Power Functions and Scaling Laws

Learning Objectives

By the end of this section, you will be able to:

1. Define a power function $f(x) = k\,x^{n}$ and tell it apart from an exponential function $f(x) = a^{x}$.
2. Recognize the qualitative shape of $x^{n}$ for any real exponent — including negative and fractional — at a single glance.
3. Use the log-log trick to turn any power law into a straight line and read off the exponent.
4. Apply scaling laws to real phenomena: the square–cube law in biology, Kepler's third law, the inverse-square law of gravity and light, and Kleiber's law of metabolism.
5. Preview the power rule $\frac{d}{dx}\,x^{n} = n\,x^{n-1}$ and compute the slope of a power function at any point.
6. Implement these ideas in Python: plot the family, and fit an unknown exponent from data.

The Big Picture: One Knob, A Whole Universe

"If you can change one number and watch the entire shape of the world change with it, you have found a power function."

A power function is a function with exactly one moving part: an exponent. Write it as $f(x) = k\,x^{n}$, and everything you care about — whether the curve grows or shrinks, how fast, what it does near zero, what it does at infinity, whether doubling the input doubles the output or makes it eight times bigger — is decided by the single number $n$. The coefficient $k$ just stretches the picture vertically.

Why this one family keeps reappearing

When a system has no preferred scale — when zooming in by a factor of 10 just rescales the picture rather than revealing new structure — the laws that govern it are power laws. That is why the same algebra describes phenomena that look unrelated:

Mathematical Definition

A power function is any function of the form

where $k$ is a real constant (the coefficient) and $n$ is a real constant (the exponent). The input $x$ is the only variable.

Power function vs exponential function — never confuse these

Both involve raising-to-a-power, but the variable lives in different places:

The Power Function Explorer

Before any algebra, build the intuition with your hand. Drag the $n$ slider through the whole range. Watch the curve morph continuously between every shape we'll catalogue below. Try the preset buttons. Drag $x_{0}$ and notice how the orange tangent rotates — that's the derivative we'll meet at the end of the section.

A Tour Through the Family

Now let's name what you just saw on the explorer. With $k = 1$ for simplicity, every value of the dial $n$ lands somewhere on this map:

The three regimes you should always recognize

Symmetry: Even, Odd, and Neither

When $n$ is an integer, the parity of $n$ controls the symmetry of the graph:

• n even (n = 2, 4, 6, …): the graph is symmetric about the y-axis. $f(-x) = f(x)$. Negative inputs produce the same output as their positives.
• n odd (n = 1, 3, 5, …): the graph is symmetric about the origin. $f(-x) = -f(x)$. A rotation by 180° leaves the graph unchanged.
• n not an integer (e.g., n = 0.5): the function isn't even defined for negative $x$ in the real numbers, because $(-1)^{1/2}$ isn't real. Restrict the domain to $x \ge 0$.

Worked Example: f(x) = 2x³

Let's put one specific power function under the microscope. Take $f(x) = 2x^{3}$ — a cubic with vertical stretch $k = 2$. We'll evaluate it, compute its slope at three points using the power rule, and verify every number against an explicit table.

The Log-Log Trick — How Scientists Discover Hidden Power Laws

Suppose you measure a quantity $y$ at several values of $x$ and suspect a power law relation $y = k\,x^{n}$, but you don't know $n$. The whole experiment looks like a curved blob on a regular plot. There is a beautiful trick — old, simple, and used everywhere from astrophysics to ML scaling laws — that turns the curve into a straight line.

The derivation in three lines

Start from the hypothesis. Take the logarithm of both sides. Use the identity $\log(ab) = \log a + \log b$ and $\log(x^{n}) = n\log x$:

Now substitute $Y = \log y$, $X = \log x$, $b = \log k$:

That is a straight line with slope $n$ and intercept $b = \log k$. Plotting your data on log-log axes turns the "is this a power law?" question into the much easier "does this look like a straight line?".

Scaling Laws in the Wild

Power functions describe nature's favorite kind of relationship: when one quantity is a fixed-power function of another. Here are four you'll meet again.

1. Kepler's third law — orbital period vs distance

For any planet orbiting the Sun, $T^{2} = a^{3}$ when $T$ is in years and $a$ is the semi-major axis in astronomical units. Equivalently:

Kepler arrived at this empirically in 1619 — a century before Newton showed it falls out of gravity being $1/r^{2}$. Two power laws nested inside each other.

2. The inverse-square law — light, gravity, radio

A point source spreads its energy over the surface of a sphere of radius $r$. Surface area is $4\pi r^{2}$, so intensity must be $\propto 1/r^{2}$:

Step twice as far from the source, get one quarter the brightness. Step ten times farther, get one hundredth. That single number $n = -2$ explains why the night sky is dark and why your phone's flashlight is useless past a few meters.

3. Kleiber's law — metabolism vs body mass

Across mammals from mice to elephants, the resting metabolic rate scales as:

An elephant is 200,000× heavier than a rabbit but burns only ~500× more energy per second. Cells in big animals work slower, and that single 3/4 exponent is the reason a mouse's heart races at 600 bpm while an elephant's plods at 30.

4. Neural-network scaling laws

A 21st-century example: empirically, the loss of a well-trained transformer language model decreases as a power of the number of parameters:

with $\alpha \approx 0.076$. Doubling parameters knocks the loss down by a fixed factor — the same algebra Kepler used for planets, applied to GPT-class models.

The Square–Cube Law — A Power Function Battle

Two power functions, fighting each other. Take a sphere of radius $r$:

• surface area grows like $r^{2}$: $A = 4\pi r^{2}$
• volume grows like $r^{3}$: $V = \tfrac{4}{3}\pi r^{3}$

The ratio surface-to-volume is what really matters for heat loss, oxygen exchange, and structural load. Divide:

That "3/r" is itself a power function with $n = -1$. So as creatures get bigger, their surface-to-volume ratio shrinks like $1/r$. Slide the radius and watch the numbers.

Calculus Preview: The Power Rule

Every power function has an absurdly clean derivative. We'll prove it carefully in Chapter 4 using limits, but you can already use it:

Combined with the constant-multiple rule (which we'll also justify):

Why does this make sense?

Look back at the explorer with the tangent on. At $x_{0} = 1$, the slope is $k\cdot n$. So:

• For $f(x) = x^{2}$ at $x = 1$, slope = 2.
• For $f(x) = x^{3}$ at $x = 1$, slope = 3.
• For $f(x) = 1/x$ at $x = 1$, slope = −1.

Higher exponent ⇒ steeper at the pivot. That is exactly what your eye sees in the interactive plot.

Python: Plotting the Family

Time to translate the math into code. The following script draws six members of the power-function family on a single axis. Read along with the side-by-side cards on the left — every line is annotated, including every NumPy function and its arguments.

np.power(np.array([1, 2, 3]), 2) → array([1, 4, 9])

np.power(2.0, 0.5) → 1.4142 (square root). np.power(2.0, -1) → 0.5 (reciprocal).

np.linspace(0.01, 4, 5) → [0.01, 1.0075, 2.005, 3.0025, 4.0]

1^-2 = 1, 1^0.5 = 1, 1^3 = 1 — all the same point.

1import numpy as np
2import matplotlib.pyplot as plt
3
4def power(x, n, k=1.0):
5    """f(x) = k * x^n — the general power function."""
6    return k * np.power(x, n)
7
8x = np.linspace(0.01, 4, 400)
9
10exponents = [-2, -1, 0.5, 1, 2, 3]
11labels    = ["1/x^2", "1/x", "sqrt(x)", "x", "x^2", "x^3"]
12colors    = ["#ef4444", "#f97316", "#eab308",
13             "#22c55e", "#3b82f6", "#8b5cf6"]
14
15plt.figure(figsize=(10, 6))
16for n, label, color in zip(exponents, labels, colors):
17    y = power(x, n)
18    plt.plot(x, y, color=color, linewidth=2, label=label)
19
20# Every member of the family passes through (1, 1) when k = 1
21plt.scatter([1], [1], color="white", edgecolors="black",
22            s=80, zorder=5, label="(1, 1)")
23
24plt.xlim(0, 4)
25plt.ylim(0, 6)
26plt.xlabel("x")
27plt.ylabel("f(x) = x^n")
28plt.title("The Power Function Family")
29plt.legend()
30plt.grid(True, alpha=0.3)
31plt.show()

Python: Discovering an Exponent from Data

Now the inverse problem: given a cloud of measurements, what power law generated them? The log-log trick makes this a one-line linear fit. We'll use synthetic data $y = 2\,x^{1.5}$ so you can verify the fit recovers the truth.

log(k · x^n) = log(k) + n · log(x).

np.polyfit([0, 1, 2], [1, 3, 5], 1) → array([2., 1.]) — slope 2, intercept 1.

1import numpy as np
2
3# Synthetic measurements: y = 2 * x^1.5 (we'll pretend we don't know n)
4x = np.array([1, 2, 4, 10, 100], dtype=float)
5y = np.array([2.0, 5.6569, 16.0, 63.2456, 2000.0])
6
7# Step 1 — take logs of both sides
8#   y = k * x^n
9#   log y = log k + n * log x
10log_x = np.log10(x)
11log_y = np.log10(y)
12
13# Step 2 — fit a straight line through (log_x, log_y)
14slope, intercept = np.polyfit(log_x, log_y, 1)
15
16# Step 3 — read off the recovered constants
17n_hat = slope
18k_hat = 10 ** intercept
19
20print(f"recovered n = {n_hat:.4f}   (true value: 1.5000)")
21print(f"recovered k = {k_hat:.4f}   (true value: 2.0000)")

Run it locally and you'll see:

recovered n = 1.5000   (true value: 1.5000)
recovered k = 2.0000   (true value: 2.0000)

Common Pitfalls

Summary

Power functions are the "one knob" family. The single exponent $n$ controls everything about the shape and is the reason this family describes phenomena from planets to mice to language models.

Key formulas

Key takeaways

1. A power function is $f(x) = k\,x^{n}$: variable base, constant exponent. Not the same as an exponential.
2. The exponent $n$ alone decides whether you get growth, decay, a root, a line, or a hyperbola.
3. Every curve in the family pivots through $(1, k)$.
4. Even integer $n$ ⇒ y-axis symmetry. Odd integer $n$ ⇒ origin symmetry.
5. The log-log trick turns any power law into a straight line of slope $n$. This is how exponents are measured in experiments.
6. Scaling laws govern biology (Kleiber), physics (inverse-square), astronomy (Kepler), and modern ML (neural scaling laws).
7. The power rule $\frac{d}{dx}\,x^{n} = n\,x^{n-1}$ will carry us through all of differential calculus.

Coming Next: Now that you can read the shape of any power function at a glance, we'll meet a different beast — exponential functions — where the variable lives in the exponent and growth becomes truly explosive.