Family 063 — The sharp tradeoff between Fano curves and divisor directions

← All digests

Family 063 · Algebraic and complex geometry
Importance 4/5 · Provisional editorial rating
Sources consulted 2026-10-12

1. The question

Projective space has very few independent divisor directions and very positive rational curves. Products of projective spaces have more divisor directions, but their dimension must grow to accommodate them. The generalized Mukai conjecture proposes that every smooth Fano manifold obeys exactly the same tradeoff.

Its three quantities are dimension $n$, Picard number $\rho$, and pseudoindex $\iota$, the smallest anticanonical degree of a rational curve. The expected inequality is

$$\rho(\iota-1)\le n.$$

Equality should be rigid: the manifold must be a product of $\rho$ copies of $\mathbb P^{\iota-1}$. Family 063 claims both assertions in all dimensions. Its proposed argument finds independent curve directions through quantum cohomology, then turns a grading restriction into the numerical bound.

2. Background and essential definitions

Fano positivity and curve degrees

A Fano manifold is a smooth connected complex projective variety whose anticanonical line bundle $-K_X$ is ample. The canonical bundle consists of top-degree differential forms, and ampleness means some positive tensor power gives a projective embedding.

A rational curve is an irreducible curve whose normalization is $\mathbb P^1$. Its anticanonical degree $-K_X\cdot C$ is a positive integer. Fano manifolds contain rational curves, so the minimum

$$\iota_X=\min_C(-K_X\cdot C)$$

is defined. This minimum is the pseudoindex.

It differs from the Fano index, the largest integer $r$ for which $-K_X\simeq rH$ in the Picard group for some line bundle $H$, using additive notation. Divisibility of the canonical bundle constrains every curve degree, but the minimum degree is not generally the same number as the divisibility index.

How many independent divisor directions?

Two divisors are numerically equivalent if they have equal intersections with every curve. Their real numerical classes form a finite-dimensional vector space. Its dimension is the Picard number $\rho_X$. It is not the number of divisors on the variety, which can be infinite.

The curve classes form the dual numerical space. To produce $\rho_X$ independent curve directions is to find enough curve classes to distinguish every divisor class. A single family sweeping out all points need not provide all those independent directions.

For $\mathbb P^a$, one has $-K=(a+1)H$, Picard number one, and pseudoindex $a+1$. For a product of $r$ copies of $\mathbb P^a$, lines in one factor have anticanonical degree $a+1$, the dimension is $ra$, and the Picard number is $r$. Thus

$$r\bigl((a+1)-1\bigr)=ra,$$

which explains both the proposed bound and the expected equality example.

Counting curves through cohomology

Ordinary cohomology multiplies classes by cup product. Quantum cohomology modifies this multiplication using counts of rational curves meeting specified conditions. These counts are Gromov–Witten invariants; their virtual definition accounts for singular or obstructed spaces of maps.

A marked point on a mapped curve has a cotangent line. Its first Chern class is denoted $\psi$. Inserting powers of $\psi$ produces descendant invariants. They retain deformation information at the marked point in addition to requiring its image to lie on a chosen cycle.

One need not compute these invariants to follow the claim. Their role is to create commuting linear operators whose degrees remember the positivity of the contributing curves.

3. Development history and previous partial results

Mukai’s original formulation used the Fano index. Wiśniewski introduced the pseudoindex and proved a strong restriction at very large pseudoindex: $2\iota_X>n+2$ forces Picard number one. Bonavero–Casagrande–Debarre–Druel formulated the generalized conjecture and established the bound in dimension at most four and in several toric ranges. Their paper also proposed an approach through chains of independent rational curves. Bonavero–Casagrande–Debarre–Druel.

Andreatta–Chierici–Occhetta extended the low-dimensional result to dimension five and proved cases under additional contraction assumptions. Large pseudoindex provided another broad range: Novelli–Occhetta handled $\iota_X\ge(n+3)/3$, and Novelli later proved the conjecture whenever $\iota_X>n/3$. At the boundary $n=3\iota_X$, further results use extra geometric hypotheses. Novelli’s theorem.

Symmetry has also made the problem tractable. Casagrande proved the toric version for suitable Gorenstein varieties in every dimension, and Pasquier treated horospherical varieties. Gagliardi–Hofscheier–Pearson subsequently proved the generalized conjecture for $\mathbb Q$-factorial spherical Fano varieties. These singular varieties with large group actions form a different scope from the unrestricted smooth theorem claimed here. The spherical result.

The difficulty with the chain approach is not the final dimension estimate. It is constructing enough independent proper families of rational curves with the required incidence properties. M1 first detects numerical curve directions through quantum operators and only needs the stronger proper-family structure when equality occurs. That is its proposed way past the missing geometric construction in the general case.

4. What the new manuscripts claim

M1 claims that every positive-dimensional smooth connected complex projective Fano manifold satisfies

$$\rho_X(\iota_X-1)\le\dim X.$$

There is no upper bound on dimension, no toric or homogeneous assumption, no prescribed extremal contraction, and no large-pseudoindex restriction. Equality holds if and only if

$$X\simeq\bigl(\mathbb P^{\iota_X-1}\bigr)^{\rho_X}$$

as an algebraic variety. Thus all factors in an equality example have the same dimension. The conclusion is an actual product isomorphism, not merely a birational description or a product after finite cover.

If $\iota_X=1$, the left side is zero, so the inequality is strict for positive-dimensional $X$. The substantive argument begins with $\iota_X\ge2$. For Picard number one, the inequality becomes $\iota_X\le n+1$, and equality identifies projective space, consistent with the classical extremal model.

A main intermediate claim concerns a curve degree $\beta$ represented by a nonconstant free rational map. Free means the pulled-back tangent bundle is globally generated, allowing the map to move through nearby points. Put $d=-K_X\cdot\beta$, and choose an integer $a>0$ for which $-aK_X$ is very ample. In the pseudoindex-at-least-two setting, the manuscript claims

$$\left\langle\tau_{d-2}(\mathrm{pt})\right\rangle_\beta \ge(ad)^{-(d-1)}.$$

The left side is the one-point descendant with a point condition and $\psi^{d-2}$. Its positivity and quantitative lower bound are used to control the spectrum of quantum multiplication. The manuscript does not assume that the entire virtual class of every map space is effective, or that quantum cohomology is semisimple.

5. Why the advance matters

Importance: 4/5. If correct, the result would settle a major sharp classification conjecture for smooth Fano manifolds, unifying numerous dimensional, positivity, and symmetry-based cases.

The inequality expresses a useful geometric constraint: many independent numerical directions and large minimum rational-curve degree cannot coexist in a fixed dimension. Its equality statement identifies the unique way to use the entire dimension allowance.

The proposed method also has broader interest. It extracts concrete independent curve degrees from commuting quantum operators without a semisimplicity assumption. That avoids requiring the quantum algebra to split into the simplest possible pieces and may make the strategy relevant to other Fano questions.

The rating reflects a substantial all-dimensional advance within Fano geometry, with a precise extremal classification, rather than an assertion that it classifies all Fano manifolds away from equality.

6. The main idea

Degrees create a limited budget

Choose a numerical divisor basis $D_1,\ldots,D_\rho$. Quantum multiplication by these divisors gives commuting matrices $A_j(q)$ on even cohomology, where the formal variables $q$ record curve degrees.

A contribution from curves of anticanonical degree $d$ lowers cohomological codimension by $d-1$. Cohomological codimension ranges only from zero to $n$. Therefore, if the manuscript can find a nonzero product using $\rho$ independent positive curve degrees $d_1,\ldots,d_\rho$, it obtains

$$\rho(\iota-1)\le\sum_{j=1}^{\rho}(d_j-1)\le n.$$

The first inequality uses the definition of pseudoindex. The second is the finite grading budget. Producing the nonzero independent product is the central task.

Positivity prevents excessive spectral dependence

The source studies maps through a very general point and contracts components away from the marking. This gives a morphism to a weighted projective parameter space. Boundary images have too small a dimension to contribute to the relevant degree, while the free-map locus provides a nonempty effective contribution. Keeping automorphism denominators gives the explicit descendant lower bound.

Divisor and recursion identities yield finite-shift recurrences for descendant vectors. After factorial normalization, these imply an exponential upper bound. If all joint eigenvalue tuples of the commuting matrices satisfied polynomial relations, the same recurrence would force much faster decay along suitable free curve degrees.

The lower bound excludes that decay. The manuscript concludes that one joint eigenvalue tuple has $\rho$ algebraically independent coordinates. This is more than the existence of a nonzero eigenvalue: it detects all numerical divisor directions simultaneously.

Turn spectra into curves, then classify equality

An alternating trace built from matrix-valued differentials is computed in two bases. In a common triangular basis it detects the independent eigenvalues. In the fixed cohomology basis it detects a nonzero product of curve correspondences in independent degrees. Triangularization, unlike diagonalization, is compatible with nonsemisimple commuting operators.

At equality, every selected degree must be minimal, $d_j=\iota$. The correspondence product then yields an actual chain of rational curves. Minimal degree prevents those curves from splitting into smaller positive-degree pieces, making the relevant full families proper.

The source applies the earlier ordered-chain estimate with the family hypotheses required by its erratum. Cyclically reordering the chain shows that every family covers the manifold. Occhetta’s characterization of products of projective spaces then yields the equality classification. This last rigidity theorem is an established input; the new proposed contribution is the route that supplies its hypotheses.

7. Scope, limitations, and open questions

The main theorem concerns smooth projective complex Fano manifolds. It does not assert the same unrestricted theorem for all singular log Fano pairs, nor for varieties with merely nef anticanonical bundle. Earlier singular results in special symmetric classes have their own assumptions.

When $\iota=1$, the inequality gives no useful upper bound on $\rho$. Even for larger pseudoindex, it is not a classification away from equality. Many different Fano manifolds can satisfy a strict inequality.

The descendant statement is restricted to degrees with a free representative and to the setting in which it is proved. It should not be generalized into positivity of every virtual curve count. Nor does the argument claim semisimplicity of quantum cohomology as a byproduct.

Natural further questions concern near-equality geometry, singular extensions with suitable correction terms, and other numerical inequalities accessible through independent quantum divisor eigenvalues. The source’s descendant and spectral arguments are reported as stated, without an independent verification of their positivity claims.

8. Sources and further reading

M1’s Theorem 1.1 contains the inequality and equality classification. Proposition 3.1 gives the point-descendant lower bound, Sections 4–5 obtain independent curve correspondences, and Section 6 invokes the ordered-chain and product-rigidity results at equality.

The original generalized formulation, its author-hosted erratum, Novelli’s large-pseudoindex theorem, and the spherical-variety theorem explain the principal earlier routes and their restrictions.

Source manuscripts

• M1. OpenAI, The generalized Mukai conjecture (September 24, 2026).

Sources consulted on 12 October 2026 at repository revision fd4aeeb. Download the manuscript citations.