Binomial Theorem

Deep Breakdown

What it tells me (Situation Awareness): The binomial theorem gives the exact expanded form of a power of a two-term sum. $(a + x)^n$ produces $n+1$ terms, each of the form $\binom{n}{k} a^{n-k} x^k$ for $k = 0, 1, \ldots, n$. The coefficient $\binom{n}{k}$ ("n choose k") counts the number of ways to pick exactly $k$ factors of $x$ from the $n$ factors in the product $(a+x)(a+x)\cdots(a+x)$. In DA/DS, the binomial theorem appears in: the binomial probability distribution $P(X = k) = \binom{n}{k} p^k (1-p)^{n-k}$ (directly this formula with $a = 1-p$ and $x = p$), Taylor/linear approximations $(1 + r)^n \approx 1 + nr$ for small $r$ (the first-order term of the expansion), compound growth and variance propagation, and polynomial feature expansion in machine learning (expanding $(x_1 + x_2)^2$ to get cross-features $x_1^2, 2x_1 x_2, x_2^2$). The theorem is also the theoretical foundation of Pascal's triangle.

Flowchart

Real-world use

Where it is used

Binomial distribution: the probability that exactly $k$ successes occur in $n$ independent Bernoulli trials with success probability $p$ is $P(X = k) = \binom{n}{k} p^k (1-p)^{n-k}$ — this is exactly the binomial theorem applied to $(p + (1-p))^n = 1$. Compound growth approximation: $(1 + r)^n \approx 1 + nr$ for small $r$ (e.g., annual growth rate compounded $n$ times). Polynomial feature generation: for two features $x_1, x_2$, the degree-2 expansion $(x_1 + x_2)^2 = x_1^2 + 2x_1 x_2 + x_2^2$ generates the polynomial basis used in kernel SVMs and polynomial regression. Variance of a sum of random variables (when terms are independent).

Problem it solves

Expanding $(a + x)^n$ by hand via repeated multiplication becomes infeasible for large $n$. The binomial theorem gives the exact result directly, without iterative multiplication. Each term is computed independently using the formula $\binom{n}{k} a^{n-k} x^k$, so you can extract any single term (e.g., the $x^3$ term in $(2 + x)^5$) without computing all others.

How it solves the problem

Consider the product $(a+x)^n = (a+x)(a+x)\cdots(a+x)$ ($n$ factors). Each term in the expansion is produced by choosing one of $\{a, x\}$ from each of the $n$ factors. To get a term with exactly $k$ factors of $x$ (and $n-k$ factors of $a$), you must choose $x$ from $k$ specific factors — the number of ways to do this is $\binom{n}{k}$. The resulting term is $\binom{n}{k} a^{n-k} x^k$. Summing over all possible values of $k$ (from 0 to $n$) gives the full expansion.

Example in a business

A conversion rate analyst wants the probability that exactly 3 out of 10 randomly selected users convert, given a 20% conversion rate. Using the binomial theorem with $a = 0.8$ (non-conversion), $x = 0.2$ (conversion), $n = 10$, $k = 3$: $P(X = 3) = \binom{10}{3} (0.8)^7 (0.2)^3 = 120 \times 0.2097 \times 0.008 = 0.2013 \approx 20.1\%$. The binomial theorem is the backbone of this calculation.

The formula, piece by piece

Breaking the notation:

• $(a + x)^n$: a binomial ("two-term") expression raised to the power $n$ — the quantity being expanded.
• $\sum_{k=0}^{n}$: sum over all integer values of $k$ from 0 to $n$ inclusive — $n+1$ terms total.
• $\binom{n}{k}$: the binomial coefficient "n choose k" — $\frac{n!}{k!(n-k)!}$ — the number of ways to select $k$ items from $n$ without order.
• $a^{n-k}$: $a$ raised to the power $n-k$ — decreases as $k$ increases (first term: $a^n$, last term: $a^0 = 1$).
• $x^k$: $x$ raised to the power $k$ — increases as $k$ increases (first term: $x^0 = 1$, last term: $x^n$). First and last terms:
• $k = 0$: $\binom{n}{0} a^n x^0 = a^n$.
• $k = n$: $\binom{n}{n} a^0 x^n = x^n$. Middle term (when $n$ is even, $k = n/2$): $\binom{n}{n/2} a^{n/2} x^{n/2}$ — has the largest binomial coefficient. Special case $a = 1$: $(1 + x)^n = \sum_{k=0}^{n} \binom{n}{k} x^k$ — the standard form used for probability generating functions and small-$x$ approximations. Linear approximation: for small $|x/a|$: $(a + x)^n \approx a^n + n a^{n-1} x$ (first two terms only).

The formula in words

"$(a + x)^n$ expands into $n+1$ terms. Each term is: how many ways to pick $k$ copies of $x$ from $n$ factors ($= \binom{n}{k}$), times $a$ to the remaining power ($= a^{n-k}$), times $x$ to the chosen power ($= x^k$). The powers of $a$ decrease and the powers of $x$ increase as you go from $k=0$ to $k=n$. In Python: from sympy import expand, symbols; expand((a+x)**n) or use scipy.special.comb(n, k) to compute individual coefficients."

Critical Concepts

1. Binomial coefficients are symmetric: $\binom{n}{k} = \binom{n}{n-k}$. The coefficient of $a^{n-k} x^k$ equals the coefficient of $a^k x^{n-k}$. This corresponds to the symmetry of Pascal's triangle.
2. Sum of coefficients = $2^n$: setting $a = x = 1$: $(1+1)^n = 2^n = \sum_{k=0}^n \binom{n}{k}$. Total number of subsets of an $n$-element set.
3. Binomial distribution connection: $P(X = k) = \binom{n}{k} p^k (1-p)^{n-k}$ is the $k$-th term of $(1-p+p)^n = 1^n = 1$ — the binomial theorem guarantees all probabilities sum to 1.
4. First-order approximation: $(1+x)^n \approx 1 + nx$ for $|x| \ll 1$. In finance: $(1 + r)^n \approx 1 + nr$ for small interest rate $r$. In ML: gradient approximations for small perturbations.
5. Polynomial features: the degree-2 expansion of $(x_1 + x_2)^2 = x_1^2 + 2x_1 x_2 + x_2^2$ generates the polynomial kernel basis.

More generally, the PolynomialFeatures transformer in scikit-learn applies the binomial (and multinomial) theorem to generate interaction terms.

Step-by-step calculation (tiny example)

Task: expand $(2 + x)^3$.

Step 1: Identify $a$, $x$, $n$ $a = 2$, $x = x$, $n = 3$. There will be 4 terms ($k = 0, 1, 2, 3$).

Step 2: Compute each binomial coefficient $\binom{3}{0} = 1$, $\binom{3}{1} = 3$, $\binom{3}{2} = 3$, $\binom{3}{3} = 1$.

Step 3: Write each term $k=0$: $1 \cdot 2^3 \cdot x^0 = 8$. $k=1$: $3 \cdot 2^2 \cdot x^1 = 12x$. $k=2$: $3 \cdot 2^1 \cdot x^2 = 6x^2$. $k=3$: $1 \cdot 2^0 \cdot x^3 = x^3$.

Step 4: Assemble $(2 + x)^3 = 8 + 12x + 6x^2 + x^3$.

Step 5: Verify by substitution $x = 1$: $(2+1)^3 = 27$. Formula: $8 + 12 + 6 + 1 = 27$.

Code examples

Beginner example (Python)

# Binomial theorem expansion
from math import comb, factorial

def binomial_term(n: int, k: int, a: float, x: float) -> float:
    """Compute the k-th term of the binomial expansion of (a + x)^n."""
    return comb(n, k) * (a ** (n - k)) * (x ** k)

def binomial_expansion(n: int, a: float, x: float) -> list[float]:
    """ Expand (a + x)^n and return all n+1 terms. Args: n: Power (non-negative integer). a: First term of the binomial. x: Second term of the binomial. Returns: List of n+1 term values. """
    return [binomial_term(n, k, a, x) for k in range(n + 1)]   # Expand (2 + x)^3 symbolically (x=1 for numeric check)
n_val = 3
a_val = 2.0
terms = binomial_expansion(n_val, a_val, 1.0)
print(f"Expansion of ({a_val} + x)^{n_val}:")
for k, term in enumerate(terms):
    coeff = comb(n_val, k) * (a_val ** (n_val - k))
print(f" k={k}: C({n_val},{k}) × {a_val}^{n_val - k} × x^{k} = {coeff:.0f}x^{k} (at x=1: {term:.1f})")
print(f" Sum at x=1: {sum(terms):.1f} vs ({a_val}+1)^{n_val} = {(a_val + 1) ** n_val:.1f}")   # Binomial distribution: P(X=k) for n=10,
p=0.2
n_trials = 10
p = 0.2
print(f"\nBinomial distribution: n={n_trials}, p={p}")
print(f"{'k':>4} {'P(X=k)':>10} {'term check':>12}")
for k in range(n_trials + 1):
    prob = binomial_term(n_trials, k, 1 - p, p)
print(f"{k:>4} {prob:>10.6f}")
total = sum(binomial_term(n_trials, k, 1 - p, p) for k in range(n_trials + 1))
print(f"Sum = {total:.8f} (should be exactly 1.0)")   # Linear approximation (1 + r)^n ≈ 1 + nr for small r
r = 0.02   # 2% growth rate
n = 5
exact = (1 + r) ** n
approx = 1 + n * r
print(f"\nLinear approximation: (1 + {r})^{n}")
print(f" Exact: {exact:.6f}")
print(f" Approx: {approx:.6f}")
print(f" Error: {abs(exact - approx):.8f}")

Beginner example (R)

# Binomial theorem in R # Expand (a + x)^n term by term
binomial_term <- function(n, k, a, x) choose(n, k) * a^(n - k) * x^k
binomial_expansion <- function(n, a, x) { sapply(0:n, function(k) binomial_term(n, k, a, x))
}  # Expand (2 + x)^3 at
x=1
n_val <- 3; a_val <- 2.0
terms <- binomial_expansion(n_val, a_val, 1.0)
cat(sprintf("Expansion of (%.0f + x)^%d at x=1:\n", a_val, n_val))
for (k in 0:n_val) { coeff <- choose(n_val, k) * a_val^(n_val - k)
  cat(sprintf(" k=%d: coeff=%.0f term=%.1f\n", k, coeff, terms[k + 1]))
}
cat(sprintf(" Sum=%.1f vs (%.0f+1)^%d=%.1f\n", sum(terms), a_val, n_val, (a_val + 1)^n_val))  # Binomial distribution
n_trials <- 10; p <- 0.2
cat(sprintf("\nBinomial: n=%d, p=%.2f\n", n_trials, p))
probs <- sapply(0:n_trials, function(k) binomial_term(n_trials, k, 1 - p, p))
cat(sprintf("Sum of all P(X=k) = %.8f\n", sum(probs)))  # Linear approximation
r <- 0.02; n <- 5
cat(sprintf("\n(1+%.2f)^%d: exact=%.6f approx=%.6f\n", r, n, (1+r)^n, 1 + n*r))

Professional example (Python)

import numpy as np
import pandas as pd
from math import comb
from scipy.stats import binom as scipy_binom   # Scenario: A/B test power analysis using the binomial distribution
# For a given baseline conversion rate and minimum detectable effect,
# compute the full binomial probability distribution for both variants
# and compare the theoretical PMF to the binomial theorem expansion
BASELINE_RATE = 0.10   # control conversion rate (p_control)
MDE = 0.03   # minimum detectable effect (absolute)
VARIANT_RATE = BASELINE_RATE + MDE   # 13% variant rate
N_TRIALS = 200   # sample size per variant

def binomial_pmf_exact(n: int, p: float) -> pd.DataFrame:
    """ Compute the exact binomial PMF for X ~ Bin(n, p) using the binomial theorem. P(X = k) = C(n, k) * p^k * (1-p)^(n-k) Args: n: Number of trials. p: Success probability per trial. Returns: DataFrame with columns: k, coefficient, p_k, 1-p_nk, pmf, cdf. """
    q = 1 - p
    rows = []
    cumulative = 0.0
    for k in range(n + 1):
        coeff = comb(n, k)
    pmf_k = coeff * (p ** k) * (q ** (n - k))
    cumulative += pmf_k
    rows.append({ "k": k, "C(n,k)": coeff, "p^k": p ** k, "(1-p)^(n-k)": q ** (n - k), "pmf": pmf_k, "cdf": cumulative, })
    return pd.DataFrame(rows)
control_pmf = binomial_pmf_exact(N_TRIALS, BASELINE_RATE)
variant_pmf = binomial_pmf_exact(N_TRIALS, VARIANT_RATE)   # Summary statistics
control_mean = (control_pmf["k"] * control_pmf["pmf"]).sum()
variant_mean = (variant_pmf["k"] * variant_pmf["pmf"]).sum()   # Probability of observing k >= threshold under each variant
k_threshold = int(np.ceil(N_TRIALS * (BASELINE_RATE + MDE / 2)))
p_exceed_control = 1 - control_pmf.loc[control_pmf["k"] < k_threshold, "pmf"].sum()
p_exceed_variant = 1 - variant_pmf.loc[variant_pmf["k"] < k_threshold, "pmf"].sum()
print(f"Binomial PMF via Binomial Theorem: P(X=k) = C(n,k)·p^k·(1-p)^(n-k)")
print(f"n = {N_TRIALS} | control p = {BASELINE_RATE} | variant p = {VARIANT_RATE}")
print("=" * 72)
print(f"{'Metric':<30} {'Control':>12} {'Variant':>12}")
print(f"{'Expected conversions E[X]':<30} {control_mean:>12.2f} {variant_mean:>12.2f}")
print(f"{'P(X >= ' + str(k_threshold) + ')':<30} {p_exceed_control:>12.4f} {p_exceed_variant:>12.4f}")
print(f"{'PMF sums to 1 (verify)':<30} {control_pmf['pmf'].sum():>12.8f} {variant_pmf['pmf'].sum():>12.8f}")
print(f"\nNote: PMF sum = 1 is guaranteed by the binomial theorem: Σ C(n,k)·p^k·(1-p)^(n-k) = (p + 1-p)^n = 1^n = 1")

Professional example (R)

# A/B test power: binomial PMF for control vs variant using binomial theorem
BASELINE_RATE <- 0.10
MDE <- 0.03
VARIANT_RATE <- BASELINE_RATE + MDE
N_TRIALS <- 200
binomial_pmf_exact <- function(n, p) { k <- 0:n
  pmf <- choose(n, k) * p^k * (1 - p)^(n - k) data.frame(k = k, pmf = pmf, cdf = cumsum(pmf))
}
control_pmf <- binomial_pmf_exact(N_TRIALS, BASELINE_RATE)
variant_pmf <- binomial_pmf_exact(N_TRIALS, VARIANT_RATE)
control_mean <- N_TRIALS * BASELINE_RATE
variant_mean <- N_TRIALS * VARIANT_RATE
k_threshold <- 25
p_exceed_control <- 1 - sum(control_pmf$pmf[control_pmf$k < k_threshold])
p_exceed_variant <- 1 - sum(variant_pmf$pmf[variant_pmf$k < k_threshold])
cat(sprintf("n = %d | control p = %.2f | variant p = %.2f\n", N_TRIALS, BASELINE_RATE, VARIANT_RATE))
cat(sprintf("E[X] control: %.2f variant: %.2f\n", control_mean, variant_mean))
cat(sprintf("P(X >= %d) control: %.4f variant: %.4f\n", k_threshold, p_exceed_control, p_exceed_variant))
cat(sprintf("PMF sum control: %.8f variant: %.8f\n", sum(control_pmf$pmf), sum(variant_pmf$pmf)))

Source: AIO p. 43

#algebra#algebra-and-factoring#binomial-theorem-and-expansions