Approximate solutions to a nonlinear functional differential equation

1. Problem Statement

The paper investigates a specific nonlinear functional differential equation (FDE) known as the $\alpha$-Riccati equation:
$$u'(t) + u(t) = u^2(\alpha t), \quad t > 0$$
with the initial condition $u(0) = 1$.

• Context: This equation is a nonlinear variant of the linear pantograph equation $u'(t) + au(t) = bu(\alpha t)$. The parameter $\alpha > 1$ introduces a "time-advanced" argument (the derivative at time $t$ depends on the future state $\alpha t$). While such equations often arise in modeling (e.g., nerve axons, control theory, Navier-Stokes scaling), the nonlinear case lacks a complete analytical theory regarding existence and uniqueness.
• Specific Focus: The authors focus on the regime $\alpha > 1$ and solutions belonging to Class (A), defined as solutions where $u(t) \to 1$ as $t \to \infty$ (specifically $u(t) = 1 + O(e^{-t})$).
• The Challenge: The problem is nonlinear, and for certain values of $\alpha$, multiple solutions may exist. Standard numerical methods struggle to find these solutions without good initial guesses, and the behavior near $\alpha = 1$ (where $\alpha \to 1^+$) is particularly complex.

2. Methodology

The authors employ a hybrid approach combining perturbation analysis, spectral collocation, and numerical continuation.

A. Linearization and Characteristic Values

The authors first analyze the linearized version of the equation around the constant solution $u=1$. By setting $u(t) = 1 + v(t)$, the linearized equation becomes:
$$v'(t) + v(t) = 2v(\alpha t)$$
They identify characteristic values of $\alpha$, denoted as $\alpha_n = 2^{1/n}$ for $n=1, 2, \dots$. At these specific values, the homogeneous linear equation admits non-trivial solutions (characteristic functions) that satisfy $v(0)=0$. These functions are given by a convergent series involving Laguerre-like terms.

B. Perturbation Analysis (Generating Initial Guesses)

To solve the nonlinear problem near the characteristic values $\alpha_n$, the authors perform a perturbation analysis. They assume $\alpha = \alpha_n + \epsilon$ (where $|\epsilon| \ll 1$) and expand the solution $v(t)$ in powers of $\epsilon$.

• Leading Order: The leading term is proportional to the characteristic function of the linear problem: $v(t) \approx C_n \epsilon E(t; \alpha_n)$.
• Determining Coefficients: The scaling constants $C_n$ are determined by matching the moments of the linear and nonlinear solutions. The authors derive a recursive formula to compute $C_n$ (values provided in Table 1).
• Utility: These perturbation approximations serve as highly accurate initial guesses for the numerical solver, which is crucial for converging to specific branches of solutions in a nonlinear system.

C. Numerical Solution: Laguerre Spectral Collocation

For the actual computation, the authors use a Laguerre spectral collocation method on a semi-infinite domain $[0, \infty)$.

• Discretization: They approximate the solution using Laguerre polynomials. To improve efficiency and stability, they employ a truncated Laguerre method, using only the first $\lceil M\sqrt{N} \rceil$ points of an $N$-point grid (with $M=6$ found to be optimal).
• Functional Terms: The delayed/advanced terms $v(\alpha t)$ and $v^2(\alpha t)$ are handled using a barycentric resampling approach, constructing an interpolation matrix $P_\alpha$.
• Solver: The resulting nonlinear algebraic system is solved using Newton iteration, seeded by the perturbation approximations.

D. Continuation

To map out the solution space over a range of $\alpha$ values, the authors use pseudo-arclength continuation. This allows them to trace solution branches even through bifurcation points where standard Newton methods might fail or jump to different branches.

3. Key Contributions and Results

A. Discovery of Multiple Solution Branches

The study reveals a rich bifurcation structure for $\alpha > 1$.

• Bifurcation Points: New non-constant solutions emerge from the trivial constant solution $u=1$ at the characteristic values $\alpha_n = 2^{1/n}$.
• Solution Classification: Solutions are categorized based on their first turning point:
  • Plus-solutions: First turning point is a local maximum.
  • Minus-solutions: First turning point is a local minimum.
• Branching Behavior:
  • For $\alpha$ slightly less than 2 ($n > 1$), a unique "minus-solution" exists.
  • As $\alpha$ decreases further (approaching $\sqrt{2}$, i.e., $n=2$), a transcritical bifurcation occurs. Two new "plus-solutions" emerge.
  • One of these plus-solutions eventually collides with the constant solution at $\alpha = \sqrt{2}$ and re-emerges as a minus-solution for $\alpha < \sqrt{2}$.
  • This pattern repeats for higher $n$: in intervals $(n-\delta, n)$, two new solutions are generated.

B. Quantitative Characterization

• The authors provide a catalog of the number of solutions for various $\alpha$ ranges.
• They compute the $L^2$ norms of these solutions, showing that as $\alpha \to 1^+$ (which corresponds to $n \to \infty$), the number of solutions increases, and their norms grow.
• Table 1 provides the first six scaling coefficients ($C_1$ to $C_6$) and the norms of the characteristic functions, which are essential for the perturbation approximations.

C. Numerical Validation

• The perturbation approximations (Eq. 3.7) are shown to be excellent qualitative and quantitative matches to the full numerical solutions near the bifurcation points.
• The residual analysis (Fig. 2) confirms that the perturbation error is $O(\epsilon^2)$ and bounded, validating the asymptotic expansion.

4. Significance and Limitations

Significance

• Methodological Novelty: The paper demonstrates a successful combination of spectral collocation, perturbation theory, and continuation methods for nonlinear FDEs. This framework is presented as a new and reliable tool for exploring complex functional differential equations where analytical proofs of existence/uniqueness are difficult.
• Insight into Nonlinear Dynamics: It provides the first detailed map of the solution space for the $\alpha$-Riccati equation, revealing complex bifurcation structures (transcritical bifurcations) that were previously unknown.
• Generalizability: The authors suggest this methodology can be extended to more general equations of the form $u' + au = b u^m(\alpha t)$.

Limitations and Open Problems

• The $\alpha \to 1$ Limit: As $\alpha$ approaches 1, the perturbation approximations become less effective (due to rapidly growing constants $C_n$), and the solutions become highly oscillatory near $t=0$. The spectral method struggles with these singular behaviors.
• Singular Limit: The authors observe a "puzzling" phenomenon where an infinite number of solutions (with increasing norms) exist for $\alpha > 1$, yet they all collapse to the unique solution $u=1$ exactly at $\alpha=1$. The nature of this singular limit remains an open problem.
• Computational Cost: For large $n$ (small $\alpha$), the Jacobian matrices in the continuation method become near-singular, making numerical stability difficult.

Conclusion

The paper successfully bridges the gap between linear theory and nonlinear complexity for a specific class of functional differential equations. By leveraging the known series solutions of the linear problem to seed numerical solvers, the authors map out a complex landscape of multiple solutions, identifying specific bifurcation mechanisms and providing a robust computational framework for future investigations into time-advanced nonlinear systems.