Boppana's Entropy Inequality

• Boppana's entropy inequality is a mathematical relation that connects the binary entropy of a squared variable to a scaled product of the variable and its entropy, establishing tight bounds for complexity problems.
• It is proven through calculus methods including auxiliary ratio analysis and monotonicity arguments, ensuring the inequality holds with equality only at specific critical points.
• Its generalizations, supported by automated formal proofs, provide essential insights for optimizing bounds in Boolean complexity and the union-closed sets conjecture.

Boppana's entropy inequality is a central analytic tool in the study of entropy functions and their combinatorial applications, most notably in the context of Boolean formula complexity and union-closed set conjectures. The inequality relates the value of the binary entropy function at a squared argument to a scaled product of the argument and the entropy itself, yielding tight bounds that are essential for both lower bound proofs in complexity theory and extremal set theory constructions.

1. Definitions and Preliminaries

The binary entropy function, $h \colon [0,1] \to [0, \ln 2]$, is defined by

$$h(x) = -x \log x - (1-x) \log(1-x), \quad h(0) = h(1) = 0,$$

where $\log$ denotes the natural logarithm. The golden ratio is given as

$\varphi = \frac{1+\sqrt{5}}{2} \approx 1.618.$

For analytic purposes, the entropy function can be extended to all real $x$ by replacing $x$ by $|x|$ in the logarithmic terms when necessary, and by setting $h(0)=h(1)=0$.

2. Statement of Boppana’s Entropy Inequality

Boppana’s original entropy inequality asserts that for all $x \in [0,1]$,

$h(x^2) \geq \varphi\, x\, h(x)$

with equality attained exactly at $x = 0$, $x = 1/\varphi$, or $x = 1$ (Boppana, 2023, Ho, 27 Jan 2026). This captures a self-similarity phenomenon in the binary entropy function under the squaring transformation and introduces the golden ratio as a critical scaling parameter.

3. Proof Techniques and Analytic Structure

The standard calculus approach introduces the auxiliary ratio

$q_2(x) = \frac{x\,h(x)}{h(x^2)}, \quad x \in (0,1)$

which extends to $q_2(0) = q_2(1) = 1/2$. Analysis of $q_2(x)$, along with its derivative structure and the monotonicity properties of the function

$U(x) = \frac{\log x \log(1-x)}{h(x)},$

demonstrates that $q_2(x)$ attains its maximum at $x = 1/\varphi$ and is strictly bounded by $\varphi$ elsewhere. The critical point solution $U(x) = U(x^2)$ within $(0,1)$ occurs precisely at $x^2 = 1-x$, i.e., $x = 1/\varphi$, leveraging strict monotonicity arguments for $U$ on $(0,1/2]$ (Ho, 27 Jan 2026).

A differential calculus proof recasts the inequality as non-negativity of

$f(x) = h(x^2) - \varphi x h(x)$

and employs explicit evaluations of roots and derivatives, including higher-order Taylor expansions near $x=0$ (using $h(x) \sim -x \log x$), to conclude positivity on the interior and specified equality points (Boppana, 2023).

4. Generalization to Power Parameters and Extrema

The inequality generalizes to real powers $k > 1$. Defining $\alpha_k$ as the unique positive solution to

$\alpha_k \, (1+\alpha_k)^{k-1} = 1$

the generalized form states that for all $x \in [0,1]$,

$\alpha_k\, h(x^k) \geq x^{k-1}\, h(x)$

with equality only at $x = 0$, $x = 1/(1+\alpha_k)$, or $x = 1$ (Ho, 27 Jan 2026). Proof proceeds via the study of the ratio

$q(x) = \frac{x^{k-1} h(x)}{h(x^k)},$

boundedly peaking at $x = 1/(1+\alpha_k)$ with boundary limits $q(0) = q(1) = 1/k < \alpha_k$. Convexity arguments and weighted AM–GM inequalities confirm $\alpha_k > 1/k$.

Table: Key Parameters in the Generalized Inequality

Parameter	Definition	Role
$\varphi$	$(1+\sqrt{5})/2$	Scaling for $k=2$
$\alpha_k$	Unique solution of $\alpha(1+\alpha)^{k-1}=1$	Scaling for general $k>1$
$x_\ast$	$1/\varphi$ or $1/(1+\alpha_k)$	Point of equality

5. Applications to Boolean Formula Complexity

In the 1980s, Boppana applied the entropy inequality to derive lower bounds for randomized Boolean formulas. By analyzing the amplification of success probabilities under repeated composition (such as series or parallel repetition), a two-variable extension comparable to

$h(p_1 p_2) \geq \varphi (p_1 h(p_2) + p_2 h(p_1))$

was formulated, with the one-variable case providing foundational bounds. These analytic tools yield results such as the minimal achievable error by formulas of size $s$ being $\exp(-\Omega(s))$ (Boppana, 2023).

6. Union-Closed Sets Conjecture and Extremal Set Theory

Boppana's entropy inequality has emerged as a central lemma in progress on Frankl's union-closed sets conjecture. Information-theoretic arguments associate marginal probabilities $p_i = \mathbb{P}(i \in S)$ for $S$ uniformly random in a union-closed family $\mathcal{F} \subseteq 2^{[n]}$. Entropy summation bounds $\sum_{i=1}^n h(p_i) \leq \ln |\mathcal{F}|$ can be strengthened using the inequality, especially when $p_i < 1/2$, improving lower bounds on the density of "popular" elements from $0.01$ to about $0.04$ (Boppana, 2023).

The generalized inequality yields analogues for approximate $k$-union closed systems. If $\mathcal{F}$ is $(1-\epsilon)$-approximate $k$-union closed, the density bound for some element is

$\frac{\alpha_k}{1+\alpha_k} - \delta$

with

$$\delta = \left(k\epsilon + 2\epsilon \frac{\log(1/\epsilon)}{\log|\mathcal{F}|}\right)^{1/(k-1)},$$

yielding $\alpha_k/(1+\alpha_k)$ for exact systems ($\epsilon = 0$) (Ho, 27 Jan 2026). This provides tight quantitative bounds in extremal set theory.

7. Formalization and Automated Proof Checking

The proof of the generalized Boppana entropy inequality for real $k > 1$ has been entirely formalized in Lean 4 using the Harmonic Aristotle library and Gemini 3 Pro Preview, integrating definitions, calculations, monotonicity, and extremal arguments in a machine-checked development. All sources are provided at https://github.com/boonsuan/entropy-inequality (Ho, 27 Jan 2026). This formal validation enhances confidence in the analytic structure and supports future exploration of entropy-based inequalities.

---

Boppana’s entropy inequality and its generalizations constitute a robust analytic framework, connecting entropy functions to structural bounds in Boolean complexity and set systems, with validated machine-checkable proofs and significant combinatorial applications.