Virasoro conjecture

Virasoro conjecture

OpenAI's claim

Manuscript 1 (family 065, 2026-09-24): Ordinary descendant Virasoro constraints are claimed for smooth projective complete intersections in every genus and curve class, retaining primitive and odd cohomology without semisimplicity. Manuscript 2 (family 065, 2026-10-05): Full descendant Virasoro constraints are claimed to pass from any smooth projective base to projectivizations of arbitrary nonsplit algebraic bundles, including all genera, integral curve classes, and odd insertions.

Mathematical statement of the claimed result

Mathematical claim 1.

Let $X\subset\mathbb P^N_{\mathbb C}$ be any smooth complete intersection. Let $F_g^X$ be its connected ordinary, unreduced genus-$g$ descendant potential, with cotangent classes at stable-map markings, arbitrary cohomology insertions and separate Novikov symbols for actual integral curve classes. Set $Z_X=\exp(\sum_{g\ge0}\hbar^{g-1}F_g^X)$. OpenAI claims the formal coefficientwise identities \[ L_k^X Z_X=0\qquad(k\ge-1). \] Here $L_k^X$ are the manuscript's normally ordered Virasoro operators built from the Poincaré pairing, first Hodge grading $\mu|_{H^{p,q}}=p-\dim(X)/2$ and $R=c_1(TX)\cup$, with its stated scalar normalization. The quantifiers include all genera, individual curve classes and primitive or odd insertions; these are descendant rather than ancestor constraints.

Mathematical claim 2.

For a smooth connected projective complex variety $B$, an algebraic vector bundle $E$ of rank $r\ge2$, and $X=\mathbb P_B(E)$ parametrizing lines in the fibers, OpenAI claims \[ \left[L_k^B Z_B=0\text{ for all }k\ge-1\right]\ \Longrightarrow\ \left[L_k^X Z_X=0\text{ for all }k\ge-1\right]. \] For each target $Y$, $Z_Y$ is its ordinary descendant partition function $\exp(\sum_g\hbar^{g-1}F_g^Y)$ with all cohomology classes, stable-map cotangent classes and actual effective integral curve labels. The normally ordered $L_k^Y$ use $\mu_Y|_{H^{p,q}}=p-\dim(Y)/2$, $R_Y=c_1(TY)\cup$, and zero-mode scalar $\chi(Y)/16-\operatorname{str}(\mu_Y^2)/4$; other modes have no scalar correction. The conclusion is coefficientwise in every genus, curve class and descendant list.

Outline of the argument

Manuscript 1 (family 065, 2026-09-24): Relative caps create and detect bounded seam-contact states with coefficientwise finite inversions. Pairing each primitive error tensor with its conjugate realizes a squared norm through these tests; Hodge--Riemann positivity then kills the full tensor without limiting the number of primitive insertions.

Manuscript 2 (family 065, 2026-10-05): Localization on the completion by a trivial summand separates the base and projective-bundle blocks. The logarithm of an equivariant shift cancels coefficient differentiation in the grading, giving quantizable modes; ancestor bounds continue their identities coefficientwise, and quantum Riemann--Roch identifies the ordinary constraints on both components.

Formal verification

Family 065 has no Lean scope page in this snapshot; no formalization of these manuscript results is listed.

This site has not executed the Lean files or checked their dependencies. These descriptions report the scope of the linked declarations, rather than a new verification of the proofs.

Status caveat

OpenAI attributes these results to an internal model. Its release notice cautions that results lacking full formalization can contain mistakes. This site has not independently assessed the proofs; a solved label records the scope of the reported claim.

Sources

• OpenAI, Virasoro Constraints for Projective Complete Intersections, 2026-09-24; citation key OAI:Virasoro-Constraints-for-Projective-Complete-Intersections-September-24-2026. PDF (GitHub); PDF (direct); Manuscript page and citation; Latest version; Statement source: thm:main.
• OpenAI, Virasoro Constraints under Projectivization, 2026-10-05; citation key OAI:Virasoro-Constraints-under-Projectivization-October-5-2026. PDF (GitHub); PDF (direct); Manuscript page and citation; Latest version; Statement source: thm:main.
• OpenAI catalogue entry.

Virasoro conjecture

Definitions and mathematical statement

Choose a homogeneous cohomology basis of $X$. Descendant Gromov--Witten correlators integrate products of evaluation classes and powers $\psi_i^{a_i}$ of cotangent-line classes over the virtual fundamental classes of stable-map moduli spaces $\overline{\mathcal M}_{g,n}(X,\beta)$. Their generating functions $F_g$ form \[ Z_X=\exp\left(\sum_{g\ge0}\hbar^{g-1}F_g\right) \] as a formal series over the Novikov ring recording effective curve classes. The conjecture asserts $L_k^XZ_X=0$ for every $k\ge-1$, where $L_k^X$ is exactly the Eguchi--Hori--Xiong--Katz operator in Getzler, Section 1.4, including constant terms and signs for odd cohomology. Those operator formulas are incorporated by reference rather than replaced by the Virasoro commutation relations alone. Equality is coefficientwise in all descendant variables, genera, and curve classes.

Formulation and concept references: [S1].

Short English statement

Do the total descendant Gromov–Witten invariants of every smooth projective complex variety satisfy all of the specified Virasoro differential constraints?

Research attempt: two operator constraints and a coefficient-level obstruction

Status and source dependence. The target fixes the actual Eguchi--Hori--Xiong--Katz operators, including odd-cohomology signs. Getzler [S1] proves their Virasoro commutation relations and explicitly records the reduction to $L_{-1}$ and $L_2$. We rederive that reduction and the even-variable coefficient equations to isolate the missing enumerative identity. Neither is asserted as a new theorem about Gromov--Witten invariants.

1. Why two specified constraints suffice

For the operators in the source, $[L_m,L_n]=(m-n)L_{m+n}$ in the permitted range. If a formal vector $Z$ obeys $L_{-1}Z=L_2Z=0$, then \[L_1Z=-\tfrac13[L_{-1},L_2]Z=0,\qquad L_0Z=-\tfrac12[L_{-1},L_1]Z=0.\] For $k\ge2$, $[L_1,L_k]=(1-k)L_{k+1}$, whose coefficient is nonzero over characteristic zero. Induction gives all higher constraints. These manipulations are coefficientwise compositions in the formal-series convention where the source defines the operators. They do not introduce a convergence claim for an analytic partition function.

The logical role of this computation is limited: the commutators propagate annihilation once it has been proved. They do not establish even the first annihilation. For example, $D_m=-x^{m+1}\partial_x$ on Laurent polynomials satisfies the same commutator formula, but $D_m(x)=-x^{m+1}\ne0$. Thus replacing the target's concrete operators and correlators by an abstract Virasoro representation would lose the problem.

2. An explicit coefficient computation

To display the structure without suppressing superalgebra signs, restrict this computation to commuting variables. Let \[\mathcal L=\frac{\hbar}{2}\sum_{i,j}A_{ij}\partial_i\partial_j+ \sum_i B_i(t)\partial_i+\frac{C(t)}{2\hbar}+D,\qquad F=\sum_{g\ge0}\hbar^{g-1}F_g,\quad Z=e^F,\] with coefficients independent of $\hbar$; take finite sums or a coefficientwise well-defined formal operator. The product rule gives the exact identity \[Z^{-1}\mathcal L Z=\frac{\hbar}{2}\sum_{i,j}A_{ij}(F_{ij}+F_iF_j)+\sum_i B_iF_i+ \frac{C}{2\hbar}+D.\] Writing $F_{-1}=0$, its coefficient at $\hbar^{g-1}$ is \[\frac12\sum_{i,j}A_{ij}\left(\partial_i\partial_jF_{g-1}+ \sum_{h=0}^{g}\partial_iF_h\,\partial_jF_{g-h}\right) +\sum_i B_i\partial_iF_g+\frac12\delta_{g0}C+\delta_{g1}D.\tag{1}\] In particular, the genus-zero quadratic term survives: it comes from the product of two $\hbar^{-1}$ terms followed by multiplication by $\hbar$. Dropping this contribution, or placing the constant $D$ in genus zero, gives the wrong recursion.

3. What a recursive proof would still need

Even when the lower-genus $F_h$ are known, $1$ contains terms involving both $F_0$ and $F_g$. It is an equation for $F_g$ with coefficients determined by genus zero, not a statement that an arbitrary $F_g$ satisfies it. A proof of the target must establish those equations for the actual virtual-cycle integrals, including each Novikov degree and descendant monomial. The string equation addresses $L_{-1}$ in the source's conventions; the unreduced task is the genuine $L_2$ identity. The commutator reduction provides no substitute for it.

For odd cohomology, ordered superderivatives and Koszul signs must be restored exactly as in [S1]. The displayed commuting-variable calculation is a diagnostic subcase, not an alternative definition of $L_k^X$. It also does not assume that arbitrary curve classes can be reordered across an infinite sum without formal finiteness.

4. Verification and gap

The numerical factors in the first step follow from $-1-2=-3$ and $-1-1=-2$; the induction never divides by the vanishing coefficient at $k=1$. Formula $1$ follows by two explicit differentiations and tracking powers of $\hbar$. The remaining gap is to prove $L_2^XZ_X=0$ for every target variety, in the full signed operator convention. No new general constraint is proved here.

Sources

• S1 E. Getzler, The Virasoro conjecture for Gromov-Witten invariants, arXiv version 4 $1999$. https://arxiv.org/pdf/math/9812026. Formulation source.