🔬 optics

Sensitivity of Fitting High Order Zernikes

The paper demonstrates that fitting high-order Zernike coefficients to wavefront data becomes increasingly unreliable and ultimately useless unless the exact position of the wavefront edge is known with high precision, as small uncertainties in this boundary significantly distort the resulting coefficients.

Original authors: Gregg M. Gallatin

Published 2026-06-25

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Gregg M. Gallatin

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: The "Fuzzy Edge" Problem

Imagine you are trying to describe the shape of a ripple in a pond using a set of standard building blocks called Zernikes. These blocks are like musical notes or mathematical Lego bricks that, when stacked together, can perfectly recreate any wave pattern on a circular surface (like a lens or a mirror).

Usually, scientists only use the first few "notes" (low-order Zernikes) because they describe the big, obvious bumps and dips in the wave. However, some people have started trying to use the "high notes" (high-order Zernikes) to describe incredibly tiny, complex details in the wave.

The Paper's Main Warning:
The author, Gregg Gallatin, argues that trying to use these "high notes" is a waste of time unless you know the exact size of the pond's edge with perfect precision. If you are even slightly off about where the edge of the data is, your description of the tiny details becomes completely wrong.

The Analogy: The Stretchy Rubber Sheet

To understand why, imagine the wavefront data is drawn on a rubber sheet.

The Normalization: To fit the "Lego blocks" (Zernikes) to the sheet, you have to stretch or shrink the rubber sheet so it fits exactly inside a perfect circle (the "unit circle").
The Edge Uncertainty: In the real world, you don't know exactly where the edge of the rubber sheet is. Maybe it's at pixel 100, or maybe it's at pixel 101. Let's say you guess it's at 100, but it's actually at 101. That's a tiny error (about 1%).
The Low-Order Blocks: If you are only describing a big, gentle hill (low-order Zernike), stretching the rubber sheet by 1% doesn't change the shape of the hill very much. You can still describe it accurately.
The High-Order Blocks: Now, imagine trying to describe a super-sharp, jagged spike right at the edge of the rubber sheet (a high-order Zernike). These shapes change very rapidly near the edge.
- If you stretch the rubber sheet by just 1%, that sharp spike gets squashed or stretched wildly.
- Because the "spike" is so sensitive to the edge, your 1% guess about the edge size causes a huge error in your description of the spike.

The Paper's Finding:
As you try to describe more and more complex, jagged details (higher orders), the math becomes incredibly sensitive to that tiny 1% error in the edge size. The results become "garbage in, garbage out." You might calculate a value for a high-order coefficient, but it will be so far from the truth that it is meaningless.

The "Cross-Talk" Problem

The paper also mentions that Zernikes are supposed to be orthogonal. In simple terms, this means they are independent. If you turn up the volume on the "bass" note (one Zernike), it shouldn't accidentally change the volume of the "treble" note (another Zernike).

However, if you get the edge size wrong:

The independence breaks.
The "bass" note starts bleeding into the "treble" note.
The math gets confused, and you end up with "ghost" values where there shouldn't be any.

What Should We Do Instead?

The paper suggests that instead of trying to force-fit thousands of tiny, high-order Zernike blocks (which fails because of the edge uncertainty), we should do this:

Fit the wavefront with a few standard, low-order blocks (the big hills and valleys).
Subtract those from the data.
Look at what is left over (the "residual").
Analyze that leftover noise using a different tool called a Fourier Transform.

This method is better because it doesn't rely on knowing the exact edge of the circle to the same degree of precision to get useful results. It tells us about the "halo" or "flare" of light that scatters around the main image, which is very useful for optical engineers.

The Bottom Line

How precise do you need to be?
The paper does some math to show that if you want to fit a Zernike up to the 40th order (very high detail) and be within a few percent of the truth, you need to know the edge of your data to within 0.1%.

If your detector (like a camera sensor) is 1,000 pixels wide, that means you need to know the edge of the light circle to within 1 pixel. If you are off by even a fraction of a pixel, your high-order calculations are useless.

In short: Trying to fit high-order Zernikes is like trying to tune a radio to a specific station while standing in a storm. If the signal (the edge of the data) isn't perfectly clear, the higher the frequency (order) you try to tune to, the more static (error) you hear, until the sound is unintelligible.

Technical Summary: Sensitivity of Fitting High Order Zernikes

Problem Statement
The paper addresses a critical limitation in the representation of optical wavefronts using Zernike polynomials. While Zernikes are a standard orthogonal basis for circular wavefronts, there is a growing trend to fit wavefront data to increasingly higher-order terms (e.g., beyond the 10th or 12th order). The author identifies a fundamental sensitivity issue: fitting these high-order polynomials requires normalizing the wavefront radius to a unit circle. Any uncertainty or error in determining the exact physical edge (radius, $a$ ) of the wavefront data introduces significant errors into the fitted coefficients. The paper posits that as the order of the Zernike polynomials increases, the sensitivity to this radius error grows rapidly, potentially rendering high-order fits meaningless unless the edge position is known with extreme precision.

Methodology
The author employs an analytical derivation to quantify this sensitivity. The core of the methodology involves:

Formulation: Defining the wavefront $w(\rho, \phi)$ as a linear superposition of Zernike functions and establishing the relationship between the true wavefront radius ( $a_w$ ) and the radius used for fitting ( $a_{fit}$ ).
Orthogonality Breakdown: Utilizing the orthogonality condition of Zernikes, the author derives an expression for the fitted coefficients ( $c^{fit}_{n, \pm m}$ ) when $a_{fit} \neq a_w$ .
Taylor Expansion: The derived integral equation is expanded in powers of the radius error ( $\delta a_w = a_{fit} - a_w$ ). This expansion explicitly reveals that the error in the fitted coefficient depends on the product of the radial slope of the Zernike polynomial and the polynomial itself.
Analytical Evaluation: The author simplifies the notation using combinatorial coefficients ( $A_{n,m,k}$ ) to evaluate the overlap integrals exactly. This allows for the calculation of fitted coefficients and cross-coefficients (orthogonality breakdown) for specific orders ( $n$ ) and error magnitudes (e.g., $\pm 0.1\%$ and $\pm 1\%$ ).
Comparison: The results are compared against the "fringe normalized" property of Zernikes, where the radial function $R_n^m(1) = 1$ , and the behavior of the radial slope as $r \to 1$ .

Key Contributions and Results

Sensitivity Scaling: The analysis demonstrates that the absolute value of the radial slope of Zernike polynomials increases significantly as the order $n$ increases and $r$ approaches 1. Consequently, the sensitivity of the fitted coefficients to radius errors scales with the order.
Quantitative Error Bounds: For a radius uncertainty of $\pm 1\%$ , the error in the fitted coefficient for a Zernike of order $n$ is roughly equal to $n$ percent. For example, fitting a 40th-order term with a 1% radius error results in a coefficient error of approximately 40%.
Loss of Orthogonality: The paper shows that when $a_{fit} \neq a_w$ , the orthogonality of the Zernike basis is broken. This leads to non-zero cross-coefficients, where a single true coefficient in the wavefront "leaks" into multiple fitted coefficients. Figures 2 and 3 illustrate this, showing that for a 1% error, the overlap integrals for high orders ( $n > 30$ ) deviate drastically from the theoretical zero (or true self-overlap), with values reaching magnitudes of -1350 or -792 for $n=40$ .
Precision Requirements: To achieve a fit within a few percent of the true value for coefficients up to the 40th order, the radius of the wavefront data must be known to within $\pm 0.1\%$ or less. This implies that for a detector array, the wavefront diameter must span approximately 1,000 pixels or more to support such high-order fitting.

Significance and Claims
The paper concludes that, from a practical standpoint, fitting wavefront data to "higher and higher order Zernikes" rapidly becomes a "useless exercise" unless the true edge of the wavefront is known with sufficient precision. The author argues that the increasing sensitivity renders the fitted coefficient values "virtually meaningless" above a certain order for any given level of radius uncertainty.

Instead of pursuing high-order Zernike fits, the paper suggests the established alternative: fitting a fixed number of lower-order Zernikes, subtracting this from the data, and analyzing the residual wavefront using Fourier Transforms to derive an annular Power Spectral Density (PSD). This approach avoids the orthogonality breakdown and sensitivity issues associated with high-order polynomial fitting on uncertain boundaries. The paper maintains that these conclusions apply generally to any set of radial polynomial basis functions defined on the unit circle.

The Big Picture: The "Fuzzy Edge" Problem

The Analogy: The Stretchy Rubber Sheet

The "Cross-Talk" Problem

What Should We Do Instead?

The Bottom Line

More like this