🔢 mathematics

Local Consistency and Higher-Order Structure of Spherical Interpolation

This paper establishes the local consistency and higher-order accuracy of the SIDER- $n$ recursive spherical interpolation method by analyzing its behavior in geodesic normal coordinates, demonstrating that it achieves $O(h^{n+1})$ convergence through the systematic cancellation of curvature-dependent error terms.

Original authors: Shingyu Leung

Published 2026-06-12

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Shingyu Leung

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: Drawing on a Ball

Imagine you have a smooth, perfect basketball (this represents the unit sphere). You want to draw a smooth, curved line connecting a few specific dots painted on the surface of that ball.

In the flat world (like drawing on a piece of paper), we have standard math tools to connect dots smoothly. But on a ball, you can't just draw a straight line between two dots because that line would cut through the inside of the ball. You have to follow the curve of the surface.

This paper is about a specific, clever method called SIDER (Spherical Interpolation of orDER n) that connects dots on a ball. The authors wanted to prove that this method is not just "okay," but actually extremely accurate and smooth, even when you try to connect many dots at once.

The Problem: Why Flat Math Fails on a Ball

If you tried to use standard computer math (which assumes a flat world) to connect dots on a ball, the resulting line would float off the surface, go through the air, and then get "squashed" back onto the ball. This squashing ruins the smoothness and accuracy of the curve.

To fix this, the SIDER method uses a tool called SLERP (Spherical Linear Interpolation).

The Analogy: Imagine you are walking from one city to another on Earth. The shortest path isn't a straight line through the ground; it's a curved path along the surface (a "great circle"). SLERP is the math that tells you how to walk that perfect curved path between two points.

The Challenge: Building a Smooth Curve from Steps

The SIDER method builds a complex curve by stacking these simple curved steps on top of each other, like building a tower.

SIDER2: Connects 3 dots. It's like drawing a simple curve through three points.
SIDER3: Takes two of those simple curves and blends them together to connect 4 dots.
SIDER4: Takes two SIDER3 curves and blends them to connect 5 dots.
And so on...

The big question the authors asked was: "As we keep stacking these curves to connect more dots, does the result stay smooth and accurate, or does the tiny error from each step pile up and make the curve messy?"

The Secret Weapon: "Normal Coordinates"

To answer this, the authors used a special mathematical trick. Instead of looking at the whole ball, they zoomed in very close to a single point on the curve and flattened that tiny spot out like a map.

The Analogy: Imagine you are standing on a huge, round hill. If you look at the whole hill, it's curved. But if you look only at the patch of grass right under your feet, it looks perfectly flat.
By doing this, the authors could compare the SIDER method (which lives on the curved ball) to standard, flat math (which lives on the flat patch). This allowed them to see exactly where the "curvature" of the ball messes with the math.

The Discovery: How the Errors Cancel Out

The authors found that the SIDER method works because of a magical "cancellation" effect.

The First Step (SIDER2): When connecting 3 dots, the method is very accurate, but it has a tiny error. The authors proved this error looks exactly like the error you'd get if you were drawing on flat paper, just shifted slightly.
The Second Step (SIDER3): When you blend two SIDER2 curves together, you might think the errors would double. But they found that the errors in the two curves are identical twins. They have the exact same "shape" and "size."
- The Analogy: Imagine two people pushing a car. If one pushes forward and the other pushes backward with the exact same force, the car doesn't move. The errors cancel each other out perfectly.
- Because the errors cancel, the new SIDER3 curve is much smoother and more accurate than the parts it was made of.
The Higher Steps (SIDER4 and beyond): The authors proved that this "twin error" pattern continues. No matter how many times you stack the curves, the errors from the lower levels always match up perfectly so they cancel out when you blend them.

The Conclusion: It Works Every Time

The paper proves that for any number of dots you want to connect, the SIDER method will produce a curve that is incredibly close to the perfect mathematical path.

If you connect 3 dots, the error is tiny.
If you connect 4 dots, the error is even tinier.
If you connect $n+1$ dots, the error is so small it is practically invisible for smooth data.

In short: The authors used a "zoom-in" map trick to show that the SIDER method is a masterful way to draw smooth lines on a ball. It works because the tiny mistakes made in the early steps are perfectly designed to cancel each other out in the later steps, leaving you with a highly accurate curve.

A Note on the "AI" Aspect

The paper includes a unique disclosure at the end. The author states that they used an advanced AI (GPT-5.5) to do the heavy lifting of the complex mathematical derivations and calculations. The author acted as the "director," setting the goals and checking the logic, but acknowledges that the math is so deep that it is currently beyond what a human could verify manually. Therefore, this paper is being shared as a preprint (a draft) to share these AI-assisted findings, rather than as a final, peer-reviewed journal article.

Technical Summary: Local Consistency and Higher-Order Structure of Spherical Interpolation

Problem Statement
The paper addresses the challenge of interpolating data constrained to a nonlinear manifold, specifically the unit sphere $S^2$ . Standard Euclidean polynomial interpolation fails in this context because the resulting curve generally leaves the manifold, and projecting it back alters both the geometry and approximation properties. While Spherical Linear Interpolation (SLERP) provides a geometrically natural, first-order method that preserves the unit-length constraint, piecewise SLERP lacks the smoothness required for systematic high-order reconstruction. Previous work introduced SIDER (Spherical Interpolation of orDER n), a recursive family of interpolants inspired by Neville interpolation but replacing affine combinations with SLERP. However, a rigorous local consistency analysis proving the expected high-order accuracy of SIDER for smooth curves had not been established.

Methodology
The analysis is conducted using geodesic normal coordinates centered at the exact evaluation point $\gamma(s)$ on a smooth curve. This coordinate system allows the SIDER recursion to be compared directly with classical Euclidean Neville interpolation while explicitly retaining curvature-dependent corrections introduced by SLERP.

The core technical approach involves:

Taylor Expansions in Normal Coordinates: Deriving local expansions for the sampled data points $p_i = \gamma(ih)$ and the exact curve in the tangent space $T_{\gamma(s)}S^2$ .
Cubic SLERP Expansion: Establishing a precise cubic expansion for the SLERP operation in normal coordinates. A critical finding is that while SLERP is affine to leading order, its first nonlinear correction appears at the cubic order. This is significant because the leading error of a quadratic reconstruction (SIDER2) is also of order $h^3$ . Therefore, rigorous proofs for higher orders cannot treat the nonlinear part of SLERP as a negligible remainder; the specific structure of the cubic term must be analyzed.
Degree-Filtered Formal Expansion: Introducing a framework where the local error is treated as a formal series in the step size $h$ , with coefficients that are polynomials in the normalized stencil variable $\xi = \theta - i$ . The paper proves that the SLERP operation preserves a specific "degree filtration" property: if the inputs have error coefficients of degree at most $r$ in $\xi$ at order $h^r$ , the output maintains this structure.

Key Contributions and Results

SIDER2 (Third-Order Accuracy): The paper derives a refined error formula for SIDER2. It proves that despite the nonlinear geometry, the leading error term is exactly $O(h^3)$ and possesses the same shifted nodal structure $\theta(\theta-1)(\theta-2)$ as ordinary Euclidean quadratic interpolation. The cubic curvature corrections from the intermediate SLERP steps do not alter the leading error coefficient.
SIDER3 (Fourth-Order Accuracy): The analysis demonstrates that two adjacent SIDER2 reconstructions entering the SIDER3 recursion share a common leading cubic coefficient (the vector $B$ representing the third derivative of the curve in normal coordinates). Because the weights in the SIDER3 recursion match those of the Neville algorithm, these leading cubic terms cancel exactly. The final SLERP blend, applied to points already $O(h^3)$ -close to the solution, introduces nonlinear errors of order $O(h^9)$ , which are negligible. This yields a local accuracy of $O(h^4)$ .
SIDER4 (Fifth-Order Accuracy): By extending the expansion to the next order, the paper verifies that the leading error coefficients for adjacent SIDER3 reconstructions are also compatible (shift-invariant). This allows the cancellation of the $O(h^4)$ terms, proving fifth-order accuracy ( $O(h^5)$ ) for SIDER4.
General All-Order Consistency (Theorem 4): The paper introduces a degree-filtered expansion framework to generalize these results. It proves that for any fixed $n$ , the SIDER- $n$ recursion preserves the polynomial degree structure of the error coefficients. Combined with the interpolation conditions at $n+1$ nodes, this forces the first $n$ error coefficients to vanish identically. Consequently, the local consistency estimate is established as:
$d_{S^2}(\gamma(\theta h), P^{[n]}_i(\theta; h)) = O(h^{n+1})$
under assumptions of smoothness ( $C^{n+1}$ ) and a small-stencil condition (points remaining in a common geodesically convex neighborhood).

Significance and Claims
The paper claims to provide a rigorous local explanation for the observed high-order behavior of the SIDER construction. Its primary significance lies in resolving the tension between the algebraic cancellation mechanism of Neville interpolation and the geometric nonlinearity of SLERP.

The authors emphasize that the success of the SIDER recursion relies not merely on the algebraic cancellation of nodal factors (as in Euclidean space) but on a deeper coefficient compatibility: the leading error vectors of adjacent lower-order reconstructions must be identical when expressed in the same normal-coordinate system. The paper proves this compatibility explicitly for SIDER2 through SIDER4 and establishes it generally for all orders via the degree-filtered framework.

The results are strictly local and asymptotic, assuming the interpolation stencil and intermediate values remain within a geodesically convex neighborhood to ensure the uniqueness of geodesics and the validity of the logarithm map. The paper does not claim global convergence or stability for the associated SENO (Spherical Essentially Non-Oscillatory) method, noting that a full non-oscillation theory would require additional analysis of the stencil-selection mechanism.

Disclosure Note
The paper explicitly discloses that the primary computational modeling and mathematical derivations were performed using GPT-5.5 (Thinking). The author acts as the principal investigator defining the scope but acknowledges that the mathematical depth of the results exceeds current manual verification capabilities. Consequently, the findings are presented as AI-assisted hypotheses intended for dissemination as a preprint, subject to future formal peer verification.