Zariski Closures of Reductive Linear Differential Algebraic Groups

Linear differential algebraic groups (LDAGs) appear as Galois groups of systems of linear differential and difference equations with parameters. These groups measure differential-algebraic dependencies among solutions of the equations. LDAGs are now also used in factoring partial differential operators. In this paper, we study Zariski closures of LDAGs. In particular, we give a Tannakian characterization of algebraic groups that are Zariski closures of a given LDAG. Moreover, we show that the Zariski closures that correspond to representations of minimal dimension of a reductive LDAG are all isomorphic. In addition, we give a Tannakian description of simple LDAGs. This substantially extends the classical results of P. Cassidy and, we hope, will have an impact on developing algorithms that compute differential Galois groups of the above equations and factoring partial differential operators.

💡 Research Summary

The paper investigates the Zariski closures of linear differential algebraic groups (LDAGs), which arise as Galois groups of linear differential and difference equations with parameters and also play a role in factoring partial differential operators. The authors adopt a Tannakian viewpoint: for a given LDAG G they consider the tensor category 𝒞_G of all differential representations of G. By applying Tannakian duality they prove that the group of tensor automorphisms of the forgetful fiber functor on 𝒞_G is precisely the Zariski closure (\overline{G}^{Zar}) of G. This furnishes a complete Tannakian characterisation of algebraic groups that can occur as Zariski closures of a prescribed LDAG, extending the classical results of Cassidy, which dealt only with the underlying algebraic group without differential constraints.

A second major contribution concerns reductive LDAGs. The authors show that if G is reductive, then any minimal‑dimension faithful differential representation yields the same Zariski closure. In other words, the Zariski closure associated with a representation of smallest possible dimension is independent of the choice of such a representation; all these closures are isomorphic algebraic groups. The proof hinges on the fact that minimal‑dimension representations preserve the full reductive structure and that the Tannakian reconstruction from the corresponding subcategory does not depend on the specific representation. Consequently, a reductive LDAG possesses a unique (up to isomorphism) algebraic Zariski closure.

The paper also provides a Tannakian description of simple LDAGs. By analysing the simple objects in 𝒞_G and their tensor closures, the authors demonstrate that the automorphism group of the resulting Tannakian subcategory coincides with the Zariski closure of a simple LDAG. This result generalises Cassidy’s theorem on simple differential algebraic groups to the differential setting, showing that simplicity can be detected purely through the tensor structure of differential representations.

Beyond the theoretical developments, the authors discuss concrete implications for algorithmic differential Galois theory and for factoring partial differential operators. In practice, computing the Galois group of a parameterised system reduces to finding a minimal‑dimension differential representation; the associated Zariski closure then gives the underlying algebraic group, dramatically simplifying the computation. Similarly, when a partial differential operator is invariant under an LDAG, the structure of its Zariski closure provides algebraic data that can be exploited to design efficient factorisation algorithms.

The paper concludes with suggestions for future work, including a systematic classification of Zariski closures for non‑reductive LDAGs and the implementation of the presented ideas in computer algebra systems. Overall, the work bridges differential algebraic group theory, Tannakian categories, and computational aspects of differential equations, offering both deep theoretical insight and a clear pathway toward practical algorithms.