Abstract
In a physical design problem, the designer chooses values of some physical parameters, within limits, to optimize the resulting field. We focus on the specific case in which each physical design parameter is the ratio of two field variables. This form occurs for photonic design with real scalar fields, diffusion-type systems, and others. We show that such problems can be reduced to a convex optimization problem, and therefore efficiently solved globally, given the sign of an optimal field at every point. This observation suggests a heuristic, in which the signs of the field are iteratively updated. This heuristic appears to have good practical performance on diffusion-type problems (including thermal design and resistive circuit design) and some control problems, while exhibiting moderate performance on photonic design problems. We also show in many practical cases there exist globally optimal designs whose design parameters are maximized or minimized at each point in the domain, i.e., that there is a discrete globally optimal structure.
Introduction
Physical design is a cornerstone of modern engineering, encompassing the optimization of material properties—such as permittivity in silicon photonics or thermal conductance in heat management systems—to achieve a desired physical field behavior. Traditionally, these problems are framed as nonconvex optimization tasks. The governing physics equations often involve the product of design parameters and field variables, leading to mathematical formulations that are notoriously difficult to solve globally and are frequently NP-hard.
Engineers typically rely on local optimization heuristics or continuous relaxations that may not yield physically realizable or globally optimal structures. This paper introduces a novel approach using convex restrictions, demonstrating that the inherent difficulty of these problems is often tied to the discrete choice of the field's sign at each point in space.
Methodology
Problem Formulation
The authors consider physical systems where the relationship between a design parameter $\theta$ and the resulting field $v$ can be expressed through a bilinear constraint, such as $u = \text{diag}(\theta)v$. The core challenge is that both $\theta$ and $v$ are variables, making the constraint nonconvex.
To address this, the researchers reformulate the physics into an "absolute upper bound" form: $|w| \le |v|$. They prove that if the sign vector $s$ of the field $v$ is known or fixed, the problem transforms into a convex optimization problem. Convex problems are significant because they can be solved efficiently to global optimality using standard numerical methods.
Sign Flip Descent Heuristic
Since the optimal signs are rarely known a priori, the authors propose a "sign flip descent" heuristic. The algorithm follows an iterative process:
- Initialize: Start with an initial guess for the sign vector $s$.
- Solve: Solve the resulting convex restricted problem to find the optimal field and design for those signs.
- Update: Use a "field-based rule" to identify points where the field is near zero or inconsistent with the current signs, and "flip" those signs for the next iteration.
This method provides a valid upper bound on the global optimum at every step and avoids the exponential complexity of checking all $2^n$ possible sign combinations.
Results and Applications
Photonic Design
The method was tested on optimizing a 2D photonic domain governed by the Helmholtz equation. The objective was to minimize field intensity within a specific "mask" region while maintaining an excitation source.
As shown in Figure 3, the algorithm successfully generates a structured design that routes the field away from the masked area. Figure 4 illustrates the resulting field distribution, where the intensity is effectively suppressed in the target region. For a $101 \times 101$ grid, the heuristic reached a solution in approximately four minutes.
Thermal Design and Control
The heuristic showed exceptional performance in diffusion-type problems. In thermal management simulations, it recovered complex "tendril-like" structures optimized for heat dissipation in just 20 seconds. In a dynamic temperature control scenario for a "smart building," the algorithm converged in under two seconds, effectively managing time-varying heat pump and vent operations to maintain comfort levels.
Discussion and Conclusion
A key theoretical insight of this work is the extremality principle: for many linear objectives, the globally optimal design is "extremal," meaning parameters are set to their minimum or maximum allowed values. This explains why optimized engineering structures often appear binary or "bang-bang" in nature.
While the current framework is highly effective for real scalar fields, future work remains to extend these convex restrictions to complex-valued fields (common in RF and phase-sensitive photonics) and to handle multi-scenario designs where a single structure must perform across varying conditions. Nonetheless, the sign-based convex restriction offers a powerful new tool for the global optimization of physical systems.
Summary updated