Conformal Corrector: Numeric Framework

• Conformal corrector is a numerical framework that enforces near-conformal deformations by integrating Cauchy–Riemann constraints to preserve angles.
• It applies penalty-based regularization in shape optimization, balancing objective fidelity and mesh quality through a variational formulation.
• It employs iterative conformal mapping algorithms with FFT-based spectral computations to achieve rapid convergence and high accuracy in complex geometries.

A conformal corrector is a numerical algorithmic framework designed to compute or enforce near-conformal deformations of geometric domains, either as part of mesh-based shape optimization or for the construction of conformal mappings in complex geometry and fluid mechanics applications. The concept integrates analytic properties—primarily the Cauchy–Riemann (CR) equations encoding conformality—into optimization and numerical mapping procedures, providing angle-preserving transformations with controlled deviations. Two principal formulations have appeared in the literature: (1) as a penalty-regularized corrector in shape optimization via least-squares Cauchy–Riemann constraints (Iglesias et al., 2017), and (2) as an iterative mapping corrector for conformal flattening of domains over nontrivial topographies (Flamarion et al., 2021).

1. Analytical Foundation: Cauchy–Riemann Structure

The conformal corrector framework is anchored in the Cauchy–Riemann (CR) equations. For a deformation field $u = (u_1, u_2)$ on a planar domain $\Omega \subset \mathbb{R}^2$, conformality (i.e., angle preservation; holomorphicity) is characterized by: $$\partial_x u_1 = \partial_y u_2, \quad \partial_y u_1 = -\partial_x u_2$$ This can be written in operator form as a two-component mapping $C u$: $$C u := \begin{bmatrix} -\partial_x u_1 + \partial_y u_2 \ \partial_y u_1 + \partial_x u_2 \end{bmatrix}$$ Satisfaction of $C u = 0$ signals that $u$ is a holomorphic map. Deviation from conformality is penalized using the least-squares functional: $$J_{CR}[u] = \int_{\Omega} |C u|^2 \, dx = \int_\Omega \left[(\partial_x u_1 - \partial_y u_2)^2 + (\partial_y u_1 + \partial_x u_2)^2\right] \, dx$$ In iterative conformal mapping (e.g., for water wave topographies), an analogous structure arises: the mapping $f(\zeta) = x(\xi, \eta) + i y(\xi, \eta)$ is constructed with $y$ harmonic and $x$ as its harmonic conjugate, ensuring $x_\xi = y_\eta$ and $x_\eta = -y_\xi$ (Flamarion et al., 2021).

2. Penalty-Based Conformal Regularization in Shape Optimization

In shape optimization, the conformal corrector is integrated as a penalty regularization to target mesh deformations that are as conformal as possible while optimizing an objective $S[u]$ (e.g., drag, compliance) (Iglesias et al., 2017). The composite variational problem becomes: $$\text{Minimize } J_{total}[u] = S[u] + \alpha J_{CR}[u]$$ Here, $\alpha>0$ serves as a regularization parameter balancing objective fidelity and angle preservation.

The Euler–Lagrange condition reads: $$DS[u](v) + \alpha \int_\Omega C u \cdot C v\, dx = 0 \quad \forall v$$ or, equivalently, introducing the $L^2$-adjoint $C^*$,

$DS[u] + \alpha C^* C u = 0$

The finite-element discretization employs a space $V_h^2 \subset [H^1(\Omega)]^2$ (e.g., continuous $P_1$ vector fields), with assembly of corresponding stiffness-like matrices for both the penalty (CR) and standard $H^1$ terms. The system solved at each optimization iteration is: $(\alpha A_{CR} + A_H) U = f$

Boundary conditions are handled either as Dirichlet (clamped, $u_h=0$) or natural (Neumann) as implied by the variational form.

3. Iterative Conformal Mapping Corrector in Water Wave Domains

The iterative conformal-corrector algorithm for mapping domains with uneven topography (e.g., for free-surface water waves) seeks to map a canonical strip $S$ onto a physical fluid domain $\Omega$ where the bottom is given by a periodic function $h(x)$ (Flamarion et al., 2021).

The procedure introduces:

• An initial guess $H^1(\xi) = h(\xi)$
• A coupled system for $(X_b, H)$, where $X_b(\xi)$ represents the mapped bottom. The update scheme:
  • Compute $$X_b^\ell(\xi) = \xi + \sum_{k\neq 0} \frac{i \coth(k D) \cosh(k D)}{\cosh(k D)} \widehat{H}^\ell(k) e^{ik \xi}$$
  • Set $H^{\ell+1}(\xi) = h(X_b^\ell(\xi))$
• Iteration proceeds until $$\|H^{\ell+1} - H^\ell\|_\infty / \|H^\ell\|_\infty < \varepsilon$$ (with $\varepsilon$ chosen as $10^{-16}$)

Spectral (FFT-based) computation is central, leading to $O(N\log N)$ complexity per iteration, with rapid convergence (5–8 steps typical).

Once converged, the conformal map and its Jacobian $J(\xi) = [X_\xi(\xi)]^2$ are evaluated. The inverse map $(\xi, \eta) = f^{-1}(x_0, y_0)$ is recovered by minimizing the squared distance between $(x(\xi, \eta), y(\xi, \eta))$ and $(x_0, y_0)$.

4. Optimization Algorithms and Workflow Integration

In mesh-based shape optimization, the conformal-corrector is executed within the main loop as follows (Iglesias et al., 2017):

• At each iteration, solve the state PDE and its adjoint to obtain $DS[u_k]$.
• Assemble right-hand side and stiffness matrices incorporating the CR-term.
• Solve the regularized linear system to obtain the deformation direction.
• Perform a line search for step size.
• Update mesh nodes and monitor convergence via deformation and objective change.
• Typically, L-BFGS or quasi-Newton variants are used, with the CR-penalty modifying the inner product structure.

For FFT-based conformal mapping correctors, the workflow is characterized by alternating iterations of spectral updates and nonlinear (pointwise) evaluation, terminating when the change in the topography representation falls below a given tolerance (Flamarion et al., 2021).

5. Mesh Quality Control and Parameter Selection

Proper mesh quality is critical. A specific metric $\eta(K) = h(K)/(2 \rho(K))$ is used per triangle $K$, where $h(K)$ is the diameter and $\rho(K)$ is the inradius. $\eta(K)=1$ for equilateral triangles and grows for elongated elements (Iglesias et al., 2017). This metric is tracked through min/max and histogram statistics at each optimization step.

Parameter $\alpha$ in the regularized problem controls the regime: small $\alpha$ prioritizes conformality but slows convergence and may lead to ill-conditioned systems; large $\alpha$ reverts to standard $H^1$-type behavior, sometimes inducing mesh distortion. Monitoring the conditioning of $(\alpha A_{CR} + A_H)$ and mesh metrics is essential.

In the iterative mapping setting, monitoring $\min |X_\xi|$ guards against "pinching," ensuring the mapping remains invertible.

6. Empirical Performance and Implementation Aspects

In shape optimization examples such as circle-to-clover mappings, the conformal-corrector with $\alpha=10^{-2}$ achieved near-perfect angle preservation; the histogram of $\eta(K)$ before and after optimization was nearly unchanged (Iglesias et al., 2017). By contrast, pure $H^1$-inner-product evolution led to mesh collapse within a few steps.

For the water wave mapping context, accuracy against the Schwarz–Christoffel toolbox revealed maximum absolute Jacobian errors ranging (for topography amplitude $A=0.01$ to $0.5$) from $1.7 \times 10^{-7}$ to $1.19 \times 10^{-5}$, with convergence in 5–8 iterations. Spectral grid refinement yielded $\ell_2$-error dropping to $O(10^{-15})$, and the inverse mapping exhibited errors $O(10^{-8})$ for the most severe tested cases (Flamarion et al., 2021). Computational cost was $O(N\log N)$, dominating in FFT transforms and negligible against alternative methods.

Assembly of the CR stiffness term is identical in cost to a standard element matrix (for mesh-based problems). For implementations in high-level PDE environments (Firedrake/FFC, Fenics/Pyadjoint), the CR-operator can be defined symbolically for automatic assembly. In RKHS-based settings, discretization leads to dense matrix systems but similarly leverages pointwise evaluation of the CR-form.

The iterative conformal-corrector algorithm for mappings can be implemented in MATLAB, Julia, or Python+NumPy/FFTW in under 200 lines of code (Flamarion et al., 2021).

7. Applications and Extensions

The conformal-corrector concept is applied in:

• Mesh-based two-dimensional shape optimization where mesh quality demands nearly conformal updates even under severe geometric transformations (e.g., Stokes drag minimization, shape matching in parametric PDEs) (Iglesias et al., 2017).
• Construction of conformal maps (including inverse mapping) for complex fluid domains, particularly free-surface hydrodynamics over variable topography; provides computational acceleration relative to Schwarz–Christoffel methods and supports stable mapping inversion (Flamarion et al., 2021).

A plausible implication is that the conformal-corrector framework can be adapted for problems wherever analytic conformality must be numerically enforced or leveraged, including in computer graphics, image registration, and physical PDE-driven design, subject to additional constraints or extensions in higher dimensions.

---

References:

• "Shape optimisation with nearly conformal transformations" (Iglesias et al., 2017)
• "An iterative method to compute conformal mappings and their inverses in the context of water waves over topographies" (Flamarion et al., 2021)