Family 043 — P = W: two ways to organize the same cohomology

← All digests

Family 043 · Algebraic and complex geometry
Importance 4/5 · Provisional editorial rating
Sources consulted 2026-10-12

1. The question

The same topological space can carry two very different algebraic descriptions. Each description may organize its holes and higher-dimensional cycles in a different way. Nonabelian Hodge theory provides a striking example: one description uses vector bundles equipped with a field, while the other uses matrices describing transport around loops on a surface.

The $P=W$ conjecture predicts that two filtrations on their common cohomology agree. One comes from the singular fibers of the Hitchin map; the other comes from mixed Hodge theory on a character variety. Their agreement is unexpected because the correspondence between the spaces is analytic and topological, rather than algebraic.

Family 043 claims the remaining composite-rank comparison for the smooth, fixed-determinant, trace-free setting in coprime degree. Combined with earlier prime-rank theorems, the manuscript says this gives the comparison in every coprime rank. It concerns the entire rational cohomology, including a part not recovered by simply imposing fixed determinant on the known general-linear-group result.

2. Background and essential definitions

From bundles to a map of moduli spaces

Let $C$ be a smooth projective complex curve of genus $g\ge2$. Topologically, it is a closed oriented surface with $g$ handles. A rank-$n$ vector bundle $E$ attaches an $n$-dimensional vector space to every point, varying holomorphically.

A Higgs field is a holomorphic map

$$\theta:E\longrightarrow E\otimes K_C,$$

where $K_C$ is the line bundle of holomorphic one-forms. Locally, it is a matrix of one-forms. We fix a line bundle $L$ of degree $d$, require $\det E\simeq L$, and impose $\operatorname{tr}\theta=0$.

A Higgs bundle is stable if each proper nonzero subbundle preserved by $\theta$ has smaller slope, where slope means degree divided by rank. The moduli space $X$ parametrizes stable objects up to isomorphism. Assuming $\gcd(n,d)=1$ places us in the smooth setting used by the theorem.

The Hitchin map remembers the coefficients of the characteristic polynomial of $\theta$:

$$h:X\longrightarrow\mathcal A, \qquad \mathcal A=\bigoplus_{i=2}^{n}H^0(C,K_C^{\otimes i}).$$

The linear term is absent because the trace is zero. Writing $R=(n^2-1)(g-1)$, the source has complex dimension $2R$ and the base dimension $R$. The fibers form an algebraic integrable system, with important singular fibers as the characteristic polynomial degenerates.

The matrix description

On the other side, assign matrices $A_j,B_j\in\mathrm{SL}_n(\mathbb C)$ to the two standard loops around each handle. The determinant-one condition explains the notation $\mathrm{SL}_n$. The matrices satisfy

$$\prod_{j=1}^{g}[A_j,B_j]=\zeta I_n, \qquad [A,B]=ABA^{-1}B^{-1}, \qquad \zeta=e^{2\pi i d/n}.$$

Tuples are identified under simultaneous conjugation, which changes the basis used to record transport. The resulting twisted character variety is denoted $X_B$. The scalar twist retains the degree information. Nonabelian Hodge theory gives a diffeomorphism between $X$ and $X_B$, hence identifies their rational cohomology.

What is being filtered?

A filtration is a nested sequence of subspaces

$$\cdots\subset F_{k-1}H\subset F_kH\subset F_{k+1}H\subset\cdots.$$

It records more than the dimension of $H$: it assigns levels to its elements. On the Higgs side, the perverse Leray filtration $P$ organizes cohomology according to the direct image of the Hitchin map, taking the dimensions of supports and singular-fiber contributions into account. “Perverse” is a technical term for the sheaf-theoretic indexing, not a negative judgment.

On the character-variety side, Deligne’s weight filtration $W$ measures the different Hodge-theoretic weights that can occur on a noncompact algebraic variety. Smooth projective varieties have pure cohomology of a given degree; these character varieties need not.

These descriptions explain the origin of the two filtrations without requiring their sheaf-theoretic construction. The theorem compares the actual subspaces under the specified nonabelian-Hodge identification.

3. Development history and previous partial results

Hitchin’s work introduced the Higgs-bundle moduli spaces and their integrable systems. The correspondence with flat connections and representations rests on foundational work of Hitchin, Donaldson, Corlette, and Simpson.

Arithmetic calculations by Hausel–Rodriguez-Villegas revealed unexpected symmetries in the weights of character-variety cohomology. De Cataldo–Hausel–Migliorini formulated $P=W$ and established the rank-two case. Their comparison explained weight symmetries on one side through the geometry of the Hitchin map on the other.

Higher-rank progress used tautological classes, cohomology classes naturally extracted from universal bundles. Results of Markman and others provided generators in the general-linear setting, while Shende determined their weights. De Cataldo–Maulik–Shen proved the general-linear comparison in genus two and obtained further bounds in arbitrary genus.

Maulik–Shen then proved the coprime $\mathrm{GL}_n$ conjecture in arbitrary rank and genus. Other proofs use the operator methods of Hausel–Mellit–Minets–Schiffmann and the Fourier-transform methods of Maulik–Shen–Yin.

Fixed determinant introduces an extra issue. The finite group of degree-zero $n$-torsion line bundles acts by tensoring $E$. The known general-linear result controls the invariant part of the fixed-determinant cohomology, but does not automatically recover its complementary part. De Cataldo–Maulik–Shen’s prime-rank work, combined with the later general-linear theorem, handles prime rank.

For composite rank, earlier endoscopic correspondences related the missing pieces to geometry on covers of the curve and were compatible with perverse filtrations. M1 identifies the remaining task as a comparison on full cohomology, rather than another proof restricted to the invariant subspace.

4. What the new manuscripts claim

Equality on every cohomological degree

For every composite $n\ge4$, every $g\ge2$, every degree $d$ coprime to $n$, and every fixed determinant $L$ of that degree, Theorem 1.1 of M1 claims

$$P_kH^m(X_B,\mathbb Q) =W_{2k}H^m(X_B,\mathbb Q) =W_{2k+1}H^m(X_B,\mathbb Q)$$

for all $m,k\ge0$. The perverse filtration has been transported from $X$ through nonabelian Hodge theory.

The factor two is part of the theorem, not an optional convention that can be dropped. M1 uses the standard Hitchin normalization in which the perverse index runs from zero through $2R$. Equality with both weight steps says no additional piece appears at the intervening odd step.

Combining this claim with established prime-rank cases gives the smooth coprime fixed-determinant statement for every $n\ge2$.

The complementary cohomology is included

Let

$$\Gamma=\operatorname{Pic}^0(C)[n]$$

be the group of line bundles whose $n$-th power is trivial. Tensoring by an element of $\Gamma$ preserves the fixed determinant. Averaging its action projects onto invariant cohomology; the kernel is called variant cohomology.

M1 asserts the filtration equality before taking invariants. After allowing complex coefficients, this includes every nontrivial character of $\Gamma$, meaning every simultaneous eigenspace for this finite abelian group action. A proof of equality only on the average would not establish the advertised theorem.

Corollary 9.2 also derives an exact weight-image compatibility for the specified endoscopic operator associated with a cyclic unramified cover of $C$. This uses the new full equality together with the known general-linear comparison on the cover side. It is a consequence for that particular transported correspondence, not a blanket assertion about every operator on cohomology.

5. Why the advance matters

Importance: 4/5. If correct, the result would finish an important fixed-determinant portion of a central comparison in nonabelian Hodge theory, including cohomology invisible to the simplest reduction from $\mathrm{GL}_n$.

It is stronger than matching Betti numbers or even the dimensions of individual graded pieces. Two filtrations can have equally sized steps while placing different vectors in those steps. The statement says the subspaces themselves coincide under a geometrically meaningful correspondence.

This allows information to travel between descriptions: the geometry of Hitchin fibers constrains Hodge weights, while weight calculations can organize the topology of the integrable system. The claimed method also illustrates how a common operator can identify structures that initially arise from unrelated constructions.

The rating reflects this substantial completion while retaining credit for the earlier general-linear and prime-rank theorems. It is not a claim to establish every version of $P=W$ for arbitrary groups or singular moduli spaces.

6. The main idea

Use one raising operator and construct one lowering operator

Cup product with a distinguished degree-two cohomology class gives an operator $e$. On each side, a version of hard Lefschetz supplies symmetry around the middle level: repeated applications of $e$ connect corresponding levels below and above the center.

M1’s strategy is to construct a single operator $f$, lowering cohomological degree by two, that simultaneously satisfies

$$f(P_k)\subset P_{k-2}, \qquad f(W_j)\subset W_{j-4}, \qquad [[f,e],e]=-2e.$$

Here $[a,b]=ab-ba$ is the commutator of linear operators. Together with the two Lefschetz symmetries, these relations allow a linear-algebra argument to recover both filtrations from eigenspaces of the same operator $[e,f]$. The task is therefore geometric construction of a genuinely common $f$, not merely dimension counting.

Move a pole and compare two limits

The construction enlarges the moduli problem by allowing a logarithmic pole at a moving point of the curve and choosing a full flag there. A full flag is a nested sequence of subspaces of dimensions one through $n$. The logarithmic connection parameter interpolates between Higgs bundles and connections.

The manuscript modifies two eigenline lattices in opposite directions, preserving the determinant condition. On the connection side, the resulting translation is interpreted through monodromy as eigenvalues move. A suitable logarithm of that monodromy, followed by operations along the flag correspondence, produces the proposed lowering operator.

Different arguments control its two filtrations. A finite diagram of algebraic fibers supplies the weight bookkeeping; specialization over a Higgs parameter supplies the perverse bound. M1 stresses that the diagram assigns weight zero to the cochains of the moving-point base. Those are not the ordinary Hodge weights of that curve viewed as an algebraic variety.

Finally, a characteristic-class calculation establishes the commutator relation. Throughout, the construction must retain full cohomology. Averaging over $\Gamma$ too early would discard the very classes the new theorem needs to reach.

7. Scope, limitations, and open questions

The hypotheses include a smooth projective complex curve of genus at least two, fixed determinant, trace-free Higgs field, and coprime rank and degree. Coprimality is important for smoothness. The claim does not settle versions involving noncoprime degree, where singular moduli spaces and intersection cohomology may enter.

The coefficients are rational. An integral comparison would need to address torsion and is not asserted here. The theorem also does not turn the nonabelian-Hodge diffeomorphism into an algebraic isomorphism: it matches these specific cohomological structures despite the distinct algebraic descriptions.

Other structure groups, parabolic conditions, and singular variants require their own statements. The endoscopic consequence retains its designated operator and normalization. These boundaries leave a broader comparison program beyond the smooth coprime $\mathrm{SL}_n$ case addressed by this family.

8. Sources and further reading

M1’s Theorem 1.1 states the composite-rank equality; its introduction explains the variant-cohomology obstruction and the common-operator strategy. Section 9 combines prime and composite ranks and gives the endoscopic consequence.

The primary papers linked in Section 3 trace the development from rank two to the general-linear and prime-rank cases. Optional background on Higgs bundles helps with the moduli-space language. The central comparison is the equality of nested subspaces displayed in Section 4, with its factor-two indexing.

Source manuscripts

• M1. OpenAI, P=W in composite rank for fixed determinant (September 24, 2026).

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