A "mock automorphism" is a polynomial map of n-dimensional space over a finite field that looks like it should be an invertible transformation in some algebraic sense (its Jacobian determinant equals 1 and it fixes the base field pointwise), but is not actually a true automorphism. A central question is: over which finite field extensions does such a map actually permute the points? The "extension spectrum" of such a map is the set of extension degrees where it behaves like a permutation. This paper gives a complete classification of which finite sets of positive integers can appear as such spectra for an explicit family of mock automorphisms, showing that any finite nonempty set closed under divisors works. This directly disproves a conjecture by Maubach and Willems, who believed every mock automorphism had to permute infinitely many finite extensions.
The paper also establishes sharp degree bounds. Any map satisfying the two conditions above (unit Jacobian and fixing the base field) that has only finitely many extensions where it permutes must have total polynomial degree at least p times (q+1), where p is the characteristic of the field and q is the size of the base field. The authors show this bound is tight by constructing explicit examples achieving it in dimension two and higher. In dimension one, the same bound is achieved over most fields, with a careful treatment of the remaining even-characteristic cases. They also compute exact values for the minimum degree of a well-behaved one-variable mock automorphism in small cases, and give a clean formula relating it to the minimum degree of certain permutation polynomials for primes five and larger.
The paper additionally provides a criterion that tells you when a map has a "cofinite" spectrum, meaning it permutes all but finitely many extensions, and isolates a remaining open question about behavior at primes not equal to the characteristic. A unifying observation across all the counterexamples constructed is that their geometric degree is always divisible by the characteristic p, suggesting this divisibility is a deep structural feature of mock automorphisms that fail to be true automorphisms.