Ellipse

, $A = \pi a b$

Deep Breakdown

What it tells me (Situation Awareness): The ellipse $\frac{x^2}{a^2} + \frac{y^2}{b^2} = 1$ is the set of points where the sum of distances to two fixed foci is constant. With semi-major axis $a$ (along $x$) and semi-minor axis $b$ (along $y$), the area is $A = \pi a b$. The circle is the special case $a = b = R$. In DA/DS and statistics, the ellipse is the confidence region shape for a bivariate normal distribution: the 1-$\sigma$ contour of $\mathcal{N}(\mu, \Sigma)$ in 2D is an ellipse whose axes align with the eigenvectors of $\Sigma$ and whose half-axes lengths are $\sqrt{\lambda_1}$ and $\sqrt{\lambda_2}$ (the square roots of the eigenvalues). PCA scatter plots routinely overlay these confidence ellipses to visualize data spread and feature correlation.

Flowchart

Real-world use

Where it is used

Used when plotting confidence regions for bivariate normal distributions in 2D (PCA biplots, scatter plot overlays), in orbital mechanics analogies for understanding non-circular feature distributions, in quality control (Hotelling's $T^2$ statistic defines an ellipsoidal acceptance region), and in 2D anomaly detection where the normal region is an ellipse defined by the covariance structure of the training data.

Problem it solves

A circle assumes equal variance in all directions. Real data rarely has this property: feature pairs often have different scales and may be correlated. The ellipse generalizes the circle to capture anisotropic (direction-dependent) spread, making it the correct geometric object for modeling 2D Gaussian data, defining confidence regions, and visualizing correlation structure.

How it solves the problem

The ellipse equation $\frac{x^2}{a^2} + \frac{y^2}{b^2} = 1$ is the standard form aligned with the coordinate axes. For a rotated ellipse (correlated features), the general form is $\mathbf{x}^T \Sigma^{-1} \mathbf{x} = c^2$ where $\Sigma$ is the covariance matrix. Eigendecomposing $\Sigma = Q\Lambda Q^T$ gives the rotation angle (eigenvectors $Q$) and the axis lengths ($\sqrt{\lambda_i}$ scaled by $c$). The area formula $A = \pi a b$ follows from the substitution $u = x/a$, $v = y/b$ which transforms the ellipse to the unit circle with Jacobian $ab$.

Example in a business

A data scientist plots the scatter of weekly TV spend vs digital spend for 104 weeks, overlaid with the 95% confidence ellipse from the sample covariance matrix. The ellipse's tilt reveals the correlation direction; the axis lengths reveal the relative variability. If the ellipse is nearly circular, the two channels vary independently; if it is elongated, the channels co-vary (possibly due to budget constraints or seasonality). This informs media mix model assumptions.

The formula, piece by piece

• $\frac{x^2}{a^2}$: the contribution of the $x$-coordinate, normalized by the squared semi-major axis $a$; $a$ controls the horizontal extent
• $\frac{y^2}{b^2}$: the contribution of the $y$-coordinate, normalized by the squared semi-minor axis $b$; $b$ controls the vertical extent
• $a$: semi-major axis length; half the width along the $x$-axis; $a \geq b$ by convention
• $b$: semi-minor axis length; half the height along the $y$-axis
• $A = \pi a b$: area; reduces to $\pi r^2$ when $a = b = r$
• Eccentricity: $e = \sqrt{1 - b^2/a^2}$ (0 = circle, approaching 1 = very elongated) Full formulas: $\dfrac{x^2}{a^2} + \dfrac{y^2}{b^2} = 1$, $A = \pi a b$

The formula in words

"An ellipse is a stretched circle: squash a circle by factor $b/a$ in the $y$-direction and you get an ellipse with semi-axes $a$ and $b$. The area is $\pi$ times both half-axis lengths — if either half-axis is $r$, you recover the circle formula $\pi r^2$." Tiny numeric example: $a = 5$, $b = 3$. Equation: $\frac{x^2}{25} + \frac{y^2}{9} = 1$. Points on ellipse: $(5, 0)$ (right vertex), $(0, 3)$ (top vertex), $(-5, 0)$, $(0, -3)$. $A = \pi \cdot 5 \cdot 3 = 15\pi \approx 47.12$.

Critical Concepts

• Standard vs general form: standard form assumes axes aligned with coordinate axes. For a rotated/correlated ellipse: $\mathbf{x}^T \Sigma^{-1} \mathbf{x} = k^2$ where $k^2 = \chi^2_{2, 1-\alpha}$ (from chi-squared distribution with 2 df) for a $(1-\alpha)$ confidence region.
• Confidence region for bivariate normal: the $p$-level confidence ellipse has $k^2 = \chi^2_{2}(p)$. For 95%, $k^2 \approx 5.991$.
• Axis alignment from PCA: eigendecompose $\Sigma = Q\Lambda Q^T$; the eigenvectors give the ellipse orientation, and $\sqrt{\lambda_i \cdot k^2}$ gives the half-axis lengths.
• Area connection to determinant: $A_{\text{confidence region}} = \pi k^2 \sqrt{\det(\Sigma)}$. The determinant of the covariance matrix controls the total area of the confidence ellipse.
• Eccentricity: $e = 0$ is a circle; $e \to 1$ is an extremely elongated ellipse. High eccentricity in a confidence ellipse signals strong correlation between the two features.
• Perimeter: there is no simple closed-form for the ellipse perimeter (unlike area).

The Ramanujan approximation: $C \approx \pi[3(a+b) - \sqrt{(3a+b)(a+3b)}]$.

Step-by-step calculation (tiny example)

Step 1: Write the equation and identify vertices $a = 4$, $b = 2$. Equation: $\frac{x^2}{16} + \frac{y^2}{4} = 1$. Vertices: $(4, 0)$, $(-4, 0)$, $(0, 2)$, $(0, -2)$. Foci: $c = \sqrt{a^2 - b^2} = \sqrt{12} \approx 3.46$ along $x$: $(\pm 3.46, 0)$.

Step 2: Compute the area $A = \pi \cdot 4 \cdot 2 = 8\pi \approx 25.13$.

Step 3: Check a point is on the ellipse Point $(2\sqrt{2}, \sqrt{2})$: $\frac{(2\sqrt{2})^2}{16} + \frac{(\sqrt{2})^2}{4} = \frac{8}{16} + \frac{2}{4} = 0.5 + 0.5 = 1$.

Code examples

Beginner example (Python)

import numpy as np
import matplotlib.pyplot as plt

# -- Ellipse: equation, points on boundary, area -----------------------
A_SEMI = 5.0   # semi-major axis (x direction)
B_SEMI = 3.0   # semi-minor axis (y direction)

# Parametric form: x = a*cos(t), y = b*sin(t)
t = np.linspace(0, 2 * np.pi, 500)
x_ellipse = A_SEMI * np.cos(t)
y_ellipse = B_SEMI * np.sin(t)

# Check a few specific points
test_points = np.array([[5.0, 0.0], [0.0, 3.0], [3.0, 2.4]])
for pt in test_points:
    val = (pt[0] / A_SEMI)**2 + (pt[1] / B_SEMI)**2
    status = "on ellipse" if np.isclose(val, 1.0) else ("inside" if val < 1 else "outside")
    print(f" Point ({pt[0]:.1f}, {pt[1]:.1f}): x^2/a^2 + y^2/b^2 = {val:.4f} -> {status}")

area = np.pi * A_SEMI * B_SEMI
eccentricity = np.sqrt(1 - (B_SEMI / A_SEMI)**2)
print(f"\nSemi-axes: a={A_SEMI}, b={B_SEMI}")
print(f"Area A = pi*a*b = {area:.4f}")
print(f"Eccentricity e = {eccentricity:.4f} (0=circle, ~1=very elongated)")

Beginner example (R)

# -- Ellipse: equation, points on boundary, area -----------------------
A_SEMI <- 5.0
B_SEMI <- 3.0

# Parametric form
t <- seq(0, 2 * pi, length.out = 500)
x_ell <- A_SEMI * cos(t)
y_ell <- B_SEMI * sin(t)

# Check points
test_pts <- rbind(c(5, 0), c(0, 3), c(3, 2.4))
for (i in seq_len(nrow(test_pts))) {
  pt <- test_pts[i, ]
  val <- (pt[1] / A_SEMI)^2 + (pt[2] / B_SEMI)^2
  status <- if (isTRUE(all.equal(val, 1.0))) "on ellipse" else if (val < 1) "inside" else "outside"
  cat(sprintf(" Point (%.1f, %.1f): val = %.4f -> %s\n", pt[1], pt[2], val, status))
}
area <- pi * A_SEMI * B_SEMI
eccentricity <- sqrt(1 - (B_SEMI / A_SEMI)^2)
cat(sprintf("\nArea A = pi*a*b = %.4f\n", area))
cat(sprintf("Eccentricity e = %.4f\n", eccentricity))

Professional example (Python)

import numpy as np
import pandas as pd
from scipy.stats import chi2
np.random.seed(42)   # -- Business scenario: 95% confidence ellipse for weekly media spend
# TV spend and digital spend for 104 weeks.
# Overlay the 95% confidence ellipse (chi-squared with 2 df) on the scatter.
# Report: semi-axes, area, eccentricity, and what the ellipse shape reveals.
N_WEEKS = 104
TV_MEAN, TV_STD = 20_000.0, 4_000.0
DIGITAL_MEAN, DIG_STD = 15_000.0, 3_000.0
CORRELATION = 0.65   # positive: both channels funded together # Simulate correlated spend data
cov_raw = np.array([ [TV_STD**2, CORRELATION * TV_STD * DIG_STD], [CORRELATION * TV_STD * DIG_STD, DIG_STD**2],
])
spend = np.random.multivariate_normal([TV_MEAN, DIGITAL_MEAN], cov_raw, size=N_WEEKS)
x_centered = spend - spend.mean(axis=0)   # Sample covariance matrix
cov_sample = np.cov(x_centered.T)   # 2x2 # Confidence ellipse: x^T Sigma^-1 x = chi2(2, 0.95) ~ 5.991
K_SQ = float(chi2.ppf(0.95, df=2))
eigvals, eigvecs = np.linalg.eigh(cov_sample)
# Sort eigenvalues descending
order = np.argsort(eigvals)[::-1]
eigvals = eigvals[order]
eigvecs = eigvecs[:, order]   # Ellipse semi-axes in original units
semi_a = float(np.sqrt(eigvals[0] * K_SQ))   # semi-major axis (along eigvec_0)
semi_b = float(np.sqrt(eigvals[1] * K_SQ))   # semi-minor axis (along eigvec_1)
area = np.pi * semi_a * semi_b
angle_deg = float(np.degrees(np.arctan2(eigvecs[1, 0], eigvecs[0, 0])))
eccentricity = float(np.sqrt(1 - (semi_b / semi_a)**2))
summary = pd.DataFrame({ "metric": [ "semi-major axis (EUR)", "semi-minor axis (EUR)", "rotation angle (deg)", "confidence ellipse area (EUR^2)", "eccentricity", "correlation implied", ], "value": [ f"{semi_a:,.0f}", f"{semi_b:,.0f}", f"{angle_deg:.2f}", f"{area:,.0f}", f"{eccentricity:.4f}", f"{CORRELATION:.2f}", ],
})
print(f"95% Confidence Ellipse for TV vs Digital Spend (chi2 threshold = {K_SQ:.3f}):")
print(summary.to_string(index=False))
print(f"\nInterpretation: eccentricity={eccentricity:.3f} confirms strong elongation along the TV-digital correlation axis. The ellipse tilt of {angle_deg:.1f} degrees reflects the positive correlation of {CORRELATION}. Area = pi * {semi_a:,.0f} * {semi_b:,.0f} = {area:,.0f} EUR^2 (not a physical unit, but proportional to total data spread).")

Source: 1300 p. 160

#linear-algebra#spatial-distance-norms-and-analytic-geometry#analytic-geometry-shapes-perimeter-area-volume