Family 044 — Hikita’s conjecture: cohomology hidden inside a fixed scheme

← All digests

Family 044 · Algebraic and complex geometry
Importance 4/5 · Provisional editorial rating
Sources consulted 2026-10-12

1. The question

A space’s cohomology records its topology, while its ring of functions records algebraic geometry. Hikita’s conjecture predicts a surprising bridge: for certain paired spaces arising from quivers, the cohomology ring of one is the ring of functions on a fixed locus in the other.

The fixed locus must be understood with its full scheme structure. Counting its ordinary points would lose the information the conjecture is designed to capture. A fixed locus consisting of one visible point can still carry nonzero functions whose powers vanish, and those functions can represent positive-degree cohomology classes.

Family 044 contains one manuscript claiming the equivariant cohomological comparison for arbitrary finite quivers, including loops and repeated arrows. Its hypotheses require a stability choice yielding a smooth quotient with free gauge-group action. The assertion preserves the natural coefficient classes and all algebraic relations, not just the dimensions of vector spaces.

2. Background and essential definitions

A quiver turns linear algebra into geometry

A quiver is a finite directed graph. Give each vertex $i$ a vector space $V_i$ of dimension $v_i$, and each arrow a linear map between the corresponding spaces. Additional spaces $W_i$, of dimensions $w_i$, provide a framing: extra linear maps that anchor the representation.

The manuscript uses the representation space

$$N=\bigoplus_{a:i\to j}\operatorname{Hom}(V_i,V_j) \oplus\bigoplus_i\operatorname{Hom}(V_i,W_i),$$

with change-of-basis group

$$G=\prod_i\operatorname{GL}(V_i).$$

Passing to $T^*N=N\oplus N^*$ adds dual variables and a natural symplectic form. A moment map $\mu$ packages the equations expressing compatibility with the group action. For $(s,p)\in N\oplus N^*$ and an infinitesimal change of basis $\xi$, its convention is

$$\langle\mu(s,p),\xi\rangle=\langle p,\xi s\rangle.$$

One imposes $\mu=0$, selects stable points using a character $\nu:G\to\mathbb C^\times$, and divides by changes of basis. A character is a multiplicative one-dimensional representation. Stability excludes configurations that would give a badly behaved quotient.

The resulting Nakajima variety is

$$X=\mu^{-1}(0)^{\nu\text{-stable}}/G.$$

Here the required regularity means that semistable and stable loci coincide and the action there is free. The free-action clause is part of the hypothesis, not an automatic consequence of choosing an arbitrary character.

Two branches and an extra symmetry

The space $X$ is called the Higgs branch. The corresponding Coulomb branch is defined by the Braverman–Finkelberg–Nakajima construction: a convolution algebra built from spaces of formal Laurent-series data gives its coordinate ring. Convolution multiplies classes by composing geometric correspondences, much as matrix multiplication composes linear maps.

A flavor torus $F$, a group isomorphic to a product of copies of $\mathbb C^\times$, acts on the linear data and commutes with $G$. On the Higgs side we retain this symmetry through equivariant cohomology $H_F^*(X;\mathbb C)$. Unlike ordinary cohomology, this also contains polynomial parameters recording the torus action.

On the Coulomb side, flavor parameters produce a family $Y_{\mathfrak f}$, where $\mathfrak f$ is the Lie algebra of $F$. A topological charge grading supplies another torus action, and the stability character determines a one-parameter subgroup of that torus.

Why a fixed scheme is more than fixed points

Consider the algebra $\mathbb C[u]/(u^2)$. It describes one ordinary point, but the class of $u$ is nonzero and has square zero. Such an element is nilpotent. Replacing the algebra by the functions on its underlying point would give only $\mathbb C$.

This toy example illustrates how topology could be stored: $H^*(\mathbb P^1;\mathbb C)=\mathbb C[u]/(u^2)$, with $u$ in degree two. Forgetting the nilpotent would erase its degree-two class. This is an illustration of the algebraic issue, not a substitute for identifying the particular spaces in the theorem.

If the Coulomb algebra $A_F$ decomposes into weight spaces for the stability subgroup, the fixed scheme has coordinate ring

$$\mathbb C[Y_{\mathfrak f}^{\nu}] =A_F\big/\langle (A_F)_m:m\ne0\rangle.$$

One quotients by all nonzero-weight elements and retains whatever nilpotents remain. This differs from merely taking the weight-zero subalgebra.

3. Development history and previous partial results

A classical precursor describes Springer-fiber cohomology through scheme-theoretic intersections of nilpotent orbit closures with a Cartan subalgebra. Work of de Concini–Procesi and Tanisaki showed how a small algebraic intersection can retain substantial topological information.

Hikita formulated the fixed-scheme correspondence in the setting of symplectic duality and established examples involving the Hilbert scheme of points in the affine plane, alongside related type-A and hypertoric examples. These cases made the role of infinitesimal scheme structure particularly clear.

Nakajima proposed a mathematical construction of Coulomb branches, and Braverman–Finkelberg–Nakajima supplied the convolution-algebra definition. This gave a precise object to compare with quiver varieties beyond the first examples. Later formulations made compatibility with tautological and equivariant parameters part of the conjecture itself.

Krylov–Shlykov proved the Gieseker, or framed Jordan-quiver, case. Dumanski–Krylov compared cohomological, equivariant, and K-theoretic versions. As M1 emphasizes, their finite-type conclusion includes types A, D, and $E_6$, while $E_7$ and $E_8$ carry framing restrictions; an earlier unrestricted ADE formulation needs that qualification.

There are also quantum refinements. Dinkins–Karpov–Krylov treat a refined quantum conjecture for ADE quivers with specified minuscule framings and for the Jordan quiver. Those stronger structures have different hypotheses and should not be conflated with the cohomological result claimed here.

The gap addressed by M1 is a uniform coefficient-compatible comparison for arbitrary finite quiver data under regular stability, including loops, multiple arrows, and arbitrary commuting flavor tori.

4. What the new manuscripts claim

Theorem 1.1 asserts a canonical isomorphism of graded algebras

$$H_F^*(X;\mathbb C)\simeq\mathbb C[Y_{\mathfrak f}^{\nu}].$$

“Canonical” is specified through the coefficient maps. Let $T\subset G$ be a diagonal maximal torus and $W$ its Weyl group, which permutes coordinates at each vertex. Both sides receive natural maps from

$$R=H_{G\times F}^*(\mathrm{pt};\mathbb C) =\mathbb C[\operatorname{Lie}T\oplus\mathfrak f]^W.$$

On the Higgs side, gauge parameters become Chern classes of tautological bundles, the bundles whose fibers are the vertex vector spaces. Flavor parameters retain their equivariant meaning. On the Coulomb side, the same parameters map through its coefficient algebra and then to the fixed-scheme quotient.

M1 claims both maps are surjective and have exactly the same kernel. Consequently the isomorphism is an identification of the same quotient of $R$, rather than an unspecified algebra isomorphism. Linear gauge and flavor parameters have degree two.

The quiver may have any finite number of vertices and arrows, with loops and multiple arrows allowed. Dimension and framing vectors are arbitrary. The stability character must satisfy the regularity condition defined above. Empty stable loci are included: the corresponding ring is the zero ring on both sides.

The comparison retains nilpotents and all flavor parameters. It is not restricted to a generic parameter value, and no reduction to the underlying ordinary fixed-point set is taken.

5. Why the advance matters

Importance: 4/5. If correct, this would establish a broad form of a central bridge in symplectic duality for quiver gauge theories, beyond the earlier special graph types and framing choices.

The result would convert topological calculations into algebraic relations on a dual geometric object. Conversely, knowledge of the cohomology ring would determine the full fixed scheme, including infinitesimal information invisible in a picture of its points.

Preserving coefficients is especially useful. It tells us where natural geometric classes go and allows deformations and symmetry parameters to remain part of the comparison. An equality of Betti numbers alone would provide much less information.

The rating reflects the breadth of the quiver theorem. Its scope remains a cohomological comparison for this class of gauge groups, rather than a universal statement covering every proposed version of Hikita duality.

6. The main idea

Compare relations before comparing spaces

The manuscript first works with the fixed scheme for the entire topological torus, an intermediate object whose coordinate ring is a quotient of $R$. It compares the kernel of that quotient with the kernel defining equivariant cohomology.

Geometric tests on stable pairs show that each relation on the Coulomb side vanishes in the Higgs cohomology ring. This gives a surjection. Such a map could still discard information, so a second argument must show the source is no larger than the target in each degree.

Add loops to simplify the calculation, then remove them

M1 temporarily adds a loop at every relevant vertex. In that setting, factors contributed by the loops cancel denominators in explicit convolution formulas. A filtration indexed by multipartitions reduces the remaining size estimate to polynomial quotients and a reachability count.

Hausel’s established formula supplies the corresponding cohomological dimensions. The degree-by-degree upper bound, combined with the already constructed surjection, forces equality for the auxiliary quiver.

An auxiliary flavor parameter then separates the added loops from the original data. Localization compares the relations near a specified nonzero parameter value. Homogeneity is used to descend this comparison to the original coefficient ideals. The proposed conclusion is therefore equality before localization, including special parameter values, not merely an isomorphism after generic denominators are inverted.

Distinguish one subgroup from the whole torus

A final issue remains: the conjecture asks for the fixed scheme of the stability one-parameter subgroup. A charge can be nonzero for the full torus yet have weight zero for that subgroup. Consequently the two fixed-scheme quotients cannot simply be identified by terminology.

M1 uses complementary convolution generators, often called monopoles, and Hall-algebra generation to analyze a hypothetical surviving charge. The resulting decomposition would produce a semistable configuration with a positive-dimensional stabilizer. That contradicts the regularity hypothesis.

This establishes the claimed equality of the two defining ideals, retaining scheme structure throughout. The coefficient comparison and the fixed-scheme comparison solve different parts of the problem.

7. Scope, limitations, and open questions

The result is over complex coefficients and uses a gauge group that is a product of general linear groups attached to quiver vertices. It does not extend automatically to arbitrary reductive gauge groups; the manuscript notes failures of proposed identifications outside the stated setting.

Regular stability is essential. Singular quotients or actions with stabilizers require additional formulations. The empty case is allowed, but that does not remove the stability conditions from nonempty cases.

The algebra here is the commutative BFN algebra without the loop-rotation quantization parameter. The theorem does not claim the full quantized, quantum-cohomological, or K-theoretic comparison for arbitrary quivers. Those versions preserve more structure and remain separate questions.

Finally, identifying two rings does not identify the Higgs and Coulomb spaces themselves. The fixed scheme can be much smaller geometrically while retaining the algebra needed to encode the other side’s cohomology.

8. Sources and further reading

M1’s Theorem 1.1 states the coefficient-compatible isomorphism. Its introduction records the earlier finite-type restrictions and separates the full-torus comparison from the stability-subgroup comparison. The four-step strategy there is the basis of Section 6 above.

The primary papers linked in Section 3 explain the original conjecture, Coulomb construction, and earlier quiver cases. Optional introductions to quivers and equivariant cohomology provide more notation. The small algebra $\mathbb C[u]/(u^2)$ is the key example for remembering why nilpotents cannot be discarded.

Source manuscripts

• M1. OpenAI, The equivariant cohomological Hikita conjecture for arbitrary quivers (September 24, 2026).

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