in the plane.

Step 1. Take the map sending a point to its distance from the origin, so the map records radius. Its level sets are circles centred at the origin. The radius-$r$ circle is the fibre over the value $r$.

Step 2. The local stretch factor of the distance map is $1$ everywhere away from the origin, because moving one unit outward changes the radius by one unit. So the Jacobian weight is $1$.

Step 3. The coarea formula then says: integrate any function over the plane by first integrating it around each circle, then integrating those circle-integrals over all radii from $0$ to infinity. In symbols, the plane integral of a function $g$ equals the integral over $r$ of the quantity obtained by integrating $g$ around the radius-$r$ circle.

Step 4. The size of the radius-$r$ circle is its circumference $2\pi r$. For a function depending only on radius, integrating around the circle multiplies by $2\pi r$, and we recover the standard rule that the plane integral becomes the radial integral with the weight $2\pi r$.

What this shows: the everyday polar-coordinate weight is not a special trick. It is the coarea formula applied to the distance map, with the circumference of each level circle supplying the geometric factor.

Check your understanding Beginner

Formal definition Intermediate+

Throughout, ${\mathcal{L}}^n$ denotes Lebesgue measure on ${\mathbb{R}}^n$ and ${\mathcal{H}}^k$ the $k$-dimensional Hausdorff measure from 02.07.02. A map $f:{\mathbb{R}}^n\to {\mathbb{R}}^m$ is Lipschitz when there is a constant $L$ with $|f(x)-f(y)|\le L|x-y|$ for all $x,y$. By Rademacher's theorem 02.07.10, such an $f$ is differentiable ${\mathcal{L}}^n$-almost everywhere, with derivative the linear map $Df(x):{\mathbb{R}}^n\to {\mathbb{R}}^m$ given a.e. by the matrix of partial derivatives.

Definition (Jacobian of a Lipschitz map). For a linear map $L:{\mathbb{R}}^n\to {\mathbb{R}}^m$, the $n$-dimensional Jacobian is $$ J L ;=; \begin{cases} \sqrt{\det!\big(L L^{T}\big)}, & m\le n,\[2pt] \sqrt{\det!\big(L^{T} L\big)}, & n\le m. \end{cases} $$ For a Lipschitz $f$ we write $Jf(x)=J(Df(x))$, defined for ${\mathcal{L}}^n$-a.e. $x$ ^{[Federer 1969 §3.2.3]}. By the singular-value decomposition, if $L$ has singular values ${\sigma }_1,\dots ,{\sigma }_k$ with $k=\min (n,m)$, then $JL={\sigma }_1\cdots {\sigma }_k$: the Jacobian is the product of the principal stretch factors, and the formula is the volume of the image of the unit cube under $L$ on the directions where $L$ is injective.

Definition (multiplicity function). For $A\subseteq {\mathbb{R}}^n$ measurable and $y\in {\mathbb{R}}^m$, the multiplicity is $$ N(f,A,y) ;=; #\big(A\cap f^{-1}(y)\big)\in{0,1,2,\dots,\infty}, $$ the number of preimages of $y$ inside $A$. The map $y\mapsto N(f,A,y)$ is ${\mathcal{H}}^n$-measurable when $n\le m$.

Theorem (area formula, $n\le m$). For Lipschitz $f:{\mathbb{R}}^n\to {\mathbb{R}}^m$ and ${\mathcal{L}}^n$-measurable $A\subseteq {\mathbb{R}}^n$, $$ \int_{A} Jf,d\mathcal L^{n} ;=; \int_{\mathbb R^{m}} N(f,A,y),d\mathcal H^{n}(y). $$ When $f$ is injective on $A$ the right side is ${\mathcal{H}}^n(f(A))$, so the integral of the Jacobian is the ${\mathcal{H}}^n$-measure of the image. More generally, for ${\mathcal{L}}^n$-summable $g$, $$ \int_{A} g(x),Jf(x),d\mathcal L^{n}(x) ;=; \int_{\mathbb R^{m}}\Big(\sum_{x\in A\cap f^{-1}(y)} g(x)\Big),d\mathcal H^{n}(y). $$

Theorem (coarea formula, $m\le n$). For Lipschitz $f:{\mathbb{R}}^n\to {\mathbb{R}}^m$ and ${\mathcal{L}}^n$-measurable $A\subseteq {\mathbb{R}}^n$, $$ \int_{A} Jf,d\mathcal L^{n} ;=; \int_{\mathbb R^{m}} \mathcal H^{n-m}\big(A\cap f^{-1}(y)\big),d\mathcal L^{m}(y). $$ Equivalently, for non-negative measurable $g$, $$ \int_{A} g,Jf,d\mathcal L^{n} ;=; \int_{\mathbb R^{m}}\Big(\int_{A\cap f^{-1}(y)} g,d\mathcal H^{n-m}\Big),d\mathcal L^{m}(y). $$ The fibres $A\cap f^{-1}(y)$ are $(n-m)$-dimensional for ${\mathcal{L}}^m$-a.e. $y$ ^{[Federer 1969 §3.2.11]}.

Counterexamples to common slips Intermediate+

• The Jacobian, not the determinant. For non-square $Df$ there is no determinant; the Jacobian is the square root of a Gram determinant, equal to the product of singular values. Writing $\det Df$ when $n\ne m$ is meaningless.
• Multiplicity is essential. The area formula returns ${\mathcal{H}}^n(f(A))$ only when $f$ is injective on $A$. For a folding map the integral of $Jf$ overcounts the image, exactly by the multiplicity, which is the content of the formula.
• Coarea needs the right Jacobian. In the coarea case ($m\le n$) the Jacobian is $\sqrt{\det (DfD{f}^T)}$, an $m\times m$ Gram determinant. On the critical set where $\mathrm{rank}Df<m$ the Jacobian vanishes, and those fibres contribute nothing to the left side even though they may be large.

Key theorem with proof Intermediate+

Theorem (linear area formula). Let $L:{\mathbb{R}}^n\to {\mathbb{R}}^m$ be linear and injective with $n\le m$. Then for every ${\mathcal{L}}^n$-measurable $A\subseteq {\mathbb{R}}^n$, $$ \mathcal H^{n}\big(L(A)\big) ;=; JL\cdot\mathcal L^{n}(A). $$

Proof. By the polar decomposition write $L=OS$ where $S=\sqrt{L^{T}L}$ is symmetric positive definite on ${\mathbb{R}}^n$ and $O:{\mathbb{R}}^n\to {\mathbb{R}}^m$ is a linear isometry onto its image. Since $O$ preserves the Euclidean metric, it preserves $n$-dimensional Hausdorff measure: ${\mathcal{H}}^n(O(E))={\mathcal{H}}^n(E)$ for $E\subseteq {\mathbb{R}}^n$, where on the source we identify ${\mathcal{H}}^n={\mathcal{L}}^n$ (the two agree on ${\mathbb{R}}^n$, a standard consequence of the isodiametric inequality from 02.07.02).

It remains to compute ${\mathcal{H}}^n(S(A))={\mathcal{L}}^n(S(A))$ for the symmetric positive map $S$ on ${\mathbb{R}}^n$. Diagonalise $S$ in an orthonormal basis with eigenvalues ${\sigma }_1,\dots ,{\sigma }_n>0$, the singular values of $L$. In that basis $S$ rescales the $i$-th axis by ${\sigma }_i$, so it maps the unit cube to a box of volume ${\sigma }_1\cdots {\sigma }_n$. By translation invariance and the scaling behaviour of Lebesgue measure under diagonal maps, $$ \mathcal L^{n}(S(A))=\sigma_{1}\cdots\sigma_{n}\cdot\mathcal L^{n}(A)=\sqrt{\det(L^{T}L)}\cdot\mathcal L^{n}(A)=JL\cdot\mathcal L^{n}(A). $$ Composing with the isometry $O$ leaves the measure unchanged, giving ${\mathcal{H}}^n(L(A))=JL\cdot {\mathcal{L}}^n(A)$. $\square$

Bridge. This linear computation is the foundational reason the full area formula holds. The general Lipschitz statement is assembled from it by localisation: Rademacher's theorem 02.07.10 supplies an approximate derivative $Df(x)$ at almost every point, and the Whitney-Lusin approximation theorem partitions the domain (up to a set of small measure) into countably many pieces on which $f$ is close to its linearisation $Df(x_0)$ uniformly, so the linear formula applies on each piece with controlled error. Putting these together, summing the piecewise contributions and passing to the limit gives the integral $\int Jfd{\mathcal{L}}^n$ on one side and the multiplicity-weighted ${\mathcal{H}}^n$-image on the other. This is exactly the pattern that builds toward the coarea formula, whose linear case is dual to this one: there the source is split along the kernel of $L$ and the fibres carry the $(n-m)$-dimensional measure. The same machinery appears again in 02.13.07 and 02.13.11, where the area formula defines the mass of a rectifiable current and the coarea formula governs the slicing of currents by level sets, and it generalises the elementary change-of-variables rule and Fubini's theorem to the non-smooth setting. The central insight is that one linear identity, localised by a.e. differentiability and $C^1$ approximation, upgrades into both global change-of-variables theorems at once.

Exercises Intermediate+

Advanced results Master

The area and coarea formulas are the analytic core of geometric measure theory. Their full power appears once they are read not as change-of-variables tricks but as the definitions that make non-smooth geometry computable.

Theorem 1 (general area formula with multiplicity). For Lipschitz $f:{\mathbb{R}}^n\to {\mathbb{R}}^m$ with $n\le m$ and ${\mathcal{L}}^n$-summable $g$, $$ \int_{\mathbb R^{n}} g(x),Jf(x),d\mathcal L^{n}(x)=\int_{\mathbb R^{m}}\sum_{x\in f^{-1}(y)} g(x),d\mathcal H^{n}(y). $$ The proof localises via Rademacher 02.07.10: the domain decomposes (up to ${\mathcal{L}}^n$-null) into countably many Borel pieces $E_k$ on which $f$ restricts to a bi-Lipschitz map with $Df$ within $\epsilon$ of a fixed injective linear map, by the Whitney-Lusin $C^1$-approximation. The linear area formula applies on each $E_k$, and the multiplicity assembles the global identity ^{[Federer 1969 §3.2.3]}.

Theorem 2 (rectifiable sets and the ${\mathcal{H}}^k$ integrand). A set $M\subseteq {\mathbb{R}}^m$ is countably $k$-rectifiable when ${\mathcal{H}}^k$-almost all of it is covered by countably many Lipschitz images of ${\mathbb{R}}^k$. The area formula gives the integral-geometric meaning of ${\mathcal{H}}^k(M)$: parametrising the pieces by Lipschitz maps and summing $\int Jf$ recovers ${\mathcal{H}}^k(M)$, and the approximate tangent plane exists ${\mathcal{H}}^k$-a.e. This is the measure-theoretic substitute for a smooth atlas and underlies the definition of the mass of a rectifiable current in 02.13.07 ^{[Simon 1983 §11]}.

Theorem 3 (coarea for BV and Sobolev functions). The coarea formula extends from Lipschitz maps to functions of bounded variation: for $u\in BV({\mathbb{R}}^n)$, $$ |Du|(\mathbb R^{n})=\int_{-\infty}^{\infty}\operatorname{Per}\big({u>t}\big),dt, $$ where $\mathrm{Per}$ is the perimeter and $\{u>t\}$ the super-level set. The total variation equals the integral of the perimeters of the level sets. For Sobolev $u\in W^{1,1}$ this reads $\int |\nabla u|=\int \mathrm{Per}(\{u>t\})dt$, the bridge between the gradient and the geometry of level sets ^{[Maggi 2012 §8]}.

Theorem 4 (coarea proof of the isoperimetric and Sobolev inequalities). The coarea formula reduces the $W^{1,1}$ Sobolev inequality $\Vert u{\Vert }_{L^{n/(n-1)}}\le C\Vert \nabla u{\Vert }_{L^1}$ to the isoperimetric inequality on each level set. Writing $u$ through its super-level sets and applying the coarea identity together with the isoperimetric inequality $\mathrm{Per}(E)\ge n{\omega }_n^{1/n}{\mathcal{L}}^n(E{)}^{(n-1)/n}$ yields the sharp Sobolev constant. This is the Federer-Fleming route to Sobolev embeddings ^{[Maggi 2012 §11]}.

Theorem 5 (eikonal and the structure of level sets). If $u$ is Lipschitz with $|\nabla u|=1$ a.e. (an eikonal solution, such as a distance function), the coarea formula gives ${\mathcal{L}}^n(A)={\int }_{\mathbb{R}}{\mathcal{H}}^{n-1}(A\cap \{u=t\})dt$, so the volume swept between two level sets equals the integral of the slice areas. This identity drives the layer-cake decomposition of currents in 02.13.11 and the tube formulas of integral geometry ^{[Federer 1959]}.

Synthesis. The area and coarea formulas are the foundational reason geometric measure theory can compute with non-smooth objects, and the central insight is that a single linear scaling law, localised by almost-everywhere differentiability, becomes two global theorems at once. The area formula is dual to the coarea formula: one pushes an $n$-dimensional source forward to its ${\mathcal{H}}^n$-image counted with multiplicity, the other pulls an $n$-dimensional region apart into $(n-m)$-dimensional fibres weighted by ${\mathcal{L}}^m$. Putting these together, the area formula generalises the change-of-variables theorem and the definition of surface area, while the coarea formula generalises Fubini's theorem and supplies the slicing identity; this is exactly the pair that defines the mass of a rectifiable current and governs the slicing of currents in 02.13.07 and 02.13.11. The bridge is that rectifiability — covering a set up to measure zero by Lipschitz images — lets the area formula assign a well-defined ${\mathcal{H}}^k$ to objects with no smooth structure, and the coarea formula then decomposes those objects fibre by fibre. The same pattern recurs in the variational theory: the coarea reduction of the Sobolev inequality to the isoperimetric inequality, and the eikonal level-set identities, all build toward the regularity theory of minimal surfaces, where these formulas appear again as the basic accounting of area and its variations.

Full proof set Master

Proposition 1 (linear coarea formula). Let $L:{\mathbb{R}}^n\to {\mathbb{R}}^m$ be linear and surjective with $m\le n$. Then for every ${\mathcal{L}}^n$-measurable $A\subseteq {\mathbb{R}}^n$, $$ \int_{\mathbb R^{m}}\mathcal H^{n-m}\big(A\cap L^{-1}(y)\big),d\mathcal L^{m}(y)=JL\cdot\mathcal L^{n}(A). $$

Proof. By the singular-value decomposition write $L=U\Sigma V^T$ with $U\in O(m)$, $V\in O(n)$, and $\Sigma$ the $m\times n$ matrix carrying singular values ${\sigma }_1,\dots ,{\sigma }_m>0$ (surjectivity forces them positive). Orthogonal changes of variable preserve both ${\mathcal{L}}^n$ and ${\mathcal{H}}^{n-m}$ and leave $JL$ unchanged, so we reduce to $L=\Sigma$, which in suitable coordinates is the weighted projection $(\xi ,\eta )\mapsto ({\sigma }_1{\xi }_1,\dots ,{\sigma }_m{\xi }_m)$ for $\xi \in {\mathbb{R}}^m$, $\eta \in {\mathbb{R}}^{n-m}$. The fibre over $y$ is the affine subspace $\{(\xi (y),\eta ):\eta \in {\mathbb{R}}^{n-m}\}$ with ${\xi }_i(y)=y_i/{\sigma }_i$, on which ${\mathcal{H}}^{n-m}$ is ordinary $(n-m)$-Lebesgue measure. By Fubini's theorem 02.07.07 applied in the split coordinates, $$ \mathcal L^{n}(A)=\int_{\mathbb R^{m}}\mathcal L^{n-m}\big(A\cap{\xi=\xi(y/\sigma)}\big),d\xi. $$ Substituting $y=\Sigma \xi$, so $d{\mathcal{L}}^m(y)={\sigma }_1\cdots {\sigma }_{m}d\xi =JLd\xi$, converts the $\xi$-integral into a $y$-integral against $d{\mathcal{L}}^m(y)/JL$, which rearranges to the claimed identity. $\square$

Proposition 2 (Jacobian equals product of singular values). For linear $L:{\mathbb{R}}^n\to {\mathbb{R}}^m$ with $k=\min (n,m)$ and singular values ${\sigma }_1,\dots ,{\sigma }_k$, $JL={\sigma }_1\cdots {\sigma }_k$.

Proof. Suppose $m\le n$, so $JL=\sqrt{\det (L{L}^T)}$. The matrix $L{L}^T$ is the $m\times m$ Gram matrix of the rows of $L$; its eigenvalues are ${\sigma }_1^2,\dots ,{\sigma }_m^2$ by the singular-value decomposition $L=U\Sigma V^T$, since $L{L}^T=U\Sigma {\Sigma }^T{U}^T$ and $\Sigma {\Sigma }^T=\mathrm{diag}({\sigma }_1^2,\dots ,{\sigma }_m^2)$. Hence $\det (L{L}^T)={\sigma }_1^2\cdots {\sigma }_m^2$ and $JL={\sigma }_1\cdots {\sigma }_m$. The case $n\le m$ is identical with $L^{T}L$ in place of $L{L}^T$. $\square$

Proposition 3 (Lipschitz images do not increase Hausdorff measure). If $f:{\mathbb{R}}^n\to {\mathbb{R}}^m$ is $L$-Lipschitz and $E\subseteq {\mathbb{R}}^n$, then ${\mathcal{H}}^k(f(E))\le L^k{\mathcal{H}}^k(E)$ for every $k\ge 0$.

Proof. Let $\{U_i\}$ cover $E$ with $\mathrm{diam}{U}_i<\delta$. Then $\{f(U_i)\}$ covers $f(E)$ with $\mathrm{diam}f(U_i)\le L\mathrm{diam}{U}_i<L\delta$, so $$ \mathcal H^{k}{L\delta}(f(E))\le\sum{i}(\operatorname{diam}f(U_{i}))^{k}\le L^{k}\sum_{i}(\operatorname{diam}U_{i})^{k}. $$ Taking the infimum over covers gives ${\mathcal{H}}_{L\delta }^k(f(E))\le L^k{\mathcal{H}}_{\delta }^k(E)$, and letting $\delta \to 0^{+}$ yields the claim. This estimate is the quantitative reason the area formula's right side is finite whenever the left side is, and it is the first step in localising the proof. $\square$

Proposition 4 (coarea Jacobian vanishes on the low-rank set). Let $f:{\mathbb{R}}^n\to {\mathbb{R}}^m$ be Lipschitz with $m\le n$, and let $Z=\{x:\mathrm{rank}Df(x)<m\}$. Then $Jf=0$ on $Z$ and the fibres contribute via ${\int }_{Z}Jfd{\mathcal{L}}^n=0$.

Proof. At a point of differentiability in $Z$ the linear map $Df(x)$ is not surjective, so its smallest singular value ${\sigma }_m$ is $0$. By Proposition 2, $Jf(x)={\sigma }_1\cdots {\sigma }_m=0$. Hence the integrand $Jf$ vanishes identically on $Z\cap \{x:f\text{ differentiable at }x\}$, and by Rademacher 02.07.10 the complementary set is ${\mathcal{L}}^n$-null, so ${\int }_{Z}Jfd{\mathcal{L}}^n=0$. The coarea formula therefore receives no contribution from the critical set, consistent with the Sard-type Exercise 5. $\square$

Connections Master

• Rademacher's theorem 02.07.10. The area and coarea formulas are stated for Lipschitz maps, which possess a derivative only almost everywhere. Rademacher's theorem is the hinge that supplies that derivative: it guarantees $Df(x)$ exists ${\mathcal{L}}^n$-a.e., so the Jacobian $Jf$ is defined off a null set and the localisation-by-linearisation proof strategy can proceed. Without a.e. differentiability there would be no integrand on the left side of either formula.

• Lebesgue outer measure, Carathéodory, and Hausdorff measure 02.07.02. The target side of both formulas is measured by Hausdorff measure ${\mathcal{H}}^k$, constructed there by the Carathéodory method, and the identification ${\mathcal{H}}^n={\mathcal{L}}^n$ on ${\mathbb{R}}^n$ (via the isodiametric inequality) is what lets the linear area formula compare source and image measures. The whole apparatus of cover-and-infimum measures built in that unit is the measure-theoretic ground on which area and coarea stand.

• Fubini-Tonelli and product measures 02.07.07. The coarea formula is the nonlinear generalisation of Fubini's theorem: when $f$ is a coordinate projection the Jacobian is $1$ and the formula reduces exactly to integrating a function by slicing the product into fibres. The linear coarea proof in the full proof set runs through Fubini in the split singular-value coordinates, so Fubini is both the special case and a step in the general proof.

• Rectifiable currents and the mass functional 02.13.07. The area formula is the definition that makes the mass of a rectifiable current well-posed: a $k$-rectifiable set is parametrised by Lipschitz images, and the area formula assigns it the ${\mathcal{H}}^k$-measure that becomes the current's mass. The currents unit invokes "by the area formula" precisely at this point, and this unit supplies the standalone statement and proof it relies on.

• Slicing of currents 02.13.11. The coarea formula is the engine of slicing: the level-set decomposition $A={\bigcup }_{t}A\cap f^{-1}(t)$ with the mass estimate $\int \mathbf{M}(⟨T,f,t⟩)dt\le (\mathrm{Lip}f)\mathbf{M}(T)$ is the current-theoretic form of ${\int }_{A}Jf=\int {\mathcal{H}}^{n-m}(A\cap f^{-1}(y))d{\mathcal{L}}^m(y)$. The slicing unit cites the coarea formula to bound and reconstruct slices; this unit is where that formula is established.

Historical & philosophical context Master

The area formula has a long prehistory in the attempt to define surface area without smoothness. Lebesgue himself struggled with the right notion of the area of a non-smooth surface, and the subject was clouded by examples (the Schwarz lantern) showing that naive polyhedral approximation can make the "area" of a cylinder arbitrarily large. Eilenberg's 1938 work on $\phi$-measures ^{[Eilenberg 1938]} introduced the integral-geometric idea of measuring a set by integrating the count of its intersections with the fibres of a map, the seed of the multiplicity integrand $\#(A\cap f^{-1}(y))$ that appears on the right of the area formula. On the coarea side, Kronrod's 1950 study of functions of two variables ^{[Kronrod 1950]} analysed the structure of level sets and is the recognised precursor of the level-set decomposition.

The decisive synthesis is due to Herbert Federer. His 1959 paper Curvature measures ^{[Federer 1959]} and then the encyclopedic 1969 treatise Geometric Measure Theory ^{[Federer 1969]} established both formulas in full generality for Lipschitz maps between Euclidean spaces, with the Jacobian defined through the singular-value/Gram-determinant machinery and the proof organised around Rademacher's theorem and the approximation of Lipschitz maps by $C^1$ pieces. The name "coarea" is Federer's, chosen to signal the duality with the area formula: where area integrates forward over the image, coarea integrates over the family of fibres. Federer's treatment placed these formulas at the foundation of the theory of currents and rectifiable sets developed jointly with Wendell Fleming, and they became the standard computational tools of the field.

Philosophically, the two formulas mark a shift in what counts as a geometric object worthy of measurement. Classical differential geometry required smoothness to define area and to integrate over submanifolds; the area and coarea formulas show that the far weaker condition of being a Lipschitz image — equivalently, rectifiability up to a null set — already suffices. This is the technical content of the idea that "almost smooth" objects can carry a genuine theory of area and integration, an idea that runs through the Plateau problem, the calculus of variations, and the regularity theory of minimal surfaces. The modern expositions of Simon ^{[Simon 1983]} and Maggi ^{[Maggi 2012]} present the coarea formula as the bridge from the analysis of Sobolev and BV functions to the geometry of their level sets, completing the arc from Lebesgue's worry about surface area to a complete non-smooth integration theory.

Bibliography Master

@book{Federer1969,
  author    = {Federer, Herbert},
  title     = {Geometric Measure Theory},
  series    = {Grundlehren der mathematischen Wissenschaften},
  volume    = {153},
  publisher = {Springer-Verlag},
  address   = {Berlin},
  year      = {1969},
  note      = {Area formula \S3.2.3, coarea formula \S3.2.11}
}

@article{Federer1959,
  author    = {Federer, Herbert},
  title     = {Curvature measures},
  journal   = {Transactions of the American Mathematical Society},
  volume    = {93},
  pages     = {418--491},
  year      = {1959}
}

@book{EvansGariepy2015,
  author    = {Evans, Lawrence C. and Gariepy, Ronald F.},
  title     = {Measure Theory and Fine Properties of Functions},
  edition   = {Revised},
  series    = {Textbooks in Mathematics},
  publisher = {CRC Press},
  address   = {Boca Raton},
  year      = {2015},
  note      = {Chapter 3, area and coarea formulas}
}

@article{Eilenberg1938,
  author    = {Eilenberg, Samuel},
  title     = {On {$\varphi$}-measures},
  journal   = {Annales de la Soci\'et\'e Polonaise de Math\'ematique},
  volume    = {17},
  pages     = {252--253},
  year      = {1938}
}

@article{Kronrod1950,
  author    = {Kronrod, Alexander S.},
  title     = {On functions of two variables},
  journal   = {Uspekhi Matematicheskikh Nauk},
  volume    = {5},
  number    = {1},
  pages     = {24--134},
  year      = {1950}
}

@book{Simon1983,
  author    = {Simon, Leon},
  title     = {Lectures on Geometric Measure Theory},
  series    = {Proceedings of the Centre for Mathematical Analysis},
  volume    = {3},
  publisher = {Australian National University},
  address   = {Canberra},
  year      = {1983}
}

@book{Maggi2012,
  author    = {Maggi, Francesco},
  title     = {Sets of Finite Perimeter and Geometric Variational Problems},
  series    = {Cambridge Studies in Advanced Mathematics},
  volume    = {135},
  publisher = {Cambridge University Press},
  address   = {Cambridge},
  year      = {2012}
}

@book{KrantzParks2008,
  author    = {Krantz, Steven G. and Parks, Harold R.},
  title     = {Geometric Integration Theory},
  series    = {Cornerstones},
  publisher = {Birkh\"auser},
  address   = {Boston},
  year      = {2008}
}