Piecewise C² Functions and Their Applications

• Piecewise twice continuously differentiable functions are functions that are C² within each subinterval, with matching conditions ensuring C¹ continuity at the interfaces.
• They enable precise analytical treatments, as seen in applications like inverse spectral theory and nonlinear PDEs, using techniques such as Sturm–Liouville theory.
• Their construction supports advanced numerical optimization and modeling of layered physical media, handling discontinuities while preserving tractable analytic properties.

A piecewise twice continuously differentiable function is a map that is twice continuously differentiable ($C^2$) on each segment of a prescribed partition of its domain, with specified matching conditions at the boundaries between segments. Such functions form a critical regularity class in a variety of mathematical contexts, including inverse spectral theory, nonlinear partial differential equations, and optimization. Notably, piecewise $C^2$ functions can exhibit substantial mathematical subtlety at the interfaces, often modeling physically relevant discontinuities or phase changes while maintaining tractable analytic properties.

1. Formal Definition and Regularity Structure

Let $0=R_0 < R_1 < \cdots < R_L=1$ partition $[0,1]$. A function $n : [0,1] \to \mathbb{R}$ is piecewise twice continuously differentiable (piecewise $C^2$) with $L$ layers if

$$n(r) = \begin{cases} n_1(r), & R_0 \le r < R_1, \ n_2(r), & R_1 \le r < R_2, \ \vdots & \vdots \ n_L(r), & R_{L-1} \le r \le R_L, \end{cases}$$

where each $n_l \in C^2([R_{l-1}, R_l])$ for $l = 1, \dots, L$. The matching conditions at interfaces require

$$\lim_{r \to R_l^-} n_l(r) = \lim_{r \to R_l^+} n_{l+1}(r), \qquad \lim_{r \to R_l^-} n_l'(r) = \lim_{r \to R_l^+} n_{l+1}'(r),$$

for all $l=1,\dots,L-1$, ensuring global $C^1$ regularity. Typically, uniform bounds are imposed: $$0 < n_* \le n_l(r) \le n^*, \quad |n_l'(r)| \le M_1, \quad |n_l''(r)| \le M_2, \quad r \in [R_{l-1}, R_l].$$ The space of all such functions is denoted $C_p^2[0,1]$ (Bu et al., 16 Jan 2026).

2. Canonical Examples and Functional Properties

A core illustrative example is the two-layer piecewise constant function on $[0,1]$: $$n^{(1)}(r) = \begin{cases} 4, & 0 \le r < \tfrac12, \ 16, & \tfrac12 \le r \le 1, \end{cases} \qquad n^{(2)}(r) = \begin{cases} 16, & 0 \le r < \tfrac12, \ 4, & \tfrac12 \le r \le 1. \end{cases}$$ Both $n^{(1)}$ and $n^{(2)}$ are in $C_p^2$ (trivially, as they are piecewise constant) and satisfy all interface conditions (Bu et al., 16 Jan 2026).

Classically, the property of being piecewise $C^2$ permits the use of strong analytic machinery (e.g., Sturm–Liouville theory) on each subinterval, allows interface matching for higher-order ODE boundary problems, and ensures the existence and Lipschitz dependence of solutions on parameters layer by layer.

3. Spectral and Analytical Implications

A salient application of piecewise $C^2$ functions is in the inverse spectral theory of radially symmetric transmission problems. Considering the equation

$y'' + k^2 n(r) y = 0, \qquad y(0)=0,\quad y'(0)=1,$

one obtains fundamental solutions $g_{l1}, g_{l2} \in C^2$ on each segment $[R_{l-1}, R_l]$, with global $C^1$-continuous solutions generated by interface matching (Bu et al., 16 Jan 2026). The associated transmission eigenvalues (zeros of the characteristic determinant) depend critically on integrals and products involving $n_l$, often only through certain aggregate quantities (such as the total optical path length) rather than the specific ordering of layers.

However, non-uniqueness can arise: for certain configurations, distinct piecewise $C^2$ functions have identical sets of special transmission eigenvalues owing to invariance under layer permutations. In the explicit example above, $n^{(1)}$ and $n^{(2)}$ yield determinants

$$d^{(1)}(k) = 3\, d^{(2)}(k) = \frac{9}{16k} (\sin 2k + \sin 4k),$$

so $\{k : d^{(1)}(k) = 0\} = \{k : d^{(2)}(k) = 0\}$ (Bu et al., 16 Jan 2026). This demonstrates that the spectrum may not distinguish the layering order unless additional regularity (e.g., global $C^2$ without non-smooth interfaces) is imposed.

4. Construction of Piecewise $C^2$ Functions with Prescribed Properties

In higher dimensions and for more delicate PDE properties, explicit constructions of functions that are piecewise $C^2$ but not globally $C^2$ are feasible. Pan and Yan (Pan et al., 2022) provide a canonical recipe for $u : \mathbb{R}^n \to \mathbb{R}$ that is:

• $C^2$ on each region of a disjoint partition,
• twice differentiable everywhere (including interfaces),
• with continuous Laplacian $\Delta u$ and globally bounded Hessian $D^2u$,
• yet $u \notin C^2(\mathbb{R}^n)$ due to discontinuity of second derivatives at prescribed points.

Such constructions typically rely on $C^\infty$ cutoff functions $\eta$, suitable scaling of smooth profiles, and careful spatial arrangement (e.g., sequences of disjoint balls shrinking to a point). Matching conditions are arranged so that $u$, $\nabla u$, and $D^2u$ vanish at interfaces, but the global $C^2$ property fails at isolated points. These examples can be extended to higher-order differentiability and to cases involving the Monge–Ampère operator.

5. Role in Optimization and Piecewise–Smooth Penalty Functions

Piecewise (but globally $C^2$) constructions are crucial in optimization frameworks, particularly in nonlinear semidefinite programming. In this context, penalty and merit functions benefiting from full twice–continuous differentiability are highly desirable for the application of second-order methods (e.g., trust-region, SQP, augmented Lagrangian) (Yamakawa, 24 Sep 2025).

A paradigmatic example is the penalty function

$$F(x; v, M, \rho, \sigma, \tau) = \rho f(x) + \frac{\sigma \tau}{2} \big\| \tfrac{1}{\tau} v - g(x) \big\|^2 + \frac{\sigma \tau}{4} \operatorname{tr} \big( [\tfrac{1}{\tau} M - G(x)]_+^4 \big),$$

where $[\cdot]_+^4$ applies the scalar function $[r]_+^4 = r^4$ if $r \ge 0$, $0$ otherwise, to each eigenvalue of the symmetric matrix $W$. The matrix function is piecewise defined in the eigenbasis but is globally $C^2$, including at the junction where eigenvalues cross zero. The construction leverages classical results in the theory of spectral functions to ensure smoothness of first and second derivatives globally, facilitating rigorous second-order convergence analysis for methods satisfying AKKT2 and CAKKT2 criteria (Yamakawa, 24 Sep 2025).

6. Uniqueness and Non-Uniqueness Phenomena

Piecewise $C^2$ regularity marks a threshold in uniqueness results for certain inverse problems. For the inverse radial transmission eigenvalue problem, piecewise $C^2$ indices may be non-uniquely determined by available spectral data, admitting reorderings of material layers that are “spectrally invisible” (Bu et al., 16 Jan 2026). By contrast, additional regularity—specifically, ensuring that first derivatives are globally continuous on $[0,1]$ (no jumps)—restores uniqueness. Thus, the interface regularity is the critical mechanism dictating whether the spectral signature uniquely determines the underlying profile.

7. Generalizations and Applications

The framework of piecewise $C^2$ functions extends to higher order regularity conditions and fully nonlinear PDEs. For instance, the construction in (Pan et al., 2022) generalizes to build functions with $k+2$ derivatives, continuous Laplacian up to $C^k$, and bounded higher derivatives, yet failing global $C^{k+2}$ regularity. With additional decay assumptions, such constructions yield continuous Monge–Ampère determinant $\det D^2 u$ despite $u \notin C^2$.

Applications of piecewise $C^2$ functions span the modeling of layered media in physics, interface and singularity analysis in PDE theory, and the development of robust numerical optimization tools where differentiability of penalty terms is essential for algorithmic tractability and theoretical guarantees (Bu et al., 16 Jan 2026, Pan et al., 2022, Yamakawa, 24 Sep 2025).