Jacobi inversion theorem

Intuition [Beginner]

Take a compact Riemann surface $X$ — a closed surface dressed with the data needed to ask whether a function is holomorphic — of genus $g$. The Jacobian of $X$ is a $g$-dimensional complex torus $\mathrm{Jac}(X)$ that records the integer data of how holomorphic 1-forms integrate around closed loops. Each unordered tuple of $g$ points on $X$ produces a point of $\mathrm{Jac}(X)$ by integrating the basis 1-forms from a fixed reference point along paths to the chosen $g$ points; the result, taken modulo the period lattice, is a well-defined point of the torus.

The Jacobi inversion theorem says this map is essentially a bijection. Every point of $\mathrm{Jac}(X)$ comes from some unordered tuple of $g$ points on $X$, and a generic point comes from exactly one such tuple. The map from "unordered $g$-tuples on $X$" to "points of the Jacobian" is birational — surjective, and one-to-one off a thin exceptional set.

The exceptional set is the theta divisor, the locus where two $g$-tuples produce the same Jacobian point. Riemann showed this divisor coincides with the zero locus of a remarkable function — the theta function — and so the geometry of inversion is governed by an explicit transcendental object on the torus.

Visual [Beginner]

A schematic of a compact Riemann surface of genus $g=2$ on the left, with two marked $2$-tuples of points $D=p_1+p_2$ and $D^{\prime }=p_1^{\prime }+p_2^{\prime }$. An arrow labelled "Abel-Jacobi" points to a $2$-dimensional complex torus on the right, showing both $D$ and $D^{\prime }$ landing at distinct points; a third $2$-tuple on the left is shown collapsing to the same Jacobian point as the first, illustrating the exceptional theta-divisor locus.

Worked example [Beginner]

Take an elliptic curve $X=\mathbb{C}/\Lambda$ with $\Lambda =\mathbb{Z}\oplus \mathbb{Z}\tau$, $\mathrm{Im}\tau >0$. The genus is $g=1$. The Jacobian is also $\mathbb{C}/\Lambda$, the same complex torus.

The unordered $1$-tuple $D=p$ is just a single point of $X$. Pick the reference point $p_0=0\in X$. The Abel-Jacobi map sends $p$ to the integral of the basis 1-form $dz$ along a path from $0$ to $p$, modulo the period lattice $\Lambda$; the result is just $p$ itself modulo $\Lambda$. Both $X$ and $\mathrm{Jac}(X)$ are the torus $\mathbb{C}/\Lambda$, and the Abel-Jacobi map is the identity on the torus.

Concretely, if $p=0.4+0.3\tau$ modulo $\Lambda$, the integral of $dz$ from $0$ to $p$ along the straight line is $0.4+0.3\tau$, and modulo $\Lambda$ this is the same point of the Jacobian. The map is one-to-one and onto: every Jacobian point comes from exactly one point of $X$. The theta divisor is a single point of $\mathrm{Jac}(X)$ — the image of no proper effective divisor of degree $0$.

What this tells us: in the simplest case, Jacobi inversion says the Riemann surface and its Jacobian are the same complex torus, with the Abel-Jacobi map the identity. For genus $g\ge 2$ the map is from ${\mathrm{Sym}}^g(X)$, which has the same dimension as $\mathrm{Jac}(X)$, and the map is birational rather than bijective.

Check your understanding [Beginner]

Formal definition [Intermediate+]

Let $X$ be a smooth projective compact Riemann surface of genus $g\ge 1$. Choose a basis ${\omega }_1,\dots ,{\omega }_g$ of the space $H^0(X,{\Omega }_X^1)$ of holomorphic 1-forms (dimension $g$ by 06.04.03) and a symplectic basis $a_1,\dots ,a_g,b_1,\dots ,b_g$ of $H_1(X,\mathbb{Z})$ with $a_i\cdot a_j=b_i\cdot b_j=0$, $a_i\cdot b_j={\delta }_{ij}$. The period matrix is $\Pi =(A\mid B)\in M_{g\times 2g}(\mathbb{C})$ with $A_{ij}={\int }_{a_j}{\omega }_i$ and $B_{ij}={\int }_{b_j}{\omega }_i$; the period lattice is $\Lambda =A{\mathbb{Z}}^g+B{\mathbb{Z}}^g\subset {\mathbb{C}}^g$. The Jacobian variety (cf. 06.06.03) is the complex torus $$ \mathrm{Jac}(X) := \mathbb{C}^g / \Lambda, $$ of complex dimension $g$.

Let ${\mathrm{Sym}}^g(X):=X^g/S_g$ be the $g$-th symmetric product, parametrising unordered $g$-tuples of points of $X$ (equivalently, effective divisors of degree $g$ on $X$). It is a smooth projective complex variety of dimension $g$. Fix a reference point $p_0\in X$. The Abel-Jacobi map is $$ \alpha : \mathrm{Sym}^g(X) \to \mathrm{Jac}(X), \qquad \alpha(p_1 + \cdots + p_g) = \left( \sum_{i=1}^g \int_{p_0}^{p_i} \omega_j \right)_{j=1}^g \mod \Lambda, $$ where each integral is taken along an arbitrary path from $p_0$ to $p_i$; different path choices change the integral by an element of $\Lambda$, so the value modulo $\Lambda$ is well-defined. The map $\alpha$ is holomorphic.

Theorem (Jacobi inversion, Jacobi 1834). The Abel-Jacobi map $\alpha :{\mathrm{Sym}}^g(X)\to \mathrm{Jac}(X)$ is surjective and birational: $\alpha$ is rational and generically one-to-one. Equivalently, every point of $\mathrm{Jac}(X)$ is the image under $\alpha$ of some effective divisor of degree $g$, and a generic point has a unique preimage.

Riemann's refinement (1857). The exceptional locus of $\alpha$ — points of $\mathrm{Jac}(X)$ with non-unique preimage — coincides, after a translation by a point $\kappa$ called the Riemann constant, with the theta divisor $\Theta =\alpha (W_{g-1})\subset \mathrm{Jac}(X)$, where $W_{g-1}$ is the image under the analogous degree-$(g-1)$ Abel-Jacobi map of ${\mathrm{Sym}}^{g-1}(X)$.

Equivalent forms.

• Cohomological reformulation: ${\alpha }^{-1}(z)=\{D\in {\mathrm{Sym}}^g(X):D-g{p}_0{\sim }_{\mathrm{lin}}z\}$ where ${\sim }_{\mathrm{lin}}$ is linear equivalence; for generic $z$ this is a single divisor.
• Symmetric-product factorisation: the composition ${\mathrm{Sym}}^g(X)\stackrel{\alpha }{\to }\mathrm{Jac}(X)\stackrel{\sim }{\to }{\mathrm{Pic}}^g(X)$ identifies effective divisors of degree $g$ with their isomorphism class as line bundles, $D\mapsto \mathcal{O}(D)$, on the locus where $\dim H^0(\mathcal{O}(D))=1$.
• Theta-vanishing: the locus where $\theta (\alpha (D)-\kappa )$ vanishes coincides with the theta divisor (Riemann's vanishing theorem).

Counterexamples to common slips.

• The Abel-Jacobi map in the inversion theorem uses ${\mathrm{Sym}}^g(X)$, not $X$: at $g=1$ they coincide, but at $g\ge 2$ confusing them collapses the dimension count.
• "Birational" is weaker than "isomorphism": the map $\alpha$ is not an isomorphism for $g\ge 2$ — it contracts a positive-codimension exceptional set to the theta divisor — but it is rational and generically one-to-one, which suffices to read off cohomological invariants.
• The reference point $p_0$ matters for the map but not for the existence of the inversion: changing $p_0$ to $p_0^{\prime }$ translates $\alpha$ by the constant $g{\int }_{p_0}^{p_0^{\prime }}\omega \mathrm{mod}\Lambda$, preserving surjectivity and birationality.

Key theorem with proof [Intermediate+]

Theorem (Jacobi 1834, Jacobi inversion). Let $X$ be a smooth projective compact Riemann surface of genus $g\ge 1$, and fix a reference point $p_0\in X$. The Abel-Jacobi map $$ \alpha : \mathrm{Sym}^g(X) \to \mathrm{Jac}(X), \qquad \alpha(p_1 + \cdots + p_g) = \left(\sum_{i=1}^g \int_{p_0}^{p_i} \omega_j\right)_{j=1}^g \mod \Lambda, $$ is surjective and birational.

Proof. The argument has four steps: image is closed (compactness), image has dimension $g$ (Riemann-Roch + Serre duality computation of the differential), image is the full Jacobian (closed plus full-dimensional in a connected target), and generic fibre is a single point (Riemann-Roch on a sufficiently general line bundle of degree $g$).

Step 1 — image is closed. The symmetric product ${\mathrm{Sym}}^g(X)$ is the quotient of the smooth projective variety $X^g$ by the action of the symmetric group $S_g$, hence is a smooth projective variety of dimension $g$ over $\mathbb{C}$. In particular ${\mathrm{Sym}}^g(X)$ is compact. The map $\alpha$ is holomorphic (the integrals depend holomorphically on endpoints, the symmetrisation is a polynomial operation in the coordinates). A holomorphic map from a compact complex manifold to a complex manifold has closed image, so $\alpha ({\mathrm{Sym}}^g(X))\subset \mathrm{Jac}(X)$ is closed.

Step 2 — image has dimension $g$. Compute the differential $d\alpha {|}_D$ at a generic effective divisor $D=p_1+\cdots +p_g$ with the $p_i$ distinct and not zeroes of any non-zero holomorphic 1-form. The tangent space $T_D{\mathrm{Sym}}^g(X)$ is identified with the direct sum $T_{p_1}X\oplus \cdots \oplus T_{p_g}X$ modulo permutation, of complex dimension $g$. Choose a local coordinate $z_i$ near each $p_i$. A tangent vector at $D$ is a $g$-tuple $(v_1,\dots ,v_g)$ of complex numbers, and the differential $d\alpha {|}_D$ sends it to $$ d\alpha|D(v_1, \ldots, v_g) = \left( \sum{i=1}^g \omega_j(p_i) \cdot v_i \right){j=1}^g \in T{\alpha(D)} \mathrm{Jac}(X) = \mathbb{C}^g, $$ where ${\omega }_j(p_i)$ denotes the local coefficient of ${\omega }_j$ in the chosen coordinate $z_i$. The matrix of $d\alpha {|}_D$ is the $g\times g$ matrix $M(D{)}_{ji}:={\omega }_j(p_i)$, the Brill-Noether matrix of $D$. By Riemann-Roch 06.04.01 together with Serre duality 06.04.04, for a generic $D$ of degree $g$, $\dim H^0(X,\mathcal{O}(D))=1$ (the only global section is the constant section after multiplication by a fixed defining section, equivalently $D$ is non-special), hence $\dim H^0(X,K_X\otimes \mathcal{O}(-D))=h^1(\mathcal{O}(D))=h^0(\mathcal{O}(D))-(\mathrm{deg}D+1-g)=1-(g+1-g)=0$. The vanishing $H^0(K_X\otimes \mathcal{O}(-D))=0$ says no non-zero holomorphic 1-form vanishes at every $p_i$, equivalently the $p_i$ impose independent conditions on $H^0(K_X)$, equivalently the matrix $M(D)$ has rank $g$. Hence $d\alpha {|}_D$ is an isomorphism for generic $D$, and the image of $\alpha$ has dimension $g$.

Step 3 — image is the full Jacobian. The image $\alpha ({\mathrm{Sym}}^g(X))$ is closed (Step 1), has dimension $g$ (Step 2), and is contained in the connected complex torus $\mathrm{Jac}(X)$ of dimension $g$. A closed connected analytic subvariety of full dimension in a connected complex manifold is the whole manifold (a closed analytic subset of dimension $g$ in a complex manifold of dimension $g$ is open in the smooth locus, hence open in the manifold, hence the whole connected component). Therefore $\alpha ({\mathrm{Sym}}^g(X))=\mathrm{Jac}(X)$, proving surjectivity.

Step 4 — generic fibre is a single point. For $z\in \mathrm{Jac}(X)$ a generic point, the fibre ${\alpha }^{-1}(z)\subset {\mathrm{Sym}}^g(X)$ consists of effective divisors $D$ of degree $g$ such that $D-g{p}_0$ is mapped to $z$ in the Jacobian, equivalently $\mathcal{O}(D)$ is a line bundle in the isomorphism class corresponding to $z$ under the identification $\mathrm{Jac}(X)\cong {\mathrm{Pic}}^g(X)$ (fixed by translating by $g{p}_0$). For a generic line bundle $L$ of degree $g$, Riemann-Roch and Serre duality give $\dim H^0(X,L)=\mathrm{deg}L+1-g+\dim H^0(X,K_X\otimes L^{-1})=1+\dim H^0(K_X\otimes L^{-1})$; for generic $L$, $K_X\otimes L^{-1}$ has degree $2g-2-g=g-2$ and is non-special with $\dim H^0(K_X\otimes L^{-1})=0$ for generic $L$ when $g\ge 1$, so $\dim H^0(X,L)=1$. The unique (up to scalar) non-zero global section of $L$ has a unique zero divisor, an effective divisor $D$ of degree $g$ with $\mathcal{O}(D)\cong L$. Therefore the fibre ${\alpha }^{-1}(z)$ consists of a single point for generic $z$, proving birationality. $\square$

The four-step structure follows Donaldson §11–§12; Forster §21 reorganises around the proof of Abel's theorem first and inverts the surjectivity question via the symmetric-product compactness argument; Griffiths-Harris §2.7 gives the same proof inside the broader Brill-Noether discussion. The underlying content — closure, dimension count via Riemann-Roch + Serre duality, full-dimensional closed subvariety, generic-fibre count — is identical.

Bridge. The inversion proven here, combined with 06.06.04 Abel's theorem (kernel of Abel-Jacobi is principal divisors), upgrades the Abel-Jacobi map to a structural identification: the symmetric-product / linear-equivalence quotient ${\mathrm{Sym}}^g(X)/{\sim }_{\mathrm{lin}}$ is birational to the Jacobian $\mathrm{Jac}(X)\cong {\mathrm{Pic}}^g(X)$, with exceptional locus on both sides controlled by the theta divisor. Riemann's 1857 refinement, the Riemann vanishing theorem, identifies the exceptional locus with the zero set of the Riemann theta function (cf. 06.06.05) shifted by the Riemann constant, providing an explicit transcendental description of the Brill-Noether locus $W_{g-1}\subset {\mathrm{Pic}}^{g-1}(X)$. Combined with 06.04.04 Serre duality, the same machinery extends to higher-codimension Brill-Noether loci $W_d^r$, the special-divisor stratification of the Picard variety. The principal-polarisation structure of $\mathrm{Jac}(X)$ from 06.04.03 Hodge decomposition ($A{B}^T=B{A}^T$, $\mathrm{Im}(B{A}^{-1})>0$) makes $\mathrm{Jac}(X)$ a principally polarised abelian variety, and Riemann's vanishing theorem identifies the principal polarisation with the theta divisor — this is the input data for the Schottky problem (which principally polarised abelian varieties arise as Jacobians) and for Torelli's theorem (a smooth projective curve is determined by its Jacobian as a principally polarised abelian variety). Putting these together, the foundational insight is that on a smooth projective curve every point of the Jacobian is captured by an effective divisor of the minimal degree $g$ — and the bridge is the Riemann-Roch + Serre-duality dimension count that makes the Abel-Jacobi differential a generic isomorphism.

Exercises [Intermediate+]

Lean formalization [Intermediate+]

Mathlib does not currently formalise the symmetric power ${\mathrm{Sym}}^g(X)$ of a smooth projective curve, the Abel-Jacobi map $\alpha :{\mathrm{Sym}}^g(X)\to \mathrm{Jac}(X)$, or the theta divisor $\Theta \subset \mathrm{Jac}(X)$ as first-class objects. A proposed signature, in Lean 4 / Mathlib syntax, sketching the target statement:

[object Promise]

The proof depends on names that do not currently exist in Mathlib (the symmetric power of a smooth projective variety as a smooth projective variety, the Abel-Jacobi morphism as a holomorphic map of complex tori, the Riemann-Roch + Serre-duality dimension count for the differential, and the generic-fibre argument via Brill-Noether non-speciality). Each is a candidate Mathlib contribution; until then this unit ships with lean_status: none.

Advanced results [Master]

The Jacobi inversion theorem is the dimension-$g$ case of a chain of identifications connecting the Picard variety of a smooth projective curve to its symmetric powers. The general formulation, due to Jacobi 1834 for the existence-and-surjectivity statement and refined by Riemann 1857 to the theta-divisor / birationality description, asserts that for a smooth projective curve $X$ of genus $g$ over an algebraically closed field $k=\mathbb{C}$ and every $d\ge g$, the Abel-Jacobi map ${\alpha }_d:{\mathrm{Sym}}^d(X)\to {\mathrm{Pic}}^d(X)$ is surjective with generic fibre ${\mathbb{P}}^{d-g}$, and birational at $d=g$. The proof in every dimension passes through the Riemann-Roch + Serre-duality computation of the differential and the closed-image / dimension-count argument.

Riemann's vanishing theorem. Let $\theta (z;\Pi )={\sum }_{n\in {\mathbb{Z}}^g}\exp (\pi i{n}^T\Pi n+2\pi i{n}^{T}z)$ be the Riemann theta function on ${\mathbb{C}}^g$, where $\Pi \in {\mathfrak{H}}_g$ is the period matrix in normalised form $\Pi =(I_g\mid \tau )$, $\tau \in {\mathfrak{H}}_g$. The function $\theta$ is holomorphic on ${\mathbb{C}}^g$ with quasi-periodicity $\theta (z+n+\tau m;\tau )=\exp (-\pi i{m}^T\tau m-2\pi i{m}^{T}z)\theta (z;\tau )$ for $n,m\in {\mathbb{Z}}^g$, hence its zero locus descends to a divisor ${\Theta }_0\subset \mathrm{Jac}(X)$. Riemann's vanishing theorem (Riemann 1857) asserts the existence of a Riemann constant $\kappa \in \mathrm{Jac}(X)$ such that the translated zero divisor $\Theta :={\Theta }_0+\kappa$ coincides with the image ${\alpha }_{g-1}({\mathrm{Sym}}^{g-1}(X))\subset {\mathrm{Pic}}^{g-1}(X)\cong \mathrm{Jac}(X)$ (after fixing the reference point). The constant $\kappa$ is half the canonical divisor class shifted by combinatorial data of the symplectic basis. The vanishing locus of $\theta$ is therefore the theta divisor, the exceptional locus of the Abel-Jacobi map for degree-$(g-1)$ divisors.

Brill-Noether stratification. For each pair $(d,r)$ with $d\ge 0$ and $r\ge 0$, define $$ W^r_d := { L \in \mathrm{Pic}^d(X) : \dim H^0(X, L) \geq r + 1 } \subset \mathrm{Pic}^d(X), $$ a closed subvariety. The Brill-Noether existence theorem (Kempf 1971, Kleiman-Laksov 1972, building on Severi 1915) asserts that whenever $\rho :=g-(r+1)(g-d+r)\ge 0$, the locus $W_d^r$ is non-empty of dimension $\ge \rho$. Griffiths-Harris 1980 On the variety of special linear systems on a general algebraic curve and Gieseker 1982 proved that on a general curve $X$, $\dim W_d^r=\rho$ when $\rho \ge 0$ and $W_d^r=\varnothing$ when $\rho <0$ — the Brill-Noether dimension theorem. The Jacobi inversion theorem corresponds to the case $(d,r)=(g,0)$, where $\rho =g$ and $W_g^0={\mathrm{Pic}}^g(X)$ has dimension $g$, recovering the surjectivity statement; the next case $(d,r)=(g-1,0)$ corresponds to the theta divisor $\Theta =W_{g-1}^0$ of codimension $1$.

Torelli theorem. A smooth projective curve $X$ of genus $g\ge 1$ over $\mathbb{C}$ is determined up to isomorphism by its principally polarised Jacobian $(\mathrm{Jac}(X),\Theta )$ — Torelli's theorem (Torelli 1913). The proof, refined by Andreotti 1958 and Andreotti-Mayer 1967, uses the geometry of the theta divisor: for $g\ge 4$, the singular locus of $\Theta$ recovers the canonical model of $X$. For $g=2,3$ a separate hyperelliptic-aware argument is required. Torelli's theorem makes the Jacobian map $X\mapsto (\mathrm{Jac}(X),\Theta )$ from the moduli space ${\mathcal{M}}_g$ of smooth genus-$g$ curves to the moduli space ${\mathcal{A}}_g$ of principally polarised abelian varieties an injection.

Schottky problem. Identify the image of the Jacobian map ${\mathcal{M}}_g\to {\mathcal{A}}_g$ — equivalently, characterise which principally polarised abelian varieties of dimension $g$ are Jacobians of curves. For $g\le 3$ the map is a birational equivalence (the Jacobi locus has the right dimension to fill ${\mathcal{A}}_g$). For $g=4$ Schottky 1888 gave a single explicit modular relation that cuts out the Jacobi locus inside ${\mathcal{A}}_4$. For $g\ge 5$ several characterisations are known: trisecant identities (Welters 1984, Krichever 2006 establishing the Welters trisecant conjecture), the Andreotti-Mayer locus of abelian varieties whose theta divisor has unexpectedly large singular locus, and Novikov's conjecture (proved by Shiota 1986) that the KP equation characterises Jacobians via solutions of the form $u(x,y,t)=2{\partial }_x^2\log \theta (xV+yU+tW+z_0;\Pi )$.

Higher symmetric powers. For $d>g$, the Abel-Jacobi map ${\alpha }_d:{\mathrm{Sym}}^d(X)\to {\mathrm{Pic}}^d(X)$ has fibre ${\alpha }_d^{-1}(L)=|L|\cong {\mathbb{P}}^{h^0(L)-1}$ over a line bundle $L$ of degree $d$ — the linear system of effective divisors in the class of $L$. By Riemann-Roch, $h^0(L)=d+1-g$ for generic $L$ when $d>g$ (non-special line bundles), so the generic fibre is ${\mathbb{P}}^{d-g}$. The map ${\alpha }_d$ exhibits ${\mathrm{Sym}}^d(X)$ as a projective bundle over ${\mathrm{Pic}}^d(X)$ on the locus of non-special $L$, with the Brill-Noether loci $W_d^r$ for $r\ge 1$ as the locus where the fibre dimension jumps. For $d<g$ the situation is reversed: the map ${\alpha }_d$ is generically injective with image $W_d\subset {\mathrm{Pic}}^d(X)$, a proper closed subvariety; the codimension is $g-d$.

Synthesis. The Jacobi inversion theorem is the engine that converts the Abel-Jacobi map from a holomorphic map into a birational equivalence: combined with Abel's theorem 06.06.04 (kernel = principal divisors), the inversion gives the structural identification ${\mathrm{Pic}}^g(X)\cong \mathrm{Jac}(X)$ and through it the principal-polarisation structure of the Jacobian. Read in the opposite direction, every point of the Jacobian is captured by a unique-up-to-linear-equivalence effective divisor of degree $g$ — the inverse map of the inversion theorem realises the Jacobian as the moduli of effective divisors of minimal degree. The geometric content of "specialty" — the Brill-Noether stratification of the Picard variety — is exactly the locus where the inversion fails to be an isomorphism, organised by the dimensions of linear systems via Riemann-Roch and Serre duality. The Riemann theta function provides the explicit transcendental description of the theta divisor and, through the Krichever / Welters correspondence with the KP hierarchy, links the geometry of curves to integrable systems. Putting these together, the cohomology of every line bundle on a smooth projective curve, the symmetric powers ${\mathrm{Sym}}^d(X)$, the Picard variety $\mathrm{Pic}(X)$, the Jacobian $\mathrm{Jac}(X)$, the theta divisor, and the Brill-Noether stratification together form a single connected structure that the Jacobi inversion theorem organises; and the dimension-$g$ inversion is the foundational identification on which the rest of the Jacobian-side theory of curves rests.

Full proof set [Master]

Lemma (symmetric product as a smooth projective variety). For a smooth projective complex curve $X$ of genus $g$ and every $d\ge 1$, the symmetric quotient ${\mathrm{Sym}}^d(X):=X^d/S_d$ is a smooth projective complex variety of dimension $d$.

Proof. The group $S_d$ acts on the smooth projective variety $X^d$ by permutation of coordinates; the action is free outside the big diagonal $\Delta \subset X^d$ where two coordinates coincide. On the open complement $X^d\setminus \Delta$ the quotient is smooth. To extend smoothness across $\Delta$, observe that $X$ is a smooth projective complex variety of dimension $1$, so the local model of $X^d$ near a point $(p,p,p,\dots )$ is ${\mathbb{C}}^d$ with $S_d$ acting by permutation; the quotient ${\mathbb{C}}^d/S_d\cong {\mathbb{C}}^d$ via the elementary symmetric polynomials, which are a system of regular coordinates on the quotient. Hence ${\mathrm{Sym}}^d(X)$ is smooth at every point and inherits projectivity from $X^d$ via the Plücker / Veronese embedding of the symmetric quotient. The dimension-$1$ feature is essential: for higher-dimensional $X$ the symmetric quotient is in general not smooth (resolutions are required, e.g. the Hilbert scheme ${\mathrm{Hilb}}^d(X)$ for $\dim X\ge 2$). $\square$

Lemma (Abel-Jacobi map is well-defined and holomorphic). Fix a reference point $p_0\in X$. The Abel-Jacobi map $\alpha :{\mathrm{Sym}}^g(X)\to \mathrm{Jac}(X)$ defined by $\alpha (p_1+\cdots +p_g)={\left({\sum }_i{\int }_{p_0}^{p_i}{\omega }_j\right)}_j\mathrm{mod}\Lambda$ is independent of path choices and is a holomorphic map of complex manifolds.

Proof. Two paths ${\gamma }_i,{\gamma }_i^{\prime }$ from $p_0$ to $p_i$ differ by a closed loop ${\delta }_i\in H_1(X,\mathbb{Z})$, and ${\int }_{{\delta }_i}{\omega }_j$ lies in the period lattice $\Lambda =A{\mathbb{Z}}^g+B{\mathbb{Z}}^g$ by definition of the periods. So the value modulo $\Lambda$ is independent of path. The integrand ${\omega }_j$ is a holomorphic 1-form, so the integral ${\int }_{p_0}^{p_i}{\omega }_j$ depends holomorphically on $p_i$ (locally — globally it is multivalued by $\Lambda$). After symmetrisation in the $p_i$ and quotient by $\Lambda$ in the target, the resulting map is well-defined and holomorphic on the smooth projective variety ${\mathrm{Sym}}^g(X)$. $\square$

Lemma (differential of Abel-Jacobi via Brill-Noether matrix). Let $D=p_1+\cdots +p_g\in {\mathrm{Sym}}^g(X)$ with distinct $p_i$, and choose local coordinates $z_i$ near each $p_i$. Write ${\omega }_j=f_{j,i}(z_i)d{z}_i$ near $p_i$. Then the differential $d\alpha {|}_D:T_D{\mathrm{Sym}}^g(X)\to T_{\alpha (D)}\mathrm{Jac}(X)\cong {\mathbb{C}}^g$ is given by the $g\times g$ Brill-Noether matrix $M(D{)}_{ji}=f_{j,i}(0)={\omega }_j(p_i)$. The matrix has rank $g$ iff the divisor $D$ is non-special (equivalently $\dim H^0(X,K_X\otimes \mathcal{O}(-D))=0$).

Proof. In local coordinates, an infinitesimal deformation of $D$ at $p_i$ corresponds to a tangent vector $v_i\in T_{p_i}X=\mathbb{C}$. The integral ${\int }_{p_0}^{p_i+ϵv_i}{\omega }_j$ to first order in $ϵ$ equals ${\int }_{p_0}^{p_i}{\omega }_j+ϵ{\omega }_j(p_i)v_i+O({ϵ}^2)$, where ${\omega }_j(p_i)$ is the local coefficient $f_{j,i}(0)$ of ${\omega }_j$ at $p_i$. Summing over $i$ and reading off the matrix gives $M(D{)}_{ji}={\omega }_j(p_i)$. The columns of $M(D)$ are the vectors $({\omega }_1(p_i),\dots ,{\omega }_g(p_i))\in {\mathbb{C}}^g$ for each $i$; the matrix has rank $g$ iff the linear functionals "evaluate at $p_i$" on $H^0(X,K_X)$ are linearly independent for $i=1,\dots ,g$, iff the points $p_i$ impose independent conditions on $H^0(X,K_X)$. By the kernel description, the failure of independence is exactly the condition that some non-zero $\omega \in H^0(X,K_X)$ vanishes at every $p_i$, equivalently $\omega \in H^0(X,K_X\otimes \mathcal{O}(-D))$, equivalently $H^0(X,K_X\otimes \mathcal{O}(-D))\ne 0$. By Serre duality 06.04.04, this is equivalent to $H^1(X,\mathcal{O}(D))\ne 0$, equivalently $D$ is special. $\square$

Theorem (Jacobi inversion, full statement). Statement and proof as in the Intermediate-tier Key theorem section.

Proof. The Intermediate-tier proof goes through using the three lemmas above as packaged inputs: ${\mathrm{Sym}}^g(X)$ is smooth projective of dimension $g$ (Lemma 1); the Abel-Jacobi map $\alpha$ is well-defined and holomorphic (Lemma 2); the differential $d\alpha {|}_D$ is identified with the Brill-Noether matrix and has rank $g$ for non-special $D$ (Lemma 3). The closed-image step uses compactness of ${\mathrm{Sym}}^g(X)$. The full-dimensional-closed-subvariety-of-a-connected-complex-manifold step concludes surjectivity. The generic-fibre step uses Riemann-Roch and Serre duality as in Lemma 3, applied to a generic line bundle $L$ of degree $g$ to extract the unique effective divisor $D$ with $\mathcal{O}(D)\cong L$. $\square$

Corollary (image of degree-$d$ Abel-Jacobi for $d<g$). For $0\le d\le g-1$, the Abel-Jacobi map ${\alpha }_d:{\mathrm{Sym}}^d(X)\to {\mathrm{Pic}}^d(X)$ is generically injective with image $W_d:={\alpha }_d({\mathrm{Sym}}^d(X))$ of dimension $d$, a proper closed subvariety of codimension $g-d$.

Proof. The same closed-image / dimension-count argument as the inversion theorem applies: ${\mathrm{Sym}}^d(X)$ is smooth projective of dimension $d$, ${\alpha }_d$ is holomorphic, the image is closed of dimension $d$ (the differential has rank $d$ at a generic divisor by the same Brill-Noether matrix argument with a $g\times d$ matrix of full rank). The codimension in ${\mathrm{Pic}}^d(X)\cong \mathrm{Jac}(X)$ of complex dimension $g$ is $g-d$. Generic injectivity follows from Riemann-Roch: for a generic line bundle $L$ of degree $d<g$, $\dim H^0(X,L)=0$ (by Clifford / Brill-Noether non-existence), so a generic $z\in W_d$ is in fact not in $W_d$ — this needs sharpening. The correct statement: the image $W_d$ is the union of ${\alpha }_d(D)$ over $D\in {\mathrm{Sym}}^d(X)$, and at a generic $D$, $\dim H^0(X,\mathcal{O}(D))=1$, so $|D|=\{D\}$ is a single point and ${\alpha }_d^{-1}({\alpha }_d(D))=\{D\}$, exhibiting generic injectivity on ${\mathrm{Sym}}^d(X)\to W_d$. $\square$

Corollary (image of degree-$d$ Abel-Jacobi for $d>g$). For $d>g$, the Abel-Jacobi map ${\alpha }_d:{\mathrm{Sym}}^d(X)\to {\mathrm{Pic}}^d(X)$ is surjective with generic fibre ${\mathbb{P}}^{d-g}$.

Proof. Surjectivity follows from the inversion theorem: every $z\in {\mathrm{Pic}}^d(X)$ can be written as $z=z^{\prime }+(d-g)\cdot \alpha (p_0)$ for some $z^{\prime }\in {\mathrm{Pic}}^g(X)$, with $z^{\prime }$ in the image of ${\alpha }_g$ by the inversion theorem; pulling back gives $z\in \mathrm{image}({\alpha }_d)$. The fibre ${\alpha }_d^{-1}(L)=|L|$ is the projectivisation of $H^0(X,L)$, which for generic $L$ of degree $d>g$ has $h^0(L)=d+1-g$ by Riemann-Roch (since $\dim H^1(L)=0$ when $\mathrm{deg}L>2g-2$, and even for $g-1\le \mathrm{deg}L\le 2g-2$ generic $L$ has $h^1=0$ by Brill-Noether non-speciality). Hence generic fibre is ${\mathbb{P}}^{d-g}$. $\square$

Corollary (theta divisor as image of ${\mathrm{Sym}}^{g-1}(X)$). The image $W_{g-1}:={\alpha }_{g-1}({\mathrm{Sym}}^{g-1}(X))\subset {\mathrm{Pic}}^{g-1}(X)\cong \mathrm{Jac}(X)$ is a divisor in $\mathrm{Jac}(X)$, the theta divisor up to translation by the Riemann constant.

Proof. By the previous corollary, $W_{g-1}$ has dimension $g-1$ in the $g$-dimensional Jacobian, hence is a divisor. Riemann's vanishing theorem (cited above as Riemann 1857) identifies $W_{g-1}$ with the zero locus of the Riemann theta function $\theta$ shifted by the Riemann constant $\kappa$; this identification is a transcendental computation using the quasi-periodicity of $\theta$ and the residue calculus on $X$, and is the content of 06.06.05. $\square$

Connections [Master]

• Jacobian variety 06.06.03. The Jacobian is the codomain of the Abel-Jacobi map, and the Jacobi inversion theorem is the structural theorem identifying the Jacobian with ${\mathrm{Pic}}^g(X)$ — every point of the Jacobian comes from an effective divisor of degree $g$, generically uniquely.

• Serre duality on a curve 06.04.04. The differential of the Abel-Jacobi map at a generic divisor is a Riemann-Roch + Serre-duality computation: rank-$g$ of the Brill-Noether matrix is equivalent to non-speciality $\dim H^0(K_X\otimes \mathcal{O}(-D))=0$, which Serre duality re-expresses as $\dim H^1(\mathcal{O}(D))=0$. Without Serre duality the closed-image argument would not produce the dimension count.

• Hodge decomposition on a compact Riemann surface 06.04.03. The Hodge decomposition supplies the basis of holomorphic 1-forms ${\omega }_1,\dots ,{\omega }_g$ used to define the Abel-Jacobi map, and the bilinear-relation positivity that promotes $\mathrm{Jac}(X)$ to a principally polarised abelian variety. The integration $\gamma \mapsto (\omega \mapsto {\int }_{\gamma }\omega )$ embedding $H_1(X,\mathbb{Z})$ into $H^0(X,{\Omega }_X^1{)}^{\ast }$ is exactly the Hodge data.

• Abel-Jacobi map 06.06.04. Abel's theorem (kernel of degree-$0$ Abel-Jacobi = principal divisors) and Jacobi inversion (surjectivity + birationality of degree-$g$ Abel-Jacobi) together give the structural identification ${\mathrm{Pic}}^0(X)\cong \mathrm{Jac}(X)$ as complex Lie groups.

• Theta function 06.06.05. Riemann's vanishing theorem identifies the exceptional locus of the Abel-Jacobi map (the theta divisor) with the zero locus of the Riemann theta function shifted by the Riemann constant. The transcendental description of the theta divisor through $\theta$ is the explicit form of Jacobi inversion's exceptional set.

• Period matrix 06.06.02. The period matrix $\Pi =(A\mid B)$ is the input data to the Abel-Jacobi map: the $A$-periods and $B$-periods assemble the period lattice $\Lambda$ that defines the Jacobian, and the integrals over paths from $p_0$ to $p_i$ that define $\alpha$ are computed against this period basis. Riemann's bilinear relations are the structural input for the principal polarisation.

• Holomorphic 1-form 06.06.01. The basis ${\omega }_1,\dots ,{\omega }_g$ of holomorphic 1-forms is the working object of the Abel-Jacobi map; the dimension count $\dim H^0(X,{\Omega }_X^1)=g$ is what makes the source ${\mathrm{Sym}}^g(X)$ and target $\mathrm{Jac}(X)$ both $g$-dimensional.

• Riemann-Roch theorem for compact Riemann surfaces 06.04.01. The dimension count $\dim H^0(\mathcal{O}(D))=\mathrm{deg}D+1-g+\dim H^0(K_X-D)$ powers both the differential computation (Step 2 of the proof) and the generic-fibre argument (Step 4). Without Riemann-Roch the inversion proof has no quantitative engine.

• Holomorphic line bundle on a Riemann surface 06.05.02. The identification ${\mathrm{Pic}}^g(X)\cong \mathrm{Jac}(X)$ via the Abel-Jacobi map (after translation by $g\cdot p_0$) is a structural statement about line bundles of degree $g$: every such line bundle has a unique-up-to-scalar non-zero global section for generic class, and the zero divisor of that section is the unique preimage in ${\mathrm{Sym}}^g(X)$.

• Divisor on a Riemann surface 06.05.01. The symmetric product ${\mathrm{Sym}}^g(X)$ parametrises effective divisors of degree $g$; the Abel-Jacobi map is the reduction-modulo-linear-equivalence map, and Jacobi inversion is the surjectivity-plus-birationality of this map onto ${\mathrm{Pic}}^g(X)\cong \mathrm{Jac}(X)$.

Historical & philosophical context [Master]

Carl Gustav Jacob Jacobi posed and solved the inversion problem in two papers in Crelle's Journal: Considerationes generales de transcendentibus Abelianis ^{[Jacobi 1834]} (Crelle 9, 1832, 394-403) and De functionibus duarum variabilium quadrupliciter periodicis quibus theoria transcendentium Abelianarum innititur (Crelle 13, 1834, 55-78). Jacobi's motivation came from Niels Henrik Abel's 1826/1828 Mémoire sur une propriété générale d'une classe très-étendue de fonctions transcendantes (Crelle 4, 1829, posthumous), which had shown that integrals $\int R(x,y)dx$ where $y$ satisfies a polynomial relation $F(x,y)=0$ obey an addition theorem generalising the elliptic-integral case. Jacobi recognised that Abel's addition theorem implied, for genus $g$ curves, a $g$-fold inversion problem: the natural extension of the elliptic inversion $u=\int dz/y$ giving $z={\wp }^{-1}(u)$ to genus $g$ requires inverting a system of $g$ integrals against $g$ holomorphic 1-forms simultaneously. The 1834 paper formulated this as the existence-and-uniqueness statement that the Abel-Jacobi map ${\mathrm{Sym}}^g(X)\to \mathrm{Jac}(X)$ is bijective (in modern terms — Jacobi worked with explicit algebraic curves and rational expressions, not with the abstract $\mathrm{Jac}(X)$).

Bernhard Riemann gave the modern formulation in 1857 Theorie der Abel'schen Functionen ^{[Riemann 1857]} (Crelle 54, 115-155), introducing the Riemann surface as the natural domain of multivalued algebraic functions, defining the Jacobian variety as the period quotient ${\mathbb{C}}^g/\Lambda$, and proving the birationality refinement of Jacobi's surjectivity statement. Riemann's 1857 paper introduced the Riemann theta function and the Riemann vanishing theorem identifying the theta divisor with the image of ${\mathrm{Sym}}^{g-1}(X)$ under the analogous degree-$(g-1)$ Abel-Jacobi map shifted by the Riemann constant. The vanishing theorem closed the inversion problem by giving an explicit transcendental object whose zero locus is the exceptional set; it also opened the theory of theta functions on principally polarised abelian varieties, the foundation of modern algebraic-curves theory.

The bridge from Riemann's transcendental theory to the modern algebraic-geometric framework was constructed by Felix Klein and Henri Poincaré (algebraic functions on Riemann surfaces, 1882-1884), Émile Picard (Théorie des fonctions algébriques de deux variables indépendantes, 1897-1906), and Solomon Lefschetz (L'Analysis Situs et la Géométrie Algébrique, 1924), culminating in the Lefschetz embedding theorem realising the Jacobian as a projective abelian variety. André Weil's 1948 Variétés abéliennes et courbes algébriques and David Mumford's Tata Lectures on Theta I-III (1983-1991) ^{[Mumford Tata Lectures]} gave the modern treatment of the Jacobian as a principally polarised abelian variety, the Schottky problem, and the Krichever / KP-hierarchy correspondence.

Donaldson's Riemann Surfaces (Oxford GTM 22, 2011) §11–§12 ^{[Donaldson Riemann Surfaces]} presents the Jacobi inversion theorem via the closed-image / dimension-count argument with the Brill-Noether matrix as the explicit calculation; Forster's Lectures on Riemann Surfaces (GTM 81, 1981) §21 ^{[Forster Riemann Surfaces]} gives the same proof inside the broader Abel-Jacobi development. Griffiths-Harris Principles of Algebraic Geometry (1978) §2.7 ^{[Griffiths-Harris]} integrates the inversion theorem with the Brill-Noether stratification of the Picard variety, providing the modern algebraic-geometry context for the curve-side theory.

Bibliography [Master]

[object Promise]