Abstract
Discrete continuation method (DCM) is a widely used homotopy method to solve optimal control problems. Starting from an auxiliary problem, DCM monotonously varies a homotopy parameter until the objective problem is solved. However, DCM often fails when the initially and manually defined homotopy path encounters limit points or diverges to infinity. In this paper, a novel homotopy method, called the TFC homotopy method, is presented to enhance DCM performance through improved homotopy function construction. Firstly, the Theory of Functional Connections (TFC) is introduced and a TFC homotopy function is established, which implicitly defines infinite homotopy paths connecting the auxiliary problem with the objective problem for the first time in literature. Secondly, a novel two-layer path tracking method is designed. DCM is used in the first layer, and a second layer is triggered when unfavorable situations are met; then, another feasible homotopy path is found by solving an optimization problem and tracked by DCM. Compared with the pseudo-arclength method (PAM), the TFC homotopy method maintains the simplicity of DCM while providing flexible path switching. Multiple simulations against PAM illustrate the effectiveness of the proposed method.
Introduction: The Challenge of Nonlinear Zero-Finding
In aerospace engineering and robotics, solving complex nonlinear equations—finding the roots where $F(x) = 0$—is a fundamental task. While Newton’s method is a standard tool, its success is highly dependent on a good initial guess. Homotopy methods mitigate this by "deforming" a simple, solvable auxiliary problem into the target objective problem across a parameter $\kappa \in [0, 1]$.
However, traditional tracking strategies like the Discrete Continuation Method (DCM) are rigid. They often fail if the chosen path hits a "limit point" (where the path turns back) or an "indefinite growth" point (where the solution diverges to infinity). Even advanced techniques like the Pseudo-Arclength Method (PAM) can struggle with paths that loop back to the start. Current methods are essentially "locked in" to a single, manually defined path; if that path is mathematically unfavorable, the solver fails.
Methodology: TFC and Infinite Homotopy Paths
The proposed TFC-based Homotopy Method replaces the standard single-path approach with a flexible framework using the Theory of Functional Connections (TFC). TFC allows for the derivation of "constrained expressions"—functions that satisfy specific boundary conditions regardless of their internal auxiliary functions.
By using TFC to construct the homotopy function, the authors introduce a free parameter matrix ($\Omega$). Since $\Omega$ can be adjusted without violating the start ($\kappa=0$) and end ($\kappa=1$) conditions, the method implicitly defines an infinite family of paths. This provides a mathematical "steering wheel" to move from a failing path to a successful one without losing progress toward the final solution.
The Two-Layer Continuation Algorithm
The algorithm operates in two distinct layers to balance computational efficiency with robustness:
- Layer 1 (DCM): The algorithm begins with a simple, monotonous increase of the homotopy parameter. This is computationally inexpensive and easy to implement.
- Layer 2 (Path Switching): If Layer 1 fails—for instance, if the Jacobian becomes singular or values exceed a safety threshold—the second layer is triggered. It solves a small optimization problem to find a new $\Omega$, steering the solver onto a different, more stable homotopy path.
Experimental Validation
The method was evaluated against three numerical benchmarks designed to challenge traditional solvers:
- Algebraic Zero-Finding: In a 2D problem where standard paths lead to infinity, the TFC method detected the divergence, switched paths, and successfully converged to the root.
- Nonlinear Optimal Control: In a shooting method problem where PAM looped back to the origin due to a limit point, the TFC method switched to a "mild" path that led smoothly to the objective solution at $\kappa=1$.
- Elastic Rod Problem: For an inverse cantilever beam problem where paths often grow indefinitely, the TFC method used its second layer to find a new starting point and path, reaching the target configuration where PAM could not.
Conclusion
The TFC homotopy method offers a significant advancement by transforming homotopy from a tracking problem into a steering problem. By regularizing the Jacobian and providing a mechanism for path switching, it retains the simplicity of DCM while achieving the robustness necessary for complex engineering problems. Future work will explore the selection of optimal basis functions and strategies for handling bifurcation points.
Summary updated