A small frame and a certificate of its injectivity

Cynthia Vinzant

I Translation to polynomials

Following the set up of , we translate the injectivity of measurements into a condition on the solutions of a system of polynomial equations.

A useful step is the following reformulation of injectivity by Bandeira et. al. [4, Lemma 9]. They observe that a frame Φ\Phi defines injective measurements if and only if the linear space

does not contain any non-zero Hermitian matrices of rank ≤2\leq 2. The existence of rank ≤2\leq 2 Hermitian matrices in LΦ\mathcal{L}_{\Phi} can be rephrased as the existence of real roots of a certain system of polynomial equations as follows.

Any 4×44\times 4 Hermitian matrix can be written as Q=Q=

where x11,…,y34x_{11},\ldots,y_{34} are real numbers. We will write our polynomial condition for injectivity in the 16 variables xjk,yjkx_{jk},y_{jk}. For 1≤j,k≤41\leq j,k\leq 4, let mjkm_{jk} denote the determinant of the 3×33\times 3 matrix obtained by removing the jjth row and kkth column from the matrix QQ. The matrix QQ has rank ≤2\leq 2 when all these minors mjkm_{jk} equal zero.

II A small injective frame

The frame Φ=(ϕ1,…,ϕ11)\Phi=(\phi_{1},\ldots,\phi_{11}), consisting of the rows of the matrix

Apart from the coordinate vectors, the vectors of Φ\Phi were chosen to have first coordinate 1 and otherwise found by a random search. This matrix is by no means unique, as further discussed in Section III.

Φ\Phi in (2) defines injective measurements.

Using the set-up of Section I, it suffices to show the equations (1) have no non-zero real solution.

Such a certificate verifies that f(x34,y34)=0f(x_{34},y_{34})=0 for all solutions to (1). Unfortunately the polynomial multipliers pjkp_{jk}, qkq_{k} involved are too large to reproduce here.

These computations complete the certification that there are no non-zero solutions to the system of equations (1). Thus there are no non-zero Hermitian matrices of rank ≤2\leq 2 in the linear space LΦ\mathcal{L}_{\Phi} and, by [4, Lemma 9], the frame Φ\Phi defines injective measurements.

The code for these computations in both Macaulay2 and Mathematica are available at http://www4.ncsu.edu/~clvinzan/smallFrame.html. ∎

Solving the system of equations (1) numerically, we see that, up to scaling, there are exactly twenty rank-2 matrices in the linear space LΦ\mathcal{L}_{\Phi}. These are in one-to-one correspondence with the twenty complex roots of the polynomial f(x34,1)f(x_{34},1). As none of these roots are real, none of the rank-2 matrices in LΦ\mathcal{L}_{\Phi} are Hermitian. For example, the solution (x34,y34)≈(1.95+2.08i,1)(x_{34},y_{34})\approx(1.95+2.08i,1) corresponds to the rank-2 matrix

and its conjugate, (x34,y34)≈(1.95−2.08i,1)(x_{34},y_{34})\approx(1.95-2.08i,1), gives

Because the certificates used in the proof of Theorem 1 are too large to give here, we now present a much smaller example of the computations involved.

Suppose we want to show that there is no rank-one Hermitian matrix of the form

III The set of injective frames

For example, we can replace the last vector of Φ\Phi,

Acknowledgements. Thanks to Bernhard Bodmann for his encouragement and interest in this problem. The author was supported by an NSF postdoc DMS-1204447.

References