Tangent splittings and product decompositions

The explanation

Picture a square video-game screen that wraps around. Leave through the right edge and you come back on the left. Leave through the top and you come back at the bottom. Mathematicians call this surface a torus, the surface of a donut. At every point you can split the possible directions into two families: horizontal and vertical. Now unroll the torus by laying copies of the square side by side forever. You get the whole flat plane, and the two families become a plain grid. The plane is a product: each point is one horizontal position paired with one vertical position.

Now tilt one family. Keep the vertical lines, but swap the horizontal lines for lines of slope √2. Start at a corner and follow one tilted line. Each time it wraps past the right edge, note its height. The heights come out at about 0.414, 0.828, 0.243, 0.657 and 0.071. Because √2 is not a fraction, the line never closes up. It winds forever across the donut. Yet on the unrolled plane the picture is still a grid, only slanted. The product survives, even though the lines on the donut look tangled.

Not every split of directions behaves this well. Take ordinary space with coordinates x, y and z. Let one family be "move in x". Let the other be "move in y while climbing in z at rate x". Walk a small square with sides 0.1, using only these moves. You end up directly above your start, 0.01 higher. Two such moves do not close up, so no layers of surfaces can follow them. A family whose directions do fit together into layers is called integrable. The layers are called leaves. The papers define integrability through the Lie bracket (the drift left over after such a small loop).

These toy pictures use real coordinates. The papers work with complex manifolds (spaces that look locally like several complex coordinates). The question needs a few more terms. A tangent-bundle splitting is a smooth choice, at every point, of two complementary families of directions. A Kähler manifold is a complex manifold with a compatible way to measure lengths and areas. Compact means it is finite in extent and has no edge, like the donut. The universal cover is the space you get by unrolling every loop, as the plane unrolls the torus. The question asks when a given splitting comes from a product on the universal cover. The given families must be the factor directions.

Arnaud Beauville formulated this as a conjecture for compact Kähler manifolds, in a paper published in 2000. For two families, his condition is that both are integrable. He proved major cases, including Kähler–Einstein manifolds and compact Kähler surfaces. Later work added other special hypotheses. Brunella, Pereira and Touzet treated a family of directions that is a single line, with an integrable complement. Druel, Pereira, Pym and Touzet treated the case where one family has a compact leaf with finite holonomy. Integrability alone gives a product only near each point. The obstacle was continuing those small products across the whole spaceSource 1. Compactness makes every smooth metric complete, but it does not make the two families line up with itSource 2.

The first paper proves the two-family form of Beauville's conjectureSource 3. Take a compact connected Kähler manifold of complex dimension at least two. Split its directions into two integrable families, each of positive size. Then its universal cover is a product of two connected, simply connected complex manifolds. The first factor runs exactly along the first family, and the second along the second. The proof extends small two-direction patches, one direction in each family, step by step across the space. A bound on areas, supplied by the Kähler structure, stops these patches from breaking downSource 4. The result speaks only about the universal cover. It does not say that the original space, or any finite cover of it, is a productSource 5. The factors and the leaves can be infinite in extent, like the √2 line.

The second paper asks where integrability comes from. It looks at rationally connected projective manifolds. Projective means the space can be cut out by polynomial equations. Rationally connected means two general points lie on one rational curve (the image of a complex sphere). Andreas Höring showed in 2007 that on such a space one integrable family already gives an actual product. He then conjectured that at least one family is always integrable. Campana and Peternell had settled Fano manifolds when one family has size one or two. Höring's 2026 preprint covers varieties of Fano type. The second paper proves that both families are always integrable, with no rank or positivity conditionSource 6. The paper adds that Höring's earlier theorem then makes the space itself a productSource 7. Smoothness matters here. Höring's 2026 preprint gives a singular rationally connected example with a non-integrable familySource 8.

The obstacle there was finding a way to test integrability everywhere at once. The paper builds, from each rational curve, a map of two spheres into the spaceSource 9. One sphere runs along the first family and the other along the second. Along this map, the drift from walking loops becomes a quantity that cannot change. It is zero at one marked point, so it is zero at anotherSource 10. Rational curves through enough points then force the drift to vanish everywhere. Beauville's own warning example shows why such a theorem is neededSource 11. On the product of a complex torus with a sphere, a tilted family of directions can fail to be integrable. That space is not rationally connected, so the second paper does not cover it.

How sure is this? A computer proof checker called Lean has checked formal versions of both main theorems. The first is the universal-cover product, with both families assumed integrable. The second is integrability of both families on a rationally connected manifold of dimension at least two. There the manifold comes with a projective embedding that witnesses rational connectedness. The product consequence of the second paper is outside that check. So are the first paper's further corollaries. No specialist review has been published.

What it doesn't show. This does not make a compact Kähler manifold with such a splitting, or any finite cover of it, a product.