Critical phase transitions in minimum-energy configurations for the exponential kernel family $e^{-|x-y|^q}$ on the unit interval

1. Problem Statement

The paper investigates the discrete energy minimization problem for $k$ ordered points $0 \le x_1 \le \dots \le x_k \le 1$ on the unit interval, governed by the pair potential kernel:
$$K_q(d) = e^{-d^q}, \quad q > 0$$
The discrete energy functional is defined as:
$$E_{k,q}(x_1, \dots, x_k) = \sum_{1 \le i < j \le k} e^{-|x_j - x_i|^q}$$

The central research question is to characterize the phase transitions in the optimal configurations (minimizers) as the parameter $q$ varies. Specifically, the study focuses on the transition from collision-free configurations (where all points are distinct) to endpoint-collapsed configurations (where multiple points coincide at the boundaries $0$ and $1$).

The parameter $q$ controls the local behavior of the kernel near the origin:

• $0 < q < 1$: The kernel has an infinite local slope (cusp-like).
• $q = 1$: The threshold case (standard exponential kernel $e^{-d}$).
• $q > 1$: The kernel is flat at the origin (Gaussian-type for $q=2$).

Unlike singular kernels (e.g., Riesz potentials $d^{-s}$) where collisions are energetically forbidden due to infinite energy, the bounded nature of $e^{-d^q}$ allows collisions. The paper seeks to determine when these collisions become energetically optimal.

2. Methodology

The author employs a combination of analytical proofs, asymptotic analysis, and numerical optimization:

• Gap Variable Reformulation: The problem is reformulated using gap variables $g_r = x_{r+1} - x_r$. The constraint becomes $\sum g_r = 1$ with $g_r \ge 0$. This transforms the problem into minimizing a function on a simplex.
• Convexity and Symmetry:
  • For $q \ge 1$, the energy function is proven to be convex on the simplex.
  • Symmetry arguments are used to reduce the search space, assuming minimizers are symmetric about the midpoint ($x \to 1-x$).
• Karush-Kuhn-Tucker (KKT) Conditions: Necessary conditions for optimality are derived to analyze the behavior of gaps, particularly when gaps approach zero.
• Asymptotic Analysis:
  • Small $q$ ($q \to 0^+$): The energy is expanded to show convergence to the Fekete/Chebyshev–Lobatto configuration.
  • Local Analysis near $d=0$: Comparing the scaling of energy changes ($\epsilon^q$ vs. $\epsilon$) to determine if a zero gap can be perturbed to lower the energy.
• Branch Comparison: For $q > 1$, the author compares the energy of "interior" branches (configurations with points strictly inside $(0,1)$) against "endpoint-only" branches (configurations where all points are at $0$ or $1$).
• Numerical Simulation: Gradient flow methods are used to visualize the dynamic evolution of point configurations and verify critical exponents numerically for $k$ up to 20.

3. Key Contributions and Results

A. The Collision-Free Regime ($0 < q < 1$)

• Result: For any $k \ge 2$ and $0 < q < 1$, all minimizers are strictly collision-free ($0 < x_1 < \dots < x_k < 1$).
• Proof Mechanism: The proof relies on the fact that for $q < 1$, the energy gain from opening a zero gap (scaling as $\epsilon^q$) dominates the energy loss from compressing other gaps (scaling as $\epsilon$) as $\epsilon \to 0$. Since $\epsilon^q \gg \epsilon$ for small $\epsilon$, any configuration with a collision can be improved by separating the points.

B. The Threshold Case ($q = 1$)

• Result: At $q=1$, collisions first become possible. The minimizers exhibit endpoint clustering.
• Law: Numerical evidence suggests a "one-third law" for the number of coincident points at each boundary:
  $$m(k) = \left\lfloor \frac{k+1}{3} \right\rfloor$$
  This implies that roughly one-third of the points cluster at $0$, one-third at $1$, and the remaining central third form an active block.

C. Critical Exponents for $q > 1$

The paper identifies critical values $q_k$ beyond which interior points are no longer optimal, and the global minimizer consists entirely of points at $\{0, 1\}$.

• Odd $k$ ($k = 2m+1$):

  • The critical exponent is universal (independent of $k$) and derived in closed form:
    $$q_{2m+1} = \frac{\log\left( \frac{1}{-\log((1+e^{-1})/2)} \right)}{\log 2} \approx 1.396363475$$
  • This value marks the exact crossing point between the energy of the symmetric "one-midpoint" branch and the "endpoint-only" branch.
  • Conjecture: This value is the true global critical exponent for all odd $k$.

• Even $k$ ($k = 2m$):

  • The critical exponent depends on $k$ and is determined numerically by comparing the symmetric "two-interior-point" branch with the endpoint-only branch.
  • The values increase monotonically with $k$ in the tested range ($4 \le k \le 20$):
    • $q_4 \approx 1.0627$
    • $q_6 \approx 1.1556$
    • ...
    • $q_{20} \approx 1.3107$
  • As $k \to \infty$, the even critical values appear to approach the odd universal value from below.

D. Asymptotic Regime ($q \to 0^+$)

• Result: As $q \to 0^+$, the minimizers do not converge to a uniform distribution. Instead, they converge to the Fekete configuration (affine Chebyshev–Lobatto points):
  $$\xi_j = \frac{1 - \cos(j\pi/(k-1))}{2}, \quad j=0, \dots, k-1$$
• Significance: This confirms that the small-$q$ limit favors configurations that maximize the product of pairwise distances (Vandermonde determinant), leading to higher point density near the boundaries.

4. Significance and Implications

• Phase Transition Characterization: The paper provides a rigorous and detailed phase diagram for a bounded kernel family, distinguishing it from the well-studied singular Riesz potentials. It establishes that bounded kernels can undergo a sharp transition from collision-free to collapsed states.
• Universality in Odd Dimensions: The discovery of a universal critical exponent for all odd $k$ is a significant theoretical finding, suggesting a deep structural symmetry in the energy landscape that is independent of the number of points.
• Methodological Insight: The use of gap variables and the comparison of specific symmetric branches provides a robust framework for analyzing similar energy minimization problems on intervals.
• Practical Application: The results are relevant for optimal experimental design, sensor placement, and numerical integration where one must balance point separation against boundary effects under different interaction potentials.

5. Open Questions

The author identifies several avenues for future research:

1. Global Optimality for Odd $k$: Proving that the universal value $q_{odd}$ is indeed the global transition point (excluding competing branches with multiple interior points).
2. Monotonicity of Even $k$: Rigorously proving that the sequence of critical exponents $q_{2m}$ is strictly increasing.
3. Large-$k$ Asymptotics: Characterizing the full phase diagram for $k \to \infty$, specifically whether the parity effect (odd vs. even) persists asymptotically and how the "active interior block" dimension scales.

The paper concludes with a visual phase diagram in the $(k, q)$-plane, summarizing the collision-free regime, the threshold line, and the critical boundaries for odd and even $k$.