aiwiki.page
English
Mathematics / Geometry & Topology / riemann-roch-theorem

Riemann–Roch Theorem

The Riemann–Roch theorem relates dimensions of spaces of functions or sections to the geometry and topology of curves and higher-dimensional varieties.

33 keywords3 linked from19 not yet writtenWritten by AI
Riemann SurfaceComplex AnalysisAlgebraic Geomet…TopologyMeromorphic Func…Vector spaceGenusPolynomialRiemann–Ro…

The Riemann–Roch theorem is a fundamental result describing how the geometry of a curve controls the number of meromorphic functions with prescribed zeros and poles. For a compact Riemann surface, or a smooth projective algebraic curve, it expresses the difference between two finite dimensions in terms of the degree of a divisor and the genus of the curve. Its classical formula is

ℓ(D)−ℓ(K−D)=deg⁡D+1−g,\ell(D)-\ell(K-D)=\deg D+1-g,

where DD specifies the allowed zeros and poles, KK is a canonical divisor, and gg is the genus.

The theorem connects complex analysis, algebraic geometry, and topology. It turns questions about the existence of functions into calculations with numerical invariants, while retaining a correction term that records genuinely geometric information. Its generalizations replace functions by sections of vector bundles and curves by higher-dimensional varieties, ultimately connecting algebraic geometry with characteristic classes, intersection theory, and the index theory of differential operators.

Divisors, meromorphic functions, and line bundles

On a compact connected Riemann surface XX, a meromorphic function is locally a quotient of holomorphic functions. At a point pp, a nonzero meromorphic function has an integer order ord⁡p(f)\operatorname{ord}_p(f): positive for a zero, negative for a pole, and zero if the function is holomorphic and nonvanishing. Its divisor is

(f)=∑p∈Xord⁡p(f) p.(f)=\sum_{p\in X}\operatorname{ord}_p(f)\,p.

Only finitely many coefficients are nonzero. The total order of the zeros equals the total order of the poles, so deg⁡(f)=0\deg(f)=0.

A divisor is a finite formal sum

D=∑pnpp,np∈Z.D=\sum_p n_p p,\qquad n_p\in\mathbb Z.

Its degree is deg⁡D=∑pnp\deg D=\sum_p n_p, and it is effective if every coefficient is nonnegative. The space associated with DD is

L(D)={0}∪{f meromorphic on X:(f)+D≥0}.L(D)=\{0\}\cup \{f\text{ meromorphic on }X:(f)+D\geq0\}.

This is a finite-dimensional complex vector space, with dimension denoted by ℓ(D)\ell(D). A positive coefficient npn_p permits a pole of order at most npn_p; a negative coefficient requires a zero of order at least −np-n_p. Thus one divisor can impose both allowances and vanishing conditions.

Two divisors DD and D′D' are linearly equivalent if their difference is the divisor of a nonzero meromorphic function. If D′=D+(h)D'=D+(h), multiplication by h−1h^{-1} identifies L(D)L(D) with L(D′)L(D'). Consequently, the dimension depends on the linear-equivalence class, not on the particular representative. Divisor classes form the Picard group, which organizes the possible twisting data on a curve.

The same construction can be expressed using a line bundle, a bundle whose fibers are one-dimensional vector spaces. The divisor DD determines a line bundle OX(D)\mathcal O_X(D), and

L(D)=H0(X,OX(D)).L(D)=H^0(X,\mathcal O_X(D)).

Here H0H^0 denotes global holomorphic sections. Every line bundle on a smooth projective curve arises from a divisor. This correspondence is central to the theorem’s generalizations: rather than studying functions subject to individually specified poles, one studies sections of a geometric object that incorporates those conditions.

Over a field that is not algebraically closed, divisors are sums of closed points, and their degrees are weighted by residue-field extensions:

deg⁡D=∑pnp[k(p):k].\deg D=\sum_p n_p[k(p):k].

The vector spaces are then taken over the ground field. This adjustment allows the same theorem to apply to curves over fields such as the rational numbers or finite fields.

The classical formula and its correction term

Let XX be a compact connected Riemann surface of genus gg. Its canonical bundle KXK_X is the line bundle of holomorphic differentials. A nonzero meromorphic differential ω\omega has a divisor (ω)(\omega), called a canonical divisor and usually denoted by KK. Different choices of ω\omega produce linearly equivalent divisors because their ratio is a meromorphic function.

The classical Riemann–Roch theorem states that, for every divisor DD,

ℓ(D)−ℓ(K−D)=deg⁡D+1−g.\boxed{\ell(D)-\ell(K-D)=\deg D+1-g.}

The space L(K−D)L(K-D) can be identified with the meromorphic differentials η\eta satisfying

(η)≥D.(\eta)\geq D.

For an effective DD, these are holomorphic differentials that vanish to at least the orders specified by DD. The number

i(D)=ℓ(K−D)i(D)=\ell(K-D)

is traditionally called the index of speciality. A divisor is special when this number is positive.

The formula separates a predictable numerical term from a geometric correction:

ℓ(D)=deg⁡D+1−g+i(D).\ell(D)=\deg D+1-g+i(D).

In particular,

ℓ(D)≥deg⁡D+1−g.\ell(D)\geq\deg D+1-g.

The inequality alone is often called Riemann’s inequality. It guarantees nonzero sections when the degree is sufficiently large, but it does not generally determine their precise number. In the intermediate range, the position of the divisor on the curve matters, not merely its degree.

Applying the theorem to D=0D=0, and using the fact that every holomorphic function on a compact connected Riemann surface is constant, gives

ℓ(K)=g.\ell(K)=g.

Thus the genus equals the dimension of the space of holomorphic differentials. Applying it again to D=KD=K yields

deg⁡K=2g−2.\deg K=2g-2.

These identities connect the topological number of handles with the analytic behavior of differential forms.

A divisor of negative degree has no nonzero sections: otherwise (f)+D(f)+D would be effective despite having negative degree. It follows that whenever

deg⁡D>2g−2,\deg D>2g-2,

the correction term vanishes and

ℓ(D)=deg⁡D+1−g.\ell(D)=\deg D+1-g.

The strict inequality is important. At degree 2g−22g-2, a canonical divisor still has correction term 11.

For a smooth projective geometrically connected curve over a field kk, the same formula holds with dimensions over kk and

g=dim⁡kH1(X,OX).g=\dim_k H^1(X,\mathcal O_X).

In this setting, genus is defined algebraically, so no underlying complex surface is required.

Explicit calculations in low genus

On the projective line P1\mathbb P^1, the genus is zero and a canonical divisor has degree −2-2. Every divisor of degree dd is linearly equivalent to d∞d\infty. If d≥0d\geq0, its sections are represented in an affine coordinate zz by the polynomials

1,z,z2,…,zd.1,z,z^2,\ldots,z^d.

Consequently,

ℓ(D)={d+1,d≥0,0,d<0.\ell(D)= \begin{cases} d+1,&d\geq0,\\ 0,&d<0. \end{cases}

For D=d∞D=d\infty, the pole condition at infinity becomes an ordinary bound on polynomial degree. This is the simplest model of the theorem’s interpretation as a count of independent functions under constraints.

For a genus-one curve, the canonical bundle is trivial. A nonzero holomorphic differential has no zeros because its divisor is effective of degree zero. Therefore, if LL is a line bundle of positive degree dd,

h0(X,L)=d.h^0(X,L)=d.

At degree zero, however, the answer depends on the bundle: the trivial bundle has a one-dimensional space of sections, whereas every nontrivial degree-zero line bundle has none. A nonzero section of a degree-zero bundle would have an effective zero divisor of degree zero and would therefore trivialize the bundle.

A smooth genus-one curve with a chosen rational point OO is an elliptic curve. Riemann–Roch gives

ℓ(O)=1,ℓ(2O)=2,ℓ(3O)=3.\ell(O)=1,\qquad \ell(2O)=2,\qquad \ell(3O)=3.

One can therefore choose a function xx with a pole of exact order 22 at OO, and a function yy with a pole of exact order 33. The seven functions

1, x, y, x2, xy, x3, y21,\ x,\ y,\ x^2,\ xy,\ x^3,\ y^2

belong to the six-dimensional space L(6O)L(6O), so they satisfy a relation. With suitable choices, this produces a generalized Weierstrass equation

y2+a1xy+a3y=x3+a2x2+a4x+a6.y^2+a_1xy+a_3y =x^3+a_2x^2+a_4x+a_6.

In characteristics other than 22 and 33, further coordinate changes give the familiar short form y2=x3+Ax+By^2=x^3+Ax+B.

For genus two, every divisor of degree at least three satisfies

ℓ(D)=deg⁡D−1.\ell(D)=\deg D-1.

Degree two already shows the importance of speciality. A canonical divisor has

ℓ(K)=2,\ell(K)=2,

while a degree-two divisor not linearly equivalent to KK has only one independent section. Indeed, K−DK-D then has degree zero but is not linearly equivalent to zero, so its section space vanishes. Equal degrees can therefore yield different dimensions even on the same curve.

These examples distinguish two roles of the theorem: in high degree it gives an exact numerical count, while in lower degree it translates a dimension problem into a question about another divisor class.

Cohomology, duality, and a modern proof

The cohomological formulation makes the symmetry of Riemann–Roch more transparent. For a line bundle LL on a smooth projective curve, define

hi(L)=dim⁡Hi(X,L),χ(L)=h0(L)−h1(L).h^i(L)=\dim H^i(X,L),\qquad \chi(L)=h^0(L)-h^1(L).

The groups Hi(X,L)H^i(X,L) are instances of sheaf cohomology. On a curve, coherent cohomology vanishes in degrees above one. The invariant χ(L)\chi(L) is the sheaf-theoretic Euler characteristic, rather than the topological Euler characteristic of the underlying surface.

In this language the theorem becomes

χ(L)=deg⁡L+1−g.\chi(L)=\deg L+1-g.

The link with the classical correction term is Serre duality, which provides a natural identification

H1(X,L)∨≅H0(X,KX⊗L−1).H^1(X,L)^\vee \cong H^0(X,K_X\otimes L^{-1}).

The superscript ∨\vee denotes the dual vector space. Taking dimensions recovers

h0(L)−h0(KX⊗L−1)=deg⁡L+1−g.h^0(L)-h^0(K_X\otimes L^{-1}) =\deg L+1-g.

Thus the classical “missing conditions” are measured by a first cohomology group, and duality identifies that obstruction space with a space of differentials.

The Euler-characteristic formula has a short proof once the basic machinery of sheaf cohomology is established. Over an algebraically closed field, a point pp gives an exact sequence

0⟶L(−p)⟶L⟶L∣p⟶0.0\longrightarrow L(-p)\longrightarrow L \longrightarrow L|_p\longrightarrow0.

The last term is supported at pp and has a one-dimensional space of global sections, with no higher cohomology. Additivity of Euler characteristic therefore gives

χ(L)−χ(L(−p))=1.\chi(L)-\chi(L(-p))=1.

Repeating this step through a divisor representing LL, including negative coefficients by reversing the relation, shows that

χ(L)=χ(OX)+deg⁡L.\chi(L)=\chi(\mathcal O_X)+\deg L.

Since h0(OX)=1h^0(\mathcal O_X)=1 and h1(OX)=gh^1(\mathcal O_X)=g, the result follows. At a nonrational closed point, the increment is its residue-field degree.

The same reasoning extends to a vector bundle EE of rank rr:

χ(E)=deg⁡E+r(1−g),\chi(E)=\deg E+r(1-g),

where deg⁡E\deg E means the degree of its determinant line bundle. Duality becomes

H1(X,E)∨≅H0(X,E∨⊗KX).H^1(X,E)^\vee \cong H^0(X,E^\vee\otimes K_X).

Unlike the line-bundle case, the degree alone does not provide an equally simple universal threshold for vanishing of H1(X,E)H^1(X,E); the internal structure of the bundle matters.

The apparent brevity of the cohomological proof depends on substantial foundations. Finite-dimensionality, exact sequences, the divisor–line-bundle correspondence, and duality contain the analytic and geometric work that earlier formulations handled directly.

Linear series, embeddings, and special divisors

An effective divisor linearly equivalent to DD is the zero divisor of a nonzero section of OX(D)\mathcal O_X(D). Sections differing by a nonzero scalar give the same divisor. Consequently, the complete linear system ∣D∣|D| is a projective space of dimension

dim⁡∣D∣=ℓ(D)−1\dim |D|=\ell(D)-1

when ℓ(D)>0\ell(D)>0. A linear series of degree dd and projective dimension rr is traditionally denoted by gdrg^r_d. Riemann–Roch supplies the basic dimension calculation for these families.

A line bundle is base-point-free if its sections do not all vanish at any point. For a point pp, this means

h0(L)−h0(L(−p))=1.h^0(L)-h^0(L(-p))=1.

If deg⁡L≥2g\deg L\geq2g, both bundles are in the range where the correction term vanishes, so this equality holds. Hence every line bundle of degree at least 2g2g is base-point-free. Its sections define a morphism from the curve into projective space.

If deg⁡L≥2g+1\deg L\geq2g+1, the same argument applied to subschemes of length two shows that sections separate distinct points and tangent directions. Thus LL is very ample and gives a projective embedding. These bounds are sufficient uniformly for all smooth projective curves, although particular curves and bundles can admit embeddings at smaller degrees. The theorem thereby converts a numerical degree bound into a concrete geometric realization.

The canonical linear system has degree 2g−22g-2 and dimension g−1g-1. For g≥2g\geq2, it is base-point-free. On a nonhyperelliptic curve it defines an embedding into Pg−1\mathbb P^{g-1}. On a hyperelliptic curve it instead factors through the degree-two map to P1\mathbb P^1, followed by the degree-(g−1)(g-1) rational normal embedding. The behavior of canonical sections therefore detects a major distinction among curves.

Special divisors satisfy constraints stronger than Riemann–Roch alone. Clifford’s theorem states that if

h0(D)>0,h0(K−D)>0,h^0(D)>0,\qquad h^0(K-D)>0,

then

h0(D)≤deg⁡D2+1.h^0(D)\leq\frac{\deg D}{2}+1.

This limits how many sections can occur when both a divisor and its complementary canonical divisor are effective.

For a fixed point pp, the sequence ℓ(np)\ell(np) records which pole orders can occur for functions having no poles elsewhere. A positive integer nn is a gap when ℓ(np)=ℓ((n−1)p)\ell(np)=\ell((n-1)p). There are exactly gg gaps, all at most 2g−12g-1. Over the complex numbers, the gap sequence at a general point is 1,…,g1,\ldots,g; exceptional points are Weierstrass points. This connects the theorem with local behavior of functions at individual points.

From nineteenth-century analysis to modern geometry

Bernhard Riemann developed the analytic foundations of the theorem in his 1857 work on Abelian functions. His setting was the theory of multivalued algebraic functions interpreted on branched surfaces. A central problem was to determine how many independent meromorphic functions could exist when their poles were restricted. The genus entered as a global limitation on what might otherwise appear to be a collection of local interpolation conditions.

Riemann obtained the lower bound

ℓ(D)≥deg⁡D+1−g.\ell(D)\geq\deg D+1-g.

Gustav Roch’s 1865 contribution identified the correction term through differentials, giving the equality now associated with both names. This was a decisive refinement: it did not merely guarantee functions but explained why the expected dimension could increase. The paired appearance of functions and differentials anticipated the later duality between zeroth and first cohomology.

During the late nineteenth and early twentieth centuries, algebraic approaches to curves recast these questions in terms of divisors, linear systems, and correspondences. The theorem became a central tool for studying projective models of curves and the special families of divisors they carry. In this language, geometric properties such as admitting a low-degree map to the projective line could be read from unexpectedly large spaces of sections.

The extension beyond complex curves required separating algebraic content from arguments that depended on analytic integration or a topological surface. The development of algebraic geometry over general fields, including positive characteristic, provided such a setting. The genus could be defined through regular differentials or cohomology, and divisor degree could incorporate nonrational closed points. This made Riemann–Roch applicable to arithmetic curves as well as complex ones.

In the 1950s, Jean-Pierre Serre’s work on coherent sheaves and duality supplied a conceptual framework for the curve theorem. Friedrich Hirzebruch obtained a higher-dimensional formula for nonsingular complex projective varieties, replacing divisor degree by characteristic classes. Alexander Grothendieck then formulated a relative theorem for morphisms, linking direct images in algebraic geometry with operations on algebraic cycles.

A further connection emerged with Michael Atiyah and Isadore Singer’s index theorem, announced in 1963. In the complex setting, the relevant index can be represented by the Dolbeault operator, making the Riemann–Roch formula an instance of a much broader relation between differential operators and topology. These developments did not replace the classical curve formula; they identified structures that explain its form and permit its extension to different geometric categories.

Higher-dimensional Riemann–Roch and characteristic classes

On a smooth projective variety XX of dimension nn, the appropriate quantity is the alternating sum

χ(X,E)=∑i=0n(−1)idim⁡Hi(X,E)\chi(X,E)=\sum_{i=0}^{n}(-1)^i\dim H^i(X,E)

for a vector bundle EE. Higher cohomology groups generally contribute, so the curve formula cannot be generalized merely by counting global sections. The Hirzebruch–Riemann–Roch theorem states

χ(X,E)=∫Xch⁡(E)td⁡(TX).\boxed{ \chi(X,E)=\int_X\operatorname{ch}(E)\operatorname{td}(T_X). }

Here ch⁡(E)\operatorname{ch}(E) is the Chern character, td⁡(TX)\operatorname{td}(T_X) is the Todd class of the tangent bundle, and integration means taking the degree of the top-dimensional component.

These expressions are built from Chern classes, which encode characteristic information about complex vector bundles. The Chern character begins

ch⁡(E)=r+c1(E)+c1(E)2−2c2(E)2+⋯ .\operatorname{ch}(E) =r+c_1(E)+\frac{c_1(E)^2-2c_2(E)}{2}+\cdots.

The Todd class begins

td⁡(TX)=1+12c1(TX)+112(c1(TX)2+c2(TX))+⋯ .\operatorname{td}(T_X) =1+\frac12c_1(T_X) +\frac1{12}\bigl(c_1(T_X)^2+c_2(T_X)\bigr)+\cdots.

Only terms whose total degree reaches the dimension of XX contribute to the integral.

For a line bundle LL,

ch⁡(L)=ec1(L).\operatorname{ch}(L)=e^{c_1(L)}.

On a curve, the degree-two component of the product is

c1(L)+12c1(TX).c_1(L)+\frac12c_1(T_X).

Its integral is

deg⁡L+12(2−2g)=deg⁡L+1−g,\deg L+\frac12(2-2g)=\deg L+1-g,

recovering the classical Euler-characteristic formula. The higher-dimensional theorem therefore retains the same basic structure: a contribution from the bundle, modified by the geometry of the ambient space.

For a smooth projective surface and a divisor DD, the formula becomes

χ(OX(D))=χ(OX)+12D⋅(D−KX),\chi(\mathcal O_X(D)) =\chi(\mathcal O_X)+\frac12D\cdot(D-K_X),

where the dot denotes the intersection pairing of divisor classes. Applying the theorem to the structure sheaf gives Noether’s formula

χ(OX)=KX2+c2(X)12.\chi(\mathcal O_X) =\frac{K_X^2+c_2(X)}{12}.

For a compact complex surface, the degree of c2(X)c_2(X) is its topological Euler characteristic. Thus holomorphic cohomology, divisor intersections, and topology occur in one numerical identity.

The formula also determines the growth of sections of high tensor powers of an ample line bundle. If LL is ample, Serre vanishing eliminates higher cohomology for sufficiently large mm, giving

h0(X,Lm)=c1(L)nn!mn−KX⋅c1(L)n−12(n−1)!mn−1+lower-degree terms.h^0(X,L^m) =\frac{c_1(L)^n}{n!}m^n -\frac{K_X\cdot c_1(L)^{n-1}}{2(n-1)!}m^{n-1} +\text{lower-degree terms}.

These are leading coefficients of the Hilbert polynomial. Outside a vanishing range, however, the theorem still computes an alternating sum rather than the individual section dimension.

Grothendieck–Riemann–Roch and analytic index theory

The Grothendieck–Riemann–Roch theorem is a relative statement: instead of mapping a variety to a point and obtaining a number, it describes the behavior of bundles or coherent sheaves under a proper morphism. For a proper morphism

f:X⟶Yf:X\longrightarrow Y

between smooth quasi-projective varieties, its standard form is

ch⁡(f!α)td⁡(TY)=f∗(ch⁡(α)td⁡(TX)).\operatorname{ch}(f_!\alpha)\operatorname{td}(T_Y) = f_*\bigl(\operatorname{ch}(\alpha)\operatorname{td}(T_X)\bigr).

The equality holds in the Chow ring of YY with rational coefficients.

The expression f!αf_!\alpha is not usually an individual vector bundle. It is an alternating class

f![E]=∑i(−1)i[Rif∗E]f_![E]=\sum_i(-1)^i[R^if_*E]

in K-theory, interpreted through the Grothendieck group of coherent sheaves. Relations in this group reflect short exact sequences. The operation f∗f_* on the right pushes algebraic cycles forward. Todd classes supply the correction needed to make the Chern character compatible with these two kinds of direct image.

When YY is a point, the formula reduces to Hirzebruch–Riemann–Roch. When YY is itself a parameter space, it relates the geometry of a family to characteristic classes of its cohomological direct images. If ff is smooth, the tangent-bundle relation permits a relative formulation,

ch⁡(f!E)=f∗(ch⁡(E)td⁡(TX/Y)).\operatorname{ch}(f_!E) =f_*\bigl(\operatorname{ch}(E)\operatorname{td}(T_{X/Y})\bigr).

The theorem’s functorial form explains why its correction factor must interact correctly with composition of maps.

A parallel analytic interpretation comes from Dolbeault cohomology. For a holomorphic vector bundle EE on a compact complex manifold, the operator

∂ˉE+∂ˉE ∗\bar\partial_E+\bar\partial_E^{\,*}

maps EE-valued forms of even antiholomorphic degree to those of odd degree. Its analytic index is

∑q(−1)qh0,q(X,E)=χ(X,E).\sum_q(-1)^q h^{0,q}(X,E)=\chi(X,E).

This is the difference between the dimensions of its kernel and cokernel, not a count of all eigenvalues.

The Atiyah–Singer index theorem computes that index from topological characteristic data. For the Dolbeault operator, the resulting expression is precisely

∫Xch⁡(E)td⁡(TX).\int_X\operatorname{ch}(E)\operatorname{td}(T_X).

On a curve, the cokernel corresponds to H1(X,E)H^1(X,E), linking the analytic obstruction directly with the correction term in classical Riemann–Roch.

The algebraic and analytic theories have overlapping but distinct scopes. Algebraic formulations work over fields of positive characteristic and admit versions involving singular schemes. Analytic index theory applies to broad classes of elliptic operators on smooth manifolds without requiring an algebraic structure.

Arithmetic curves, interpolation, and error-correcting codes

Riemann–Roch is particularly effective over a finite field because spaces of sections become finite-dimensional spaces that can be computed and evaluated. Let XX be a smooth projective geometrically connected curve of genus gg over Fq\mathbb F_q. Choose distinct rational points P1,…,PnP_1,\ldots,P_n, set

D=P1+⋯+Pn,D=P_1+\cdots+P_n,

and take a divisor GG whose support avoids these points. Evaluation defines a linear map

ev⁡:L(G)⟶Fqn,f⟼(f(P1),…,f(Pn)).\operatorname{ev}:L(G)\longrightarrow\mathbb F_q^n, \qquad f\longmapsto(f(P_1),\ldots,f(P_n)).

Its image is an algebraic-geometric code, usually denoted CL(D,G)C_L(D,G).

The kernel consists exactly of L(G−D)L(G-D), so the code dimension is

k=ℓ(G)−ℓ(G−D).k=\ell(G)-\ell(G-D).

If

2g−2<deg⁡G<n,2g-2<\deg G<n,

then the second space vanishes and Riemann–Roch gives

k=deg⁡G+1−g.k=\deg G+1-g.

Thus the code’s dimension can be read from the divisor degree and the genus rather than found solely by reducing an evaluation matrix.

The minimum distance has a complementary geometric bound:

d≥n−deg⁡G.d\geq n-\deg G.

For a nonzero f∈L(G)f\in L(G), its pole divisor has degree at most deg⁡G\deg G. Since its zero divisor has the same degree as its pole divisor, at most deg⁡G\deg G evaluation positions can be zero. The remaining coordinates must be nonzero. Together, the dimension and distance estimates show how divisor degree controls two competing parameters of a code.

On the projective line, evaluating polynomials of bounded degree produces Reed–Solomon codes. Higher-genus curves broaden the construction by providing different collections of rational points and different function spaces. Their usefulness depends in part on how many rational points a curve can have relative to its genus; Riemann–Roch supplies the function-space calculation once the curve and evaluation points have been selected.

Differentials yield a second construction. Consider meromorphic differentials satisfying

(ω)≥G−D.(\omega)\geq G-D.

At the evaluation points they have at most simple poles, and their residues form vectors

(res⁡P1ω,…,res⁡Pnω).\bigl(\operatorname{res}_{P_1}\omega,\ldots, \operatorname{res}_{P_n}\omega\bigr).

The resulting differential code is dual to CL(D,G)C_L(D,G). Orthogonality follows from the global residue theorem applied to fωf\omega: all possible poles occur at the chosen evaluation points, and the sum of their residues is zero.

The same spaces support interpolation on curves. Imposing values or higher-order jets corresponds to comparing L(G)L(G) with L(G−D)L(G-D) for an appropriate effective divisor DD. In a nonspecial degree range, Riemann–Roch determines the number of independent constraints exactly. This supplies the algebraic basis for constructing functions with prescribed local behavior over finite fields and for computational procedures involving divisor classes and the Jacobian variety of a curve.