Boundary Behavior of Optimal Characteristic Roots

Hello! Welcome back to our course on understanding the paper "Extremal problems for moments of polynomials and entire functions".

In our previous lesson, we introduced the Maximum Modulus Principle (MMP) and established its main consequence for functions on compact domains: a non-constant holomorphic function must attain its maximum modulus on the boundary of its domain, not in the interior.

Today, we will directly apply this powerful principle to the paper's core optimization problem. Our goal is to use the MMP to justify a crucial simplification: to maximize the amplitude |x_t|, we only need to consider characteristic roots z_k that lie on the boundary of the allowed disc, i.e., where |z_k| = r.

1. The Amplitude x_t as a Function of its Roots

Let's start by recalling the setup from our first lesson. The solution to a homogeneous linear difference equation of order n is given by:
$x_t={\sum }_{k=1}^n{c}_k{z}_k^t$
where z_1, ..., z_n are the characteristic roots, and the coefficients c_k are determined by the initial conditions x_0, ..., x_{n-1}. For now, let's assume the roots are distinct.

The problem is to maximize |x_t| over all possible choices of roots z_k within the disk D_r = \{z \in \mathbb{C} : |z| \le r\} and all initial conditions x_j in the unit disc.

To apply the tools of complex analysis, we need to see how x_t behaves as a function of its roots. For a fixed set of initial conditions, x_t is a function of the n characteristic roots: x_t = F(z_1, z_2, ..., z_n).

A key property of this function F is that it is analytic with respect to each of its variables z_k. While a full proof is involved, the intuition is that the coefficients c_k are solved via a linear system (involving a Vandermonde matrix) whose entries are powers of the roots. The solutions for c_k are rational functions of the roots. This makes x_t a sum of analytic functions. Even when roots coincide, the resulting singularities in this formulation are removable, and the function F remains analytic.

2. Applying the Maximum Modulus Principle, One Root at a Time

Now we have the necessary ingredients:

1. We want to maximize |F(z_1, ..., z_n)|.
2. The domain for each root z_k is the compact set D_r.
3. The function F is analytic in each variable z_k.

Let's use an iterative argument to show that any optimal configuration of roots must have them all on the boundary circle |z| = r.

Suppose we have found a set of roots (z_1^*, ..., z_n^*) that maximizes |x_t|. Now, let's assume for the sake of contradiction that at least one of these roots, say z_j^*, lies in the interior of the disk: |z_j^*| < r.

Let's "freeze" all the other roots at their optimal values z_k^* (for k \neq j) and consider x_t as a function of only the j-th root, which we'll call z.
$g(z)=F(z_1^{\ast },...,z_{j-1}^{\ast },z,z_{j+1}^{\ast },...,z_n^{\ast })$

This function g(z) of a single complex variable has the following properties:

• It is defined on the closed disk D_r = \{z : |z| \le r\}.
• It is analytic on this domain.

According to the Maximum Modulus Principle we studied in the last lesson, the maximum value of |g(z)| on the compact set D_r must be attained on its boundary, \partial D_r = \{z : |z| = r\}.

Our point z_j^* is in the interior of this disk. This leads to two possibilities:

1. g(z) is not a constant function. In this case, the MMP states that |g(z)| cannot have a maximum at the interior point z_j^*. There must be some point z_j^{**} on the boundary (|z_j^{**}| = r) where |g(z_j^{**})| > |g(z_j^*)|. But this is a contradiction! It would mean our original set of roots (z_1^*, ..., z_n^*) did not actually maximize |x_t|, because we could get a larger value by replacing z_j^* with z_j^{**}.
2. g(z) is a constant function. If g(z) is constant across the entire disk D_r, then |g(z_j^*)| is equal to the value everywhere, including on the boundary. In this case, we can simply replace the interior root z_j^* with any root z_j^{**} on the boundary |z|=r without changing the value of |x_t|.

In either case, we can conclude that for any set of maximizing roots, if one root is in the interior, it can be moved to the boundary without decreasing (and possibly increasing) the value of |x_t|.

We can repeat this argument for every root z_1, ..., z_n. Therefore, we can state that an optimal configuration of roots that maximizes |x_t| can always be found on the boundary of the allowed domain.

This is a profound simplification. We have reduced the search space for each root from a 2D disk to a 1D circle.

3. A Note on the Full Problem

The paper's problem involves maximizing over both the roots z_k and the initial conditions x_j. The function x_t is linear in the initial conditions.
$x_t={\sum }_{j=0}^{n-1}{A}_j(z_1,...,z_n)\cdot x_j$
For any fixed set of roots, maximizing |x_t| over the x_j in the unit disc (|x_j| \le 1) is a standard problem. The maximum is achieved by setting |x_j|=1 and aligning their phases with the phases of the coefficients A_j.

Our focus here, however, is on the more complex, non-linear dependence on the roots z_k. The argument we just constructed using the MMP is the first and most critical step in taming that complexity. It tells us that the search for the optimal roots can be constrained to the n-dimensional torus \{ (z_1, ..., z_n) \in \mathbb{C}^n : |z_k|=r \text{ for all } k \}.

For a more rigorous treatment, one would typically use the Hartogs' extension theorem or the maximum principle for several complex variables, which generalizes this idea to polydiscs. However, the iterative one-variable-at-a-time argument we've used is intuitive and captures the essential logic.

To solidify your understanding, let's look at a formal statement of the MMP that is most relevant to our argument.

Conclusion

In this lesson, we bridged the gap between the abstract theory of the Maximum Modulus Principle and the concrete optimization problem in the paper.

Key Takeaways:

• The solution x_t of the difference equation can be viewed as an analytic function of each of the characteristic roots z_k.
• By applying the Maximum Modulus Principle iteratively to each root z_k on the compact disk |z| \le r, we can conclude that a maximum value of |x_t| must be attained when all roots are on the boundary.
• This reduces the search space for the optimal roots from n disks to n circles, a significant simplification of the problem.

Preview of the Next Lesson:
Now that we know the optimal roots must lie on the circle |z|=r, the problem is still not fully solved. We need to find the specific arrangement of roots on that circle that yields the maximum amplitude. The next step in the paper's argument is to reframe this problem in a new light: as a problem of polynomial interpolation. We will explore this reformulation in our next lesson.