The Hamming weight of $x \in \{0, 1\}^n$, i.e. the number of coordinates equal to
true; defined as the Hamming distance from the all-false vector.
Instances For
The open Hamming ball of radius $r$ centred at the origin in $\{0, 1\}^n$: $\{x : \operatorname{wt}(x) < r\}$.
Instances For
The bit-flip involution on the Hamming cube: $(\operatorname{flip}(x))_i = \neg x_i$ for every coordinate $i$.
Instances For
The bit-flip map is an involution: $\operatorname{flip}(\operatorname{flip}(x)) = x$.
Flipping all bits complements the weight: $\operatorname{wt}(\operatorname{flip}(x)) = n - \operatorname{wt}(x)$.
The bit-flip map is injective (in fact, a bijection, since it is its own inverse).
The Hamming ball of radius $\lfloor n/2 \rfloor$ contains at most half of the cube: $|B(n, n/2)| \le 2^{n-1}$. The proof uses bit-flipping to inject the ball into its complement.
The open Hamming ball of radius $n/2$ around the origin coincides with the closed Hamming ball of radius $n/2 - 1$, hence is a "Hamming ball" in the Harper sense.
The $0$-expansion of a set is the set itself: $S_0 = S$.
The $t$-expansion of the empty set is empty.
Specialization of Harper's vertex-isoperimetric inequality: any $A \subseteq \{0, 1\}^n$ with $|A| \ge 2^{n-1}$ has $t$-expansion at least as large as that of the Hamming ball of radius $\lfloor n/2 \rfloor$.
The Hamming weight equals the number of true coordinates: $\operatorname{wt}(x) = |\{i : x_i = \text{true}\}|$.
The Hamming distance from any point to itself is zero.
Triangle-style inequality for the Hamming weight: $\operatorname{wt}(x) \le \operatorname{wt}(y) + d(x, y)$.
Flipping a subset $T$ of the $1$-coordinates of $x$ to $0$ produces a vector $y$ at Hamming distance $|T|$ from $x$ and of weight $\operatorname{wt}(x) - |T|$.
The $t$-expansion of the Hamming ball $B(n, n/2)$ is the Hamming ball $B(n, n/2 + t)$ of larger radius.
Chernoff upper-tail bound for the cardinality of the heavy slice of the cube: the number of points $x \in \{0, 1\}^n$ with $\operatorname{wt}(x) \ge n/2 + t$ is strictly less than $e^{-2t^{2}/n} \cdot 2^{n}$.
Rapid expansion from half to $1 - \varepsilon$ (Theorem 9.4.5). For any $A \subseteq \{0, 1\}^n$ with $|A| \ge 2^{n-1}$ and any $t \ge 1$, the $t$-expansion satisfies $|A_t| > (1 - e^{-2t^{2}/n}) \cdot 2^{n}$.
Hamming distance is symmetric: $d(x, y) = d(y, x)$.
Triangle inequality for the Hamming distance: $d(x, z) \le d(x, y) + d(y, z)$.
The $t$-expansion of the complement of $A_t$ is disjoint from $A$: any point reachable within distance $t$ from outside $A_t$ cannot lie in $A$ itself.
The iterated expansion is contained in a single larger expansion: $(A_s)_t \subseteq A_{s + t}$.
The Hamming cube $\{0, 1\}^n$ has exactly $2^n$ elements.
Bootstrap lemma: if $|A| \ge e^{-2t^{2}/n} \cdot 2^{n}$, then the $t$-expansion of $A$ already has size at least $2^{n-1}$.
Rapid expansion from $\varepsilon$ to $1 - \varepsilon$ (Theorem 9.4.6). If $|A| \ge e^{-2t^{2}/n} \cdot 2^{n}$, then $|A_{2t}| \ge (1 - e^{-2t^{2}/n}) \cdot 2^{n}$.
Rapid expansion in the Hamming cube (continuous form). If $|A| \ge \varepsilon \cdot 2^{n}$ for some $\varepsilon \in (0, 1)$, then for $t = \lceil \sqrt{\log(1/\varepsilon) \cdot n / 2} \rceil$, $|A_{2t}| \ge (1 - \varepsilon) \cdot 2^{n}$.