Lee's Conjecture: Sharp Frobenius Inequality

• Lee’s Conjecture is defined as the optimal Frobenius norm inequality for matrices, demonstrating that the norm of A+B is bounded by sqrt((1+√2)/2) times the norm of |A|+|B|.
• The proof utilizes polar decompositions, trace identities, and the matrix Cauchy–Schwarz inequality to simplify earlier, more complex angle-based methods.
• Extensions of the conjecture to many-summand scenarios and Schatten p-norm comparisons underline its significance and broad impact in matrix analysis.

Lee’s conjecture is a sharp matrix inequality in $M_n$, the algebra of all $n\times n$ complex matrices, comparing the Frobenius norm of a sum $A+B$ with the Frobenius norm of the sum of the absolute values $|A|+|B|$. In the formulation treated by Zhang, it is the $p=2$ case of Eun-Young Lee’s 2010 problem of determining the smallest constant $c_p$ for which $\|A+B\|_p\le c_p\||A|+|B|\|_p$ holds in Schatten $p$-norms. The conjectured Frobenius constant is

$c_2=\sqrt{\frac{1+\sqrt2}{2}},$

and the inequality is now known to be correct and optimal. Lin and Zhang proved it in 2022, and Zhang later gave a substantially shorter proof based only on the matrix Cauchy–Schwarz inequality (Zhang, 3 Jul 2025).

1. Mathematical formulation

For $A\in M_n$, the matrix absolute value is

$n\times n$0

and similarly $n\times n$1. The Frobenius norm is

$n\times n$2

where $n\times n$3 are the singular values. In this setting, Lee’s conjecture asserts that for all $n\times n$4,

$n\times n$5

Equivalently,

$n\times n$6

These two forms are identical after dividing by $n\times n$7 and taking square roots (Zhang, 3 Jul 2025).

A persistent point of confusion is the value of the constant. In reproduced versions of the theorem statement, the formula sometimes appears with typographical corruption. The correct norm inequality constant is

$n\times n$8

whereas

$n\times n$9

is the coefficient in the squared inequality. This distinction is explicit in the reconstructed theorem and proof (Zhang, 3 Jul 2025).

2. Origin within the Schatten-norm problem

The conjecture originated in Eun-Young Lee’s 2010 paper “Rotfel’d type inequalities for norms,” which asked for the optimal constant $A+B$0 in

$A+B$1

The Frobenius case corresponds to $A+B$2. According to Zhang’s account, Lee observed that even determining $A+B$3 appeared difficult and conjectured the exact value

$A+B$4

Thus Lee’s conjecture is not an isolated inequality but the first sharp nontrivial benchmark in a broader program on Schatten-norm comparison between $A+B$5 and $A+B$6 (Zhang, 3 Jul 2025).

Historically, the first proof was obtained by Lin and Zhang in 2022. Their argument used inequalities for the angle between two matrices induced by the Frobenius inner product. Zhang’s 2025 paper does not strengthen the bound; its novelty is methodological. It replaces the angle-based approach with a short proof using only polar decomposition, trace identities, positivity, contractions, and matrix Cauchy–Schwarz. This suggests that the conjecture’s sharp constant is encoded in a comparatively elementary trace optimization rather than in a specifically geometric theory of matrix angles (Zhang, 3 Jul 2025).

3. Core mechanism of the short proof

The short proof begins from the polar decompositions

$A+B$7

with $A+B$8 and $A+B$9 partial isometries, and defines

$|A|+|B|$0

Because $|A|+|B|$1 arises from polar decomposition, it is a contraction: $|A|+|B|$2

The Frobenius norm of the sum is expanded as

$|A|+|B|$3

Using cyclicity of the trace, the mixed term is written in the form

$|A|+|B|$4

Hence

$|A|+|B|$5

The argument then depends on two auxiliary inequalities. The first is a weighted trace Cauchy–Schwarz estimate: for any $|A|+|B|$6 and $|A|+|B|$7,

$|A|+|B|$8

The second is the technical heart of the proof: if $|A|+|B|$9 and $p=2$0 is a contraction, then for any $p=2$1,

$p=2$2

Applying this with

$p=2$3

gives

$p=2$4

At this point the proof becomes a one-parameter optimization. Since

$p=2$5

one chooses $p=2$6 so that the two coefficients coincide. The optimal value is

$p=2$7

for which

$p=2$8

Therefore

$p=2$9

which is exactly the conjecture (Zhang, 3 Jul 2025).

4. Sharpness and equality issues

The inequality is optimal. Zhang’s paper recalls that Lin and Zhang had already shown sharpness using the canonical $c_p$0 example

$c_p$1

This construction shows that the constant

$c_p$2

cannot be improved (Zhang, 3 Jul 2025).

The 2025 short proof does not develop equality cases in detail. Its emphasis is instead on proof compression: the conjecture is reduced to a mixed-term estimate for $c_p$3 and a scalar choice of $c_p$4. A plausible implication is that the sharpness mechanism is already visible at the level of $c_p$5 configurations and does not require high-dimensional structure.

5. Later extensions and the many-summand problem

A later extension generalizes Lee’s two-summand problem to arbitrary finite families. For $c_p$6, the corresponding Frobenius inequality is

$c_p$7

and this constant is sharp. Equality is attained by an equiangular rank-one family (Tang et al., 19 Oct 2025).

In this formulation the original conjecture is precisely the case $c_p$8. The same paper introduces the optimal constant $c_p$9 for Schatten $\|A+B\|_p\le c_p\||A|+|B|\|_p$0-norms,

$\|A+B\|_p\le c_p\||A|+|B|\|_p$1

and proves the general upper bound

$\|A+B\|_p\le c_p\||A|+|B|\|_p$2

It also records the exact endpoint and special values

$\|A+B\|_p\le c_p\||A|+|B|\|_p$3

For $\|A+B\|_p\le c_p\||A|+|B|\|_p$4, this recovers

$\|A+B\|_p\le c_p\||A|+|B|\|_p$5

The same paper proposes a closed-form conjecture for $\|A+B\|_p\le c_p\||A|+|B|\|_p$6 for general $\|A+B\|_p\le c_p\||A|+|B|\|_p$7, motivated by the same equiangular rank-one extremizers that saturate the Frobenius bound (Tang et al., 19 Oct 2025).

6. Terminological scope and ambiguity

The label “Lee’s conjecture” is not uniform across the literature. In matrix analysis, it refers to the Frobenius-norm conjecture just described. In other areas, however, the phrase may denote unrelated statements or may not denote any named conjecture at all. Zhuang’s paper on knot cobordism develops inequalities for Lee’s perturbation of Khovanov homology, but it does not formulate or resolve a conjecture explicitly called “Lee’s Conjecture” (Zhuang, 2022). Abe and Tagami study a different conjectural question, namely whether a divisibility-based invariant $\|A+B\|_p\le c_p\||A|+|B|\|_p$8 coincides with Rasmussen’s $\|A+B\|_p\le c_p\||A|+|B|\|_p$9-invariant (Sano, 2018). A separate note on the Lee–Yang circle theorem is explicitly about a theorem, not a conjecture, and treats “Lee” as part of the Lee–Yang name rather than as the source of a standalone conjecture (Ghosh, 2012).

Within matrix analysis, by contrast, the terminology is stable: Lee’s conjecture means the sharp Frobenius inequality

$p$0

Its current significance lies in two facts. First, the exact constant is known and sharp. Second, the 2025 proof shows that the conjecture can be derived from a compact chain of standard tools—polar decomposition, contraction estimates, trace expansions, and matrix Cauchy–Schwarz—while the 2025 many-summand extension indicates that the same phenomenon belongs to a larger family of sharp inequalities for sums of absolute values (Zhang, 3 Jul 2025).