The Minimum Cardinality of a Dependent Finite Gabor System Is Four

Theorem 1 of Heil–Ramanathan–Topiwala gives linear independence for every configuration of at most three distinct points [HRT96, pp. 2791–2792], so N_*\ge4. Theorem 1.1 gives a four-point Schwartz example, so N_*\le4. The same argument proves the Schwartz-space statement.

Since |-iuv|=1, the two matrices have the same eigenlines and the same ratio of eigenvalue moduli. From (?)–(?),

If the two eigenvalues had the same modulus \rho, then |\operatorname{tr} P|^2\le4\rho^2=4|\det P|, contrary to 27/4>39/8. Compactness gives a uniform gap.

The smaller root of the characteristic polynomial of the full return is

where the square root is the branch near 1. Indeed,

Hence the square-root argument remains in the disk of radius 13/18 about 1, and the second factor in (?) is homotopic through nonvanishing functions to 1. Equations (?)–(?) show that \operatorname{tr} R and \det R both have winding (1,-1). Their quotient therefore has winding zero, proving \operatorname{wind}(\lambda_s^{\mathrm{full}})=(0,0).

The three inequalities in (3.2) imply that the closed graph disk |\xi|\le1/4 is mapped strictly into itself by the projective action of J, with a uniform contraction factor strictly below one. The contraction mapping theorem gives a unique invariant graph. Since J is projectively P^{-1}, this graph is precisely the eigenline corresponding to the smaller-modulus eigenvalue of P, equivalently of R. Projection along the second frame direction identifies this graph with the trivial line H.

We make the perturbation estimate deliberately coarse. On the unit torus,

Only the second and third factors of (?) move with \tau. A telescoping product estimate therefore gives

For 2\times2 matrices, adjugation is linear in the entries and preserves the spectral norm up to the fixed unitary transpose identification. Hence

The explicit unit section h_0 satisfies

Differentiating the rational product gives \|\partial_xJ_\theta\|,\|\partial_\omega J_\theta\|\le6\pi. Since h=J_\theta h_0, it follows that

and the same derivative bound holds for n. Since \|w-z\|_\infty\le3\delta,

Combining (Proof)–(Proof) in (?) gives, uniformly in z,

Under (4.1) we may therefore set

and use |k_{ij}^\tau-k_{ij}^\theta|\le\eta.

Let r_0=1/4, and denote the three rational certificate bounds in (3.2) by

If

then

At the rational parameter, N_\theta\le B d_\theta and d_\theta\ge D. Hence

Thus the perturbed projective map still sends the closed radius-1/4 graph disk strictly into itself.

It remains to retain contraction. From (Proof) and (Proof), the rational coefficients satisfy |k_{ij}^\theta|\le512. Consequently

Moreover d_\tau\ge D-(1+r_0)\eta>15.4044. Using |\det(k_{ij}^\theta)|\le C d_\theta^2 and d_\theta\le d_\tau+(1+r_0)\eta, equations (?) and (3.2) give

This proves a uniform contraction on the whole parameter square.

The contraction mapping theorem now gives a unique continuous invariant graph L_\tau, and the usual fixed-point estimate gives continuous dependence on \tau. Projection of the graph onto H is a bundle isomorphism, so L_\tau is trivial. Smoothness in z follows directly by differentiating the graph iteration: at derivative order j, the only term containing the jth derivative of the previous iterate is multiplied by \partial_\xi\Phi_{\tau,z}, whose norm is uniformly <0.773; all remaining terms contain only lower derivatives. Induction in j therefore gives uniform convergence of every derivative and hence L_\tau\in C^\infty.

For a two-dimensional invertible cocycle, uniform contraction of the inverse projective graph transform on an invariant graph is the standard graph-transform criterion for a dominated splitting; the invariant graph is its unique stable line. Hence the fixed graph above is the stable line of a dominated splitting for the three-step cocycle. The cocycle identity

shows that

is another stable line for D_\tau. Indeed, iterating (?) conjugates the three-step dynamics on L_\tau(T_\tau z) to that on \widetilde L_\tau(z) by the bounded invertible map C_\tau. The stable line of a dominated splitting is unique, so \widetilde L_\tau=L_\tau. This is (4.1).

Finally, for (1.1),

which proves admissibility of the announced parameter.

Write \varepsilon=10^{-12}=1/b with b=10^{12}. Let \ell be a nearest integer to m\alpha+n\beta and set B=|m|+|n|\ge1. In the biquadratic field K=\mathbb Q(\sqrt2,\sqrt3), the number

is an algebraic integer. It is nonzero because 1,\sqrt2,\sqrt3 are linearly independent over \mathbb Q. Therefore |N_{K/\mathbb Q}(\zeta)|\ge1.

Since \ell is a nearest integer to m\alpha+n\beta, one has |\ell|=O(B). Under each of the other three real embeddings of K, the conjugate of (Proof) is therefore O(B). Taking the norm gives

for a constant C depending only on the fixed parameter. Division by 3b yields (5.1).