Higher Haantjes Brackets: Geometry & Integrability

• Higher Haantjes brackets are an infinite family of symmetric geometric brackets that extend the Frölicher–Nijenhuis bracket, unifying classical invariants such as the Nijenhuis and Haantjes torsions.
• Their recursive construction defines vector-valued 2-forms that govern the integrability of eigen-distributions and facilitate block-diagonalisation of (1,1)-tensor fields.
• The vanishing of these higher torsions implies generalized Frobenius integrability, providing local normal forms for both semisimple and non-semisimple tensor operators.

Higher Haantjes brackets constitute an infinite class of geometric brackets generalizing the classical Frölicher–Nijenhuis bracket. Developed within the context of smooth manifolds $M$, these brackets induce a hierarchy of vector-valued $2$-forms tied closely to the integrability of eigen-distributions of $(1,1)$-tensor fields. The construction unifies and extends known invariants such as the Nijenhuis torsion and the Haantjes torsion, offering new algebraic and analytic tools for analyzing the block-diagonalisation and integrability of tensor operators without requiring detailed spectral information. The vanishing of a suitable higher Haantjes torsion yields generalized Frobenius integrability for distributions, encompassing both semisimple and non-semisimple cases and providing local normal forms.

1. Inductive Construction of Higher Haantjes Brackets

Let $M$ be a smooth manifold and $A, B$ be $(1,1)$-tensor fields, with $X, Y \in \mathfrak X(M)$ vector fields. The family of higher Haantjes brackets, denoted $\mathcal H^{(m)}_{A,B}(X,Y)$ for $m \ge 1$, is defined by induction:

• For $m=1$, the bracket coincides with the Frölicher–Nijenhuis bracket:

$$\mathcal H^{(1)}_{A,B}(X,Y) := [A,B](X,Y) = (AB+BA)[X,Y] + [AX,BY] + [BX,AY] - A([X,BY]+[BX,Y]) - B([X,AY]+[AX,Y]).$$

• For $m \ge 2$, the recursive formula is:

$$\begin{aligned} \mathcal H^{(m)}_{A,B}(X,Y) &= (AB+BA) \mathcal H^{(m-1)}_{A,B}(X,Y) + \mathcal H^{(m-1)}_{A,B}(A X,B Y) + \mathcal H^{(m-1)}_{A,B}(B X,A Y) \ &\quad -A\bigl(\mathcal H^{(m-1)}_{A,B}(X,B Y) + \mathcal H^{(m-1)}_{A,B}(A X,Y)\bigr) - B\bigl(\mathcal H^{(m-1)}_{A,B}(X,A Y) + \mathcal H^{(m-1)}_{A,B}(B X,Y)\bigr). \end{aligned}$$

It follows by induction that the bracket $\mathcal H^{(m)}_{A,B}$ is symmetric in $A, B$ for all $m$, but fails to be $\mathcal C^\infty(M)$-bilinear when $m \ge 2$.

2. Generalized Nijenhuis Torsions

Setting $A = B = N$ in the above, one obtains higher generalized torsions of a $(1,1)$-tensor $N$:

$T_N^{(m)}(X,Y) := \mathcal H^{(m)}_{N,N}(X,Y).$

Notable specific cases include:

• $m=1$, recovering twice the Nijenhuis torsion:

$$T_N^{(1)}(X,Y) = N^2[X,Y] + [NX,NY] - N([X,NY] + [NX,Y]).$$

• $m=2$, yielding the Haantjes torsion:

$$T_N^{(2)}(X,Y) = N^2 T_N^{(1)}(X,Y) + T_N^{(1)}(N X, N Y) - N\bigl( T_N^{(1)}(X, N Y) + T_N^{(1)}(N X, Y) \bigr).$$

Moreover, for arbitrary $m \ge 1$,

$$T_N^{(m)}(X,Y) = \sum_{p=0}^m\sum_{q=0}^m (-1)^{p+q} \binom{m}{p}\binom{m}{q} N^{p+q} \left[ N^{m-p}X,\ N^{m-q}Y \right].$$

This polynomial expansion formula generalizes classical torsions and produces an infinite hierarchy of higher-order invariants.

3. Integrability via Vanishing Higher Torsion

Let $N$ be a $(1,1)$-tensor field with real spectrum, and denote the generalized eigendistribution for eigenvalue $\lambda$ (with Riesz index $p_\lambda$) as

$$\mathcal D_{\lambda} = \ker(N - \lambda I)^{p_\lambda}.$$

The generalized Haantjes integrability theorem asserts:

• If there exists $m \ge 1$ such that

$$T_N^{(m)}(X,Y) = 0,\quad \forall X,Y \in \mathfrak X(M),$$

then 1. Each eigendistribution $\mathcal D_\lambda$ is involutive (Frobenius-integrable). 2. Any finite direct sum $$\mathcal D_{\lambda_1} \oplus \cdots \oplus \mathcal D_{\lambda_r}$$ is also involutive.

The proof applies annihilators to the torsion formula, showing that the vanishing of $T_N^{(m)}$ enforces closure under commutators for sections. This result generalizes Haantjes's classical theorem ($m=2$) to arbitrary $m$ and non-semisimple $N$.

4. Block-Diagonalisation of Tensor Operators

Under the vanishing condition $T_N^{(m)} \equiv 0$, it follows that there exists a local coordinate chart adapted to the web of generalized eigendistributions such that the matrix representation of $N$ is block-diagonal:

$$N \simeq \begin{pmatrix} N_1(x^1) & 0 & \cdots & 0 \ 0 & N_2(x^2) & \ddots & \vdots \ \vdots & \ddots & \ddots & 0 \ 0 & \cdots & 0 & N_s(x^s) \end{pmatrix}$$

Here, each block $N_i$ acts on a subset of $r_i$ coordinates associated to the $i$-th eigendistribution. The construction of the adapted local chart follows from Frobenius integrability: choosing $r_i$ functions constant along all other eigendistributions yields a full chart, and $N$ preserves each coordinate leaf.

5. Classical and Low-Level Cases

The hierarchy of higher Haantjes brackets recovers classical cases at low levels:

• $k=1$: Frölicher–Nijenhuis bracket ($\mathcal H^{(1)}$), giving twice the Nijenhuis torsion for $A = B = N$.
• $k=2$: The Haantjes bracket ($\mathcal H^{(2)}$), yielding the Haantjes torsion for $A = B = N$.

Explicitly,

$$\mathcal H^{(1)}_{A,B}(X,Y) = (AB+BA)[X,Y] + [AX,BY] + [BX,AY] - A([X,BY] + [BX,Y]) - B([X,AY] + [AX,Y]).$$

The $k=2$ bracket is constructed recursively using $\mathcal H^{(1)}$. These cases serve as prototypes, while higher $k \ge 3$ yield genuinely new invariants with the same integrability implications upon vanishing.

6. Algebraic Structure and Applications

Each higher Haantjes bracket is symmetric in the pair $(A, B)$, though not $\mathcal C^\infty(M)$-bilinear beyond $m = 1$. The framework provides a systematic approach to generating invariants of tensor fields that probe the analytic and differential structure of distributions, without dependence on spectral knowledge. Accordingly, higher torsions are valuable in the classification and normalization problems for operators, with direct implications for block-diagonalization and characteristic foliations.

A plausible implication is the potential to extend these integrability and normalization criteria to settings beyond smooth manifolds, wherever $(1,1)$-tensor structures and generalized eigen-distributions arise. The comprehensive nature of this hierarchy situates higher Haantjes brackets as fundamental objects in the modern theory of integrable systems and differential geometry (Tempesta et al., 2018).