Towards non-commutative crepant resolutions of affine toric Gorenstein varieties

The Big Picture

Imagine trying to iron out a wrinkle in fabric, but every time you smooth one spot, it pops up somewhere else in a more complicated form. That’s roughly what mathematicians face when resolving the singularities of affine toric Gorenstein varieties: geometric objects defined by grids of integer points that show up in string theory, mirror symmetry, and combinatorial algebra.

Singularities are points where the space pinches or becomes pathological. The standard fix, a crepant resolution, removes them without introducing extra complexity. But crepant resolutions don’t always exist.

For decades, an alternative approach has gained traction: replace the singular space with a non-commutative algebraic object (a ring where the order of multiplication matters) that encodes the same geometric information. These are non-commutative crepant resolutions, or NCCRs. A long-standing conjecture holds that every affine Gorenstein toric variety should have one. IAIFI mathematicians Aimeric Malter and Artan Sheshmani have now proved this conjecture for a broad new family of these spaces, unifying and extending previous partial results into a single theorem.

Key Insight: By showing that the endomorphism ring of a partial tilting complex is itself a non-commutative resolution, the authors handle singularities that resist geometric treatment, connecting derived categories, toric geometry, and mirror symmetry.

How It Works

The strategy rests on tilting theory, a toolkit from homological algebra. A tilting complex on a geometric space captures its full derived category (a complete catalog of the space’s geometric data, tracking objects and all the ways they relate and transform). Find one, and you can often build a non-commutative resolution.

Malter and Sheshmani’s central result (Theorem 3.12) proceeds in four steps:

1. Start with a cone. Every affine toric Gorenstein variety comes from a cone $\sigma = \text{Cone}(P \times {1})$ over a polytope $P$ (a multidimensional polygon with integer-coordinate vertices). The ring of functions on the variety is $R = k[\sigma^\vee \cap M]$.

2. Triangulate the polytope. A regular triangulation of $P$ defines a fan $\Sigma$ (a collection of cones encoding the variety’s local geometry), which gives rise to a toric DM stack $X_\Sigma$. This is a generalized space that handles singularities gracefully by allowing points to carry extra symmetry data.

3. Find a partial tilting complex. On $X_\Sigma$, construct a partial tilting complex $T$ whose endomorphism algebra $\Lambda = \text{End}{X\Sigma}(T)$ has finite global dimension, meaning any module over $\Lambda$ admits a projective resolution that terminates in finitely many steps.

4. Conclude the NCCR exists. Under these conditions, $\Lambda$ is an NCCR for $R$.

This framework generalizes two major earlier results (one by Špenko and Van den Bergh, another by Iyama and Wemyss) into a single theorem. Špenko–Van den Bergh required a full tilting bundle on the Cox stack of a simplicialization of $\sigma$. Iyama–Wemyss required a projective, birational, crepant map. Theorem 3.12 replaces both conditions with the weaker requirement of a partial tilting complex of finite global dimension.

To handle cones over polytopes with interior lattice points, the authors study total spaces of canonical bundles over toric varieties. If $X$ is a simplicial projective toric variety, the total space $V = \text{tot}, \omega_X$ is itself toric, and its fan encodes a Gorenstein cone. They prove (Theorem 4.12) that if $X_\Sigma$ carries a full tilting complex $T$ and certain cohomology vanishing conditions hold, a tilting complex also exists on $X_V$. The conditions are checkable in practice, and the authors verify them through explicit examples.

Why It Matters

The open conjecture that every affine Gorenstein toric variety has an NCCR draws on algebraic geometry, representation theory, and string theory. Malter and Sheshmani prove it for all reflexive polytopes with at most $n + 2$ vertices (Theorem 5.1), building on results of Borisov and Hua.

Reflexive polytopes are tied to Fano varieties, the positive-curvature spaces that appear in string compactifications. This makes the result directly relevant to theoretical physics and mirror symmetry.

The work also fits into a broader program in derived-category theory. The Bondal–Orlov and Kawamata conjecture says that all crepant resolutions of a given space are derived equivalent: the deep invariants of singular spaces live at the level of derived categories, not individual resolutions. NCCRs make this idea concrete. They are algebraic stand-ins for geometric resolutions when no geometric resolution exists.

Bottom Line: Malter and Sheshmani unify two major threads of non-commutative algebraic geometry into one framework, proving NCCR existence for a new class of singular spaces. Their approach is constructive enough to guide explicit computations in mirror symmetry and toric geometry.