Non-Integrable Almost-Complex Structures for Homogeneous Spaces

Let $M$ be a smooth homogeneous $G$-space for a Lie group $G$, and let $J$ be a $G$-invariant almost-complex structure for $M$. Do there exist succinct sufficient (and neccessary) conditions for $J$ to be integrable? Besides the six sphere, what other examples of a non-integrable invariant almost-complex structure for a smooth homogeneous space are there? Do there exist non-integrable almost-complex structures for any flag manifolds?

-
When you give flag manifolds as examples of homogeneous spaces, do you mean homogeneous for $G$ the compact group? (I'm pretty sure you are.) – Allen Knutson Nov 19 '11 at 21:10
Yes, that's what I mean – Ago Szekeres Nov 19 '11 at 22:53

Oops, I was thinking of a Lie group. Edit: On a Lie group, it is pretty easy to test for integrability. Take a basis of complex linear left invariant 1-forms (i.e. left translate from a choice of such 1-forms on the Lie algebra). Then compute their exterior derivatives. You have an integrable almost complex structure if and only if the (0,2) parts of the exterior derivatives all vanish. To see this, you express the exterior derivatives in linear combinations of the wedge products of the original 1-forms and their complex conjugates. So examples are easy to check, and only require the differential familiar from Lie algebra cohomology, i.e. a pure Lie algebra calculation.

On a homogeneous space $G/H$ the computation is a little trickier. You take $\omega=g^{-1}dg$, the left invariant Maurer Cartan form on $G$, and then $\omega+\mathfrak{h}$ is semibasic for the quotient map $G \to G/H$ and splits into a complex linear part and a conjugate linear part on each tangent space of $G/H$. Let $\eta$ be the complex linear part. Pick a splitting of $\mathfrak{g}$ into $\mathfrak{g}/\mathfrak{h}$ and some complement, identified with $\mathfrak{h}$, and let $\Omega$ be the projection of $\omega$ to complement. The equation $d \omega + (1/2)[\omega,\omega]=0$ gives an equation $d \eta = - \Omega \wedge \eta + a \eta \wedge \eta + b \eta \wedge \bar{\eta} + c \bar{\eta} \wedge \bar{\eta}$. The Nijenhuis tensor vanishes just when $c=0$. I am pretty sure that the generic invariant almost complex structure on a flag manifold (invariant under the compact form of the automorphism group) is not integrable, but I would have to check.

-
Thanks for your answer Ben. But to have a basis of left-invariant wouldn't my homogeneous need to be parallelizable? Or do you mean on the Lie group? In which case I don't understand your answer. – Ago Szekeres Nov 19 '11 at 22:59
All clear now, thanks – Ago Szekeres Nov 20 '11 at 16:10

Presumably, you are looking for something other than "the Nijenhaus tensor, possibly simplified by the action of $(\rho_{G})_*$"?

In the local case, there have been a few studies constructing families of almost-complex manifolds in low dimensions, along with conditions for varying amounts of not-quite-integrability. You might look at http://projecteuclid.org/euclid.ajm/1175789046 (Robert Bryant's AJM paper on almost-complex 6-manifolds and the references therein). However, this only addresses the local question and even then would require refinement to study the case of homogeneous or flag manifolds. I suspect very little is known otherwise.

-