Positivity of divisors

The positivity of a Cartier divisor can be measured by sections, intersection numbers, or its numerical class. Ample divisors give projective embeddings after taking a multiple; nef divisors are their numerical limits; big divisors have the maximum possible order of section growth. Nefness and bigness are different conditions.

A Cartier divisor is semiample if some positive multiple is a basepoint-free divisor, equivalently its divisor line bundle has a globally generated positive power. Such a power defines a Kodaira map $f$ with $\mathcal{O}(mD)=f^{\ast }\mathcal{O}(1)$. Semiampleness implies nefness but need not imply ampleness: a fibre divisor of a morphism to a curve is a basic example.

A smooth rational curve $C$ with $C^2=0$ on a smooth projective surface is semiample and has Iitaka dimension one. Its normal bundle is trivial, so the restriction sequences for $mC$ have quotient ${\mathcal{O}}_C$ and zero quotient $H^1$. The finite dimensions of $H^1(X,mC)$ decrease and stabilize. Restriction of sections to $C$ is then surjective; a lift of $1$ and the canonical section generate globally. Exactness also gives $h^0(X,mC)-h^0(X,(m-1)C)=1$ eventually.

A semiample divisor on an integral projective variety is ample if it has positive degree on every integral projective curve. Its Kodaira map cannot contract a positive-dimensional fibre, since such a fibre contains a curve and the pulled-back hyperplane bundle has degree zero there. Thus the morphism is proper and a quasi-finite morphism, hence a finite morphism. The finite pullback of an ample line bundle is ample.

If $E$ is an effective Cartier divisor on a projective scheme and ${\mathcal{O}}_E(E)$ is ample, then $E$ is semiample. The divisor restriction exact sequence and Serre vanishing make $H^1(X,(m-1)E)\to H^1(X,mE)$ surjective for large $m$. Their finite dimensions stabilize, so restriction on global sections is eventually surjective. Lift generators on $E$; off $E$, the canonical section of $mE$ generates. Together these generate ${\mathcal{O}}_X(mE)$, including on nonreduced $X$.

A Cartier divisor $D$ on an integral $n$-dimensional projective variety is big when $h^0(X,mD)\ge c{m}^n$ for some $c>0$ and infinitely many positive integers $m$. Equivalently, its Iitaka dimension is $n$. Kodaira's lemma characterizes bigness by an ample-plus-effective decomposition, and the birational linear system criterion for bigness shows why this growth captures the full variety.

For ample rational Cartier classes $B,C$ on an integral projective $n$-fold with $n\ge 1$, $B^n>n{B}^{n-1}\cdot C$ implies $B-C$ is big. Scale to very ample integral divisors. Choose an effective Cartier divisor $G\in |C|$. Repeated restriction gives $h^0(m(B-C))\ge h^0(mB)-m{h}^0(G,mB)$: multiply each negatively twisted restriction by a section avoiding its associated points to inject it into ${\mathcal{O}}_G(mB)$. Asymptotic Riemann–Roch and Serre vanishing give the positive leading lower bound along sufficiently divisible section indices after undoing the scaling. No complex-analytic Morse theory or characteristic-zero vanishing is used.

Let $\nu :X^{\nu }\to X$ be the finite normalization of an integral projective variety. The coherent sheaf $\mathcal{Q}={\nu }_{\ast }{\mathcal{O}}_{X^{\nu }}/{\mathcal{O}}_X$ is supported in dimension at most $n-1$. For a Cartier divisor $D$, the projection formula for sheaves and the resulting long exact sequence in sheaf cohomology giveThe last step uses the polynomial bound for sections of a fixed divisor. Therefore the leading order $m^n$ growth, and hence bigness, is preserved in both directions. This handles nonnormal varieties without assuming a resolution of singularities in positive characteristic.

A big real divisor is an actual positive real combination of big Cartier divisors. Equivalently it is real linearly equivalent, or numerically equivalent, to an ample real divisor plus an effective real divisor. The rational approximation of an ample-plus-effective real divisor and Kodaira's lemma connect these formulations. Fujino's notes on big real divisors give the definition also for nonnormal varieties; bigness under finite normalization relates it to section growth.

Write a big real Cartier divisor on an integral projective variety as $D{\sim }_{\mathbb{R}}B+E$, with $B$ ample and $E$ effective real Cartier. Every curve outside $\mathrm{Supp}E$ has $D\cdot C=B\cdot C+E\cdot C>0$. Therefore curves negative against $D$ lie in finitely many support components of codimension one. On a projective scheme use componentwise bigness on a projective scheme and collect the supports on its finitely many integral components; codimension one is taken in the relevant component.

If a big class $D=B+E$ has ample restriction to every component $E_i$ of the effective support, then for any ample $A$ a sufficiently small $ϵ>0$ makes $D-ϵA$ nef. Use openness of the ample cone to keep $B-ϵA$ and all $(D-ϵA){|}_{E_i}$ ample with one finite minimum of bounds. Curves inside the support use these restrictions; curves outside use effectivity of $E$. Thus one does not need a uniform choice over infinitely many subvarieties.

For a possibly reducible projective scheme, componentwise bigness means that a real Cartier class restricts to a big real divisor on every reduced irreducible component. This specifies the convention needed by positivity arguments that treat all curves. Maximal total section growth alone is weaker: on ${\mathbb{P}}^2⨿{\mathbb{P}}^2$, the bundle $(\mathcal{O}(1),\mathcal{O}(-1))$ has quadratic total growth but negative degree on every line in the second component. Thus its negative curves cannot be confined to finitely many divisors.

Suppose $D{\sim }_{\mathbb{R}}tA+E$ on an integral projective variety, where $t>0$, $A$ is ample Cartier and $E$ is effective real Cartier. Then $D$ is an actual positive combination of big Cartier divisors.

Here is a finite-dimensional proof. First assume $X$ is normal. Express $D$ and $E$ in finite Cartier bases and write $D-tA-E$ as a finite combination of principal Cartier divisors. The union of the supports of these finitely many divisors has finitely many prime components. Their integer multiplicities turn the equality into finitely many rational linear equations and effectivity into finitely many rational linear inequalities. The given coefficient tuple lies in a rational polyhedron. Take its smallest face; within that face it lies in the relative interior, and $t>0$ is an open condition. A small simplex with rational vertices in this relative interior contains the tuple. Each vertex gives $D_{\nu }{\sim }_{\mathbb{Q}}{t}_{\nu }A+E_{\nu }$ with $t_{\nu }>0$ and $E_{\nu }\ge 0$. Clearing denominators and applying Kodaira's lemma shows that $D_{\nu }$ is a positive rational multiple of a big Cartier divisor. Taking the original convex weights proves the required actual equality.

If $X$ is nonnormal, pull the finite Cartier bases and the relation to its finite normalization and impose the same rational equations and effectivity inequalities there. The vertex divisors remain rational Cartier divisors on $X$ because they were constructed in bases from $X$. They are big on the normalization, hence big on $X$ by bigness under finite normalization. This proves the same conclusion. Effectivity is used in the usual effective Cartier sense so that pullback is effective; arbitrary cycles on a nonnormal variety cannot be substituted without defining a compatible divisor theory.

The big cone in $N^1(X{)}_{\mathbb{R}}$ is open and convex: write a big class as ample plus effective, perturb only the ample summand, and use openness of the ample cone. For positive-dimensional $X$, intersection with $H^{n-1}$ for any very ample divisor $H$ is strictly positive on every big class. Hence this cone contains no line and does not contain the zero class.

A Cartier divisor on an integral projective variety is big if and only if some multiple has a complete linear system of a divisor giving a rational map birational onto its image. Kodaira's lemma embeds a very ample subsystem in a suitable multiple. Conversely, $n$ algebraically independent elements among the section ratios give $\left(\genfrac{}{}{0ex}{}{q+n}{n}\right)$ independent degree-$q$ section monomials, where $n=\dim X$. Smoothness is not necessary.

A Cartier divisor $D$ on an integral projective variety is big exactly when, for every ample divisor $A$, some positive integer $j$ satisfies $jD\sim A+E$ with $E$ effective. To prove the forward direction, subtract an effective high multiple of $A$ using the section subtraction lemma for big divisors, then add an effective representative of the remaining multiple. For the reverse direction, multiply sections of $qA$ by the section of $qE$ and use the ample Hilbert polynomial.

If $D$ is a big divisor and $F$ an effective Cartier divisor, infinitely many $m$ have $H^0(X,mD-F)\ne 0$. The divisor restriction exact sequence bounds the dimension lost upon restriction to $F$ by $O(m^{n-1})$, using the polynomial bound for sections of a fixed divisor; this cannot exhaust the $c{m}^n$ sections along the infinite growth sequence.

Two Cartier divisors are numerically equivalent when they have equal intersection numbers with every integral complete curve. A numerically trivial divisor has zero intersection with every such curve. Unlike linear equivalence of Cartier divisors, numerical equivalence does not require the same line bundle.

The finite-dimensional real vector space $N^1(X{)}_{\mathbb{R}}$ consists of real combinations of Cartier divisors modulo numerical equivalence of divisors. Its elements are numerical divisor classes. The space of curve classes paired with it is $N_1(X{)}_{\mathbb{R}}$.

The nef cone is the closed convex cone of nef divisor classes. It is dual to the closed cone of curves. Its interior is the ample cone on a projective scheme.

The ample cone consists of real numerical divisor classes represented by ample real divisors. It is an open convex cone, and Kleiman's criterion identifies it with the interior of the nef cone.

The sum of a nef real Cartier class and an ample real Cartier class on a projective scheme is ample. By Kleiman's criterion, the ample cone is the interior of the nef cone. If a ball about an ample class lies in that convex cone, translating it by a nef class still lies in the cone. The sum is consequently still an interior point.

The closed cone of curves is the closure in $N_1(X{)}_{\mathbb{R}}$ of the convex cone generated by classes of integral curves. A nef divisor pairs nonnegatively with this entire closed cone. Its boundary can contain limiting classes which are not represented by one curve.

A line bundle is nef when its degree on every integral complete curve is nonnegative. Equivalently a Cartier divisor $D$ is nef when $D\cdot C\ge 0$ for every such curve. On a projective scheme, adding a positive ample class to a nef class makes it ample; taking limits in the intersection product shows that a nef divisor has nonnegative top self-intersection number.

For a nef divisor on an integral projective variety of dimension $n$,Indeed, combine asymptotic Riemann–Roch with cohomology growth for nef twists. Thus its normalized section-growth limit is $D^n/n!$, and it is big exactly when $D^n>0$.

A Cartier divisor is ample when its associated line bundle is ample. The Nakai–Moishezon criterion and Kleiman's criterion characterize this condition numerically.

A Cartier divisor is ample if for every positive-dimensional integral closed subvariety some positive multiple restricts to a bundle with a nonzero section having a nonempty zero locus. Inductively the divisor is ample on all lower-dimensional subschemes. On an integral component the chosen section cuts out a nonempty effective Cartier divisor $E$ whose restriction bundle ${\mathcal{O}}_E(E)$ is ample. The restriction ampleness implies semiampleness for an effective divisor lemma makes the original divisor semiample. On a curve the vanishing section forces positive degree, so the semiample and curve-positive ampleness criterion proves ampleness.

A Cartier divisor $D$ on a projective scheme is ample exactly when $\chi (V,{\mathcal{O}}_V(mD))\to +\infty$ for every positive-dimensional integral closed subvariety $V$, including irreducible components. For the converse induct on dimension: restrictions to hyperplane sections are ample, so higher cohomology vanishing from an ample hyperplane restriction gives $h^0=\chi +h^1\to \infty$. A nonzero section vanishing at a chosen point then exists. The vanishing-section ampleness criterion finishes. Divergence alone does not imply a positive top-degree coefficient.

For a projective scheme, a divisor class is in the ample cone exactly when it is strictly positive on every nonzero element of the closed cone of curves. Equivalently, the ample cone is the interior of the nef cone. The projectivity assumption matters: the same characterization is not asserted here for arbitrary proper schemes.

An integral Cartier divisor $D$ on a projective scheme is ample if and only if $D^{\dim Y}\cdot Y>0$ for every positive-dimensional integral closed subvariety $Y$. Testing only curves is sufficient in dimension one, but not in higher dimensions.

A real Cartier class on a projective scheme is ample if and only if its top self-intersection on every positive-dimensional integral subvariety is positive. For the converse curve tests give nefness. Small ample perturbations and rational approximation give ample rational $B,C$ with $D-A=B-C$ satisfying the algebraic Morse inequality for ample divisors, hence bigness. Induction gives ample restrictions on codimension-one subvarieties; uniform ample subtraction from a big divisor with ample exceptional restrictions makes $D-ϵA$ nef. The nef-plus-ample ampleness lemma concludes. For reducible schemes perform the finite component tests simultaneously.