AFTD Auto-Formalizing Theoretical Domains

Research Blogs

Positive Trace-Preserving Maps Violate the Universal Spectral Bound $\operatorname{Tr}\Phi \le d\,m(\Phi) + d^2 - d$ in Every Dimension $d \ge 3$

medium tier2026-11-01OP-7332 research-level results in one postreviewed by Claude Code中文版

Abstract

The question whether every positive trace-preserving linear map $\Phi: M_d(\mathbb{C}) \to M_d(\mathbb{C})$ in dimension $d \ge 3$ satisfies the universal spectral bound $\operatorname{Tr}\Phi \le d\,m(\Phi) + d^2 - d$, where $m(\Phi) = \min\{\operatorname{Re}\lambda : \lambda \in \sigma(\Phi)\}$ and $\operatorname{Tr}\Phi$ is the operator trace of $\Phi$, was raised in arXiv:2506.02145 and left open in Section 9 of arXiv:2610.37187. The bound is known for all 2-positive maps. We show, with proofs verified in Lean 4, that the bound fails in every dimension $d \ge 3$. In dimension $d = 3$ it fails for the positive trace-preserving map $\Phi(X) = K X K + \frac{3}{4} P X^T P$ on $M_3(\mathbb{C})$, with diagonal matrices $K = \operatorname{diag}(1/2, 1/2, 1)$ and $P = \operatorname{diag}(1, 1, 0)$. Being the sum of a completely positive map and a completely copositive map, $\Phi$ is positive, and the identity $K^2 + \frac{3}{4} P = I_3$ ensures that $\Phi$ is trace-preserving. It has operator trace $11/2$ and the eigenvalue $-1/2$ (eigenvector $E_{12} - E_{21}$). Consequently, the bound evaluates to $3\,m(\Phi) + 6 \le 9/2$, which is strictly less than the operator trace $11/2$. The same construction with $K = \operatorname{diag}(a, a, 1, \dots, 1)$, $P$ the projection onto the first two coordinates, the weight $1 - a^2$ in place of $3/4$ and $a = (d-2)/(d-1)$ violates the bound in dimension $d$ by exactly $2/(d-1)$. So 2-positivity cannot be weakened to positivity in the trace-preserving bound.

1. Introduction

In the study of matrix algebras and linear operator theory, the structure of positive linear maps on $M_d(\mathbb{C})$ represents a central subject of inquiry. A linear map $\Phi: M_d(\mathbb{C}) \to M_d(\mathbb{C})$ is called positive if it maps every positive semidefinite matrix to a positive semidefinite matrix. In quantum information theory and non-commutative analysis, such maps describe physical transformations, and additional properties such as trace preservation—where $\operatorname{tr}(\Phi(X)) = \operatorname{tr}(X)$ for all $X \in M_d(\mathbb{C})$—are standard requirements for quantum operations. Beyond simple positivity (often termed 1-positivity), higher degrees of positivity play a crucial role. A map $\Phi$ is called $k$-positive if the amplified map $\Phi \otimes \operatorname{id}_k: M_d(\mathbb{C}) \otimes M_k(\mathbb{C}) \to M_d(\mathbb{C}) \otimes M_k(\mathbb{C})$ is positive, and completely positive if it is $k$-positive for all integers $k \ge 1$.

Understanding the spectral properties of these maps provides deep geometric and algebraic information about the cones they preserve. For a linear operator $\Phi$ acting on $M_d(\mathbb{C})$, its spectrum $\sigma(\Phi)$ comprises all complex eigenvalues $\lambda$ of $\Phi$, and its operator trace $\operatorname{Tr}\Phi$ is defined as the trace of $\Phi$ viewed as an endomorphism of the $d^2$-dimensional complex vector space $M_d(\mathbb{C})$. A significant spectral parameter is the minimal real part among the eigenvalues of $\Phi$, defined by $m(\Phi) = \min\{\operatorname{Re}\lambda : \lambda \in \sigma(\Phi)\}$.

A fundamental question concerning such maps is whether their global spectral invariants satisfy universal geometric bounds. In recent work, universal spectral inequalities connecting the operator trace $\operatorname{Tr}\Phi$ to the minimal real part $m(\Phi)$ have been developed. In particular, it was shown that every 2-positive map satisfies the universal spectral bound: $$\operatorname{Tr}\Phi \le d\,m(\Phi) + d^2 - d.$$ The paper arXiv:2610.37187, titled 'Removing Trace Preservation from a Universal Spectral Bound for 2-Positive Maps', investigated this inequality and demonstrated that the condition of trace preservation can be removed from the universal spectral bound for all 2-positive maps.

In Section 9 of arXiv:2610.37187, within its concluding remarks, the paper highlighted two remaining open problems originally raised in arXiv:2506.02145. The first concerns whether complete positivity can be spectrally distinguished from 2-positivity. The second question asks whether positive trace-preserving maps in dimension $d \ge 3$ can violate the universal bound $\operatorname{Tr}\Phi \le d\,m(\Phi) + d^2 - d$. Specifically, does every positive trace-preserving map satisfy this bound, or does there exist some dimension $d \ge 3$ and a positive trace-preserving map $\Phi$ on $M_d(\mathbb{C})$ such that $\operatorname{Tr}\Phi > d\,m(\Phi) + d^2 - d$? Because 2-positivity is strictly stronger than positivity, this question probes whether the universal bound is an intrinsic consequence of trace-preserving positivity alone, or whether 2-positivity is genuinely required.

This post presents the resolution of this question, verified in the Lean 4 proof assistant. We establish that the universal spectral bound fails for positive trace-preserving maps in every dimension $d \ge 3$. The main theorem, not_universal_spectral_bound_positive_trace_preserving, proves the negation of the statement that every positive trace-preserving map across all dimensions $d \ge 3$ satisfies the inequality. The proof is based on an explicit three-dimensional counterexample, formalized as not_universal_spectral_bound_positive_tp_dim_three. The constructed map is decomposable—formed as the sum of a congruence map and a transposed projection map—and satisfies $\operatorname{Tr}\Phi = 11/2$ while its minimal real eigenvalue part satisfies $3\,m(\Phi) + 6 \le 9/2$. The theorem exists_positive_tp_violating_universal_spectral_bound extends the construction to every $d \ge 3$.

2. Preliminaries

We formalize linear maps on matrix algebras and their spectral invariants over the complex numbers $\mathbb{C}$. Throughout, $M_d(\mathbb{C})$ denotes the complex vector space of $d \times d$ matrices, represented in Lean as the matrix type over $d \times d$ indices with complex entries. The space of linear endomorphisms on $M_d(\mathbb{C})$ is denoted by $\operatorname{End}_{\mathbb{C}}(M_d(\mathbb{C}))$.

Definition 1 (is_positive_linear_map). Let $d \in \mathbb{N}$. A linear map $\Phi: M_d(\mathbb{C}) \to M_d(\mathbb{C})$ is positive if it maps every positive semidefinite matrix to a positive semidefinite matrix: $$\forall X \in M_d(\mathbb{C}), \quad X \ge 0 \implies \Phi(X) \ge 0.$$ In the formalization, this is defined by is_positive_linear_map d Φ, which states that for all $X$, $X$ being positive semidefinite implies $\Phi(X)$ is positive semidefinite within the scoped complex order.

*Remark on Definition 1.* The condition $X \ge 0$ is the standard weak positive semidefiniteness (that is, $X$ is Hermitian with non-negative eigenvalues), rather than strict positive definiteness. The definition specifies 1-positivity on $M_d(\mathbb{C})$; it does not require $k$-positivity for $k \ge 2$.

Definition 2 (is_trace_preserving_linear_map). Let $d \in \mathbb{N}$. A linear map $\Phi: M_d(\mathbb{C}) \to M_d(\mathbb{C})$ is trace-preserving if the matrix trace of $\Phi(X)$ is equal to the matrix trace of $X$ for all $X \in M_d(\mathbb{C})$: $$\forall X \in M_d(\mathbb{C}), \quad \operatorname{tr}(\Phi(X)) = \operatorname{tr}(X),$$ where $\operatorname{tr}(X) = \sum_{i=1}^d X_{ii}$ is the trace on $M_d(\mathbb{C})$. In the formalization, this is defined by is_trace_preserving_linear_map d Φ, asserting that the matrix trace of $\Phi(X)$ equals the matrix trace of $X$.

*Remark on Definition 2.* The trace $\operatorname{tr}$ in Definition 2 is the matrix trace mapping $M_d(\mathbb{C}) \to \mathbb{C}$. It should not be confused with the operator trace $\operatorname{Tr}\Phi$, which acts on the linear map $\Phi$ itself as an element of $\operatorname{End}_{\mathbb{C}}(M_d(\mathbb{C}))$.

Definition 3 (spectral_min_re). Let $d \in \mathbb{N}$. For a linear map $\Phi: M_d(\mathbb{C}) \to M_d(\mathbb{C})$, the minimal real part of an eigenvalue of $\Phi$, denoted $m(\Phi)$, is the infimum of the real parts of the complex eigenvalues of $\Phi$: $$m(\Phi) := \inf \{ \operatorname{Re}(\lambda) : \lambda \in \sigma(\Phi) \},$$ where $\sigma(\Phi)$ is the spectrum of $\Phi$ as a $\mathbb{C}$-linear operator. In Lean, this is defined by spectral_min_re d Φ as the infimum of the real parts of the spectrum.

*Remark on Definition 3.* Because $M_d(\mathbb{C})$ is finite-dimensional, whenever $d \ge 1$ the spectrum $\sigma(\Phi)$ is a non-empty finite subset of $\mathbb{C}$, and the infimum is attained as a minimum. The Lean definition uses the general set infimum on $\mathbb{R}$. For $d \ge 3$, the space $M_d(\mathbb{C})$ has dimension $d^2 \ge 9$, guaranteeing that the spectrum is non-empty.

Definition 4 (Operator Trace). Let $d \in \mathbb{N}$. For a linear operator $\Phi \in \operatorname{End}_{\mathbb{C}}(M_d(\mathbb{C}))$, the operator trace $\operatorname{Tr}\Phi$ is the trace of $\Phi$ as a linear endomorphism of the $d^2$-dimensional space $M_d(\mathbb{C})$.

*Remark on Definition 4.* In the inequality being evaluated, Lean takes the real part of this operator trace. When the operator trace is real (as occurs in our counterexample where all eigenvalues are real), this real part equals the operator trace itself.

3. Main result

The main theorem establishes that positive trace-preserving maps do not universally satisfy the spectral bound $\operatorname{Tr}\Phi \le d\,m(\Phi) + d^2 - d$ for dimensions $d \ge 3$.

Theorem (not_universal_spectral_bound_positive_trace_preserving). It is not the case that for every integer $d \ge 3$ and every positive trace-preserving linear map $\Phi: M_d(\mathbb{C}) \to M_d(\mathbb{C})$, the real part of the operator trace satisfies $$\operatorname{Re}(\operatorname{Tr}\Phi) \le d\,m(\Phi) + d^2 - d,$$ where $m(\Phi) = \operatorname{spectral\_min\_re} d \Phi$.

The Lean statement matches the English formulation by placing the negation in front of universal quantifiers over the dimension $d$ of type $\mathbb{N}$ restricted by the inequality $3 \le d$, the linear map $\Phi$ on $d \times d$ complex matrices, and hypotheses asserting that $\Phi$ satisfies is_positive_linear_map and is_trace_preserving_linear_map.

The refutation rests on an earlier result for dimension three, which constructs an explicit counterexample.

Theorem (not_universal_spectral_bound_positive_tp_dim_three). There exists a positive trace-preserving linear map $\Phi: M_3(\mathbb{C}) \to M_3(\mathbb{C})$ such that $$\operatorname{Tr}\Phi > 3\,m(\Phi) + 3^2 - 3.$$ Specifically, the linear map $\Phi(X) = K X K + \frac{3}{4} P X^T P$, with diagonal matrices $$K = \begin{pmatrix} 1/2 & 0 & 0 \\ 0 & 1/2 & 0 \\ 0 & 0 & 1 \end{pmatrix}, \qquad P = \begin{pmatrix} 1 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 0 \end{pmatrix},$$ is positive and trace-preserving, has operator trace $\operatorname{Tr}\Phi = 11/2$, and has $-1/2$ as an eigenvalue with eigenvector $E_{12} - E_{21}$, so that $3\,m(\Phi) + 6 \le 9/2 < 11/2$.

This dimension-three result supplies the explicit construction and eigenvalue computation behind the main theorem. The construction extends to every dimension.

Theorem (exists_positive_tp_violating_universal_spectral_bound). For every integer $d \ge 3$ there is a positive trace-preserving linear map $\Phi: M_d(\mathbb{C}) \to M_d(\mathbb{C})$ with $$d\,m(\Phi) + d^2 - d < \operatorname{Re}(\operatorname{Tr}\Phi).$$ Specifically, let $a = (d-2)/(d-1)$, let $K = \operatorname{diag}(a, a, 1, \dots, 1)$ and let $P$ be the diagonal projection onto the first two coordinates. Then $\Phi(X) = K X K + (1 - a^2)\, P X^T P$ has operator trace $(2a + d - 2)^2 + 2(1 - a^2)$ and the eigenvalue $2a^2 - 1$, and the trace exceeds $d(2a^2 - 1) + d^2 - d$ by $2/(d-1)$. For $d = 3$ this is the map above ($a = 1/2$).

Theorem not_universal_spectral_bound_positive_trace_preserving (Lean 4, verified)
theorem not_universal_spectral_bound_positive_trace_preserving :
    ¬ ∀ (d : ℕ), 3 ≤ d →
      ∀ (Φ : Matrix (Fin d) (Fin d) ℂ →ₗ[ℂ] Matrix (Fin d) (Fin d) ℂ),
      is_positive_linear_map d Φ → is_trace_preserving_linear_map d Φ →
      ((LinearMap.trace ℂ (Matrix (Fin d) (Fin d) ℂ)) Φ).re ≤
        (d : ℝ) * spectral_min_re d Φ + (d : ℝ) ^ 2 - (d : ℝ)

4. Proof

The proof idea is to show that the universal statement leads to an immediate arithmetic contradiction when instantiated at dimension $d = 3$. We construct an explicit candidate map $\Phi: M_3(\mathbb{C}) \to M_3(\mathbb{C})$ that is decomposable—the sum of a congruence transformation and a transposed projection—and verify step-by-step that it is positive, trace-preserving, has operator trace $11/2$, and admits an eigenvalue of $-1/2$.

We break the verification into four preliminary lemmas corresponding to the properties of $\Phi$, follow with the proof of not_universal_spectral_bound_positive_tp_dim_three, and conclude with the formal proof of not_universal_spectral_bound_positive_trace_preserving.

Let $K = \operatorname{diag}(1/2, 1/2, 1)$ and $P = \operatorname{diag}(1, 1, 0)$ in $M_3(\mathbb{C})$, and define $\Phi: M_3(\mathbb{C}) \to M_3(\mathbb{C})$ by $$\Phi(X) = K X K + \frac{3}{4} P X^T P.$$

Lemma 1. The map $\Phi$ satisfies is_positive_linear_map 3 Φ.

Proof of Lemma 1. Let $X \in M_3(\mathbb{C})$ be positive semidefinite ($X \ge 0$). We must show that $\Phi(X) \ge 0$. Since $\Phi(X) = K X K + \frac{3}{4} P X^T P$, it suffices to show that each summand is positive semidefinite. First, $K$ is a real diagonal matrix, hence Hermitian: $K^* = K$. For any vector $v \in \mathbb{C}^3$, the quadratic form evaluates to $$\langle v, K X K v \rangle = \langle K v, X (K v) \rangle \ge 0,$$ since $X \ge 0$. Thus $K X K \ge 0$. Second, the transpose map preserves positive semidefiniteness: if $X$ is positive semidefinite, then $X^* = X$ and $\langle w, X w \rangle \ge 0$ for all $w \in \mathbb{C}^3$. Taking complex conjugates shows that $\langle \overline{w}, X^T \overline{w} \rangle = \overline{\langle w, X w \rangle} \ge 0$, so $X^T \ge 0$. The matrix $P$ is also real diagonal, so $P^* = P$. For any vector $v \in \mathbb{C}^3$, $$\langle v, P X^T P v \rangle = \langle P v, X^T (P v) \rangle \ge 0.$$ Hence $P X^T P \ge 0$. Multiplying by the non-negative scalar $3/4$ gives $\frac{3}{4} P X^T P \ge 0$. The sum of two positive semidefinite matrices is positive semidefinite, so $\Phi(X) \ge 0$. This establishes that $\Phi$ is positive.

Lemma 2. The map $\Phi$ satisfies is_trace_preserving_linear_map 3 Φ.

Proof of Lemma 2. Let $X \in M_3(\mathbb{C})$. We compute the matrix trace $\operatorname{tr}(\Phi(X)).$ By linearity, $$\operatorname{tr}(\Phi(X)) = \operatorname{tr}(K X K) + \frac{3}{4} \operatorname{tr}(P X^T P).$$ Applying the cyclic property of the matrix trace, $\operatorname{tr}(A B) = \operatorname{tr}(B A)$, the first term satisfies $$\operatorname{tr}(K X K) = \operatorname{tr}(K^2 X).$$ Since $K = \operatorname{diag}(1/2, 1/2, 1)$, its square is $$K^2 = \operatorname{diag}((1/2)^2, (1/2)^2, 1^2) = \operatorname{diag}(1/4, 1/4, 1).$$ For the second term, using cyclicity and the fact that $P$ is a projection ($P^2 = P$): $$\operatorname{tr}(P X^T P) = \operatorname{tr}(P^2 X^T) = \operatorname{tr}(P X^T).$$ Since the matrix trace is invariant under transpose, and $P^T = P$, we have $$\operatorname{tr}(P X^T) = \operatorname{tr}((P X^T)^T) = \operatorname{tr}(X P) = \operatorname{tr}(P X).$$ Therefore, $$\operatorname{tr}(\Phi(X)) = \operatorname{tr}(K^2 X) + \frac{3}{4} \operatorname{tr}(P X) = \operatorname{tr}\left(\left(K^2 + \frac{3}{4} P\right) X\right).$$ Now compute the matrix sum $K^2 + \frac{3}{4} P$: $$K^2 + \frac{3}{4} P = \begin{pmatrix} 1/4 & 0 & 0 \\ 0 & 1/4 & 0 \\ 0 & 0 & 1 \end{pmatrix} + \begin{pmatrix} 3/4 & 0 & 0 \\ 0 & 3/4 & 0 \\ 0 & 0 & 0 \end{pmatrix} = \begin{pmatrix} 1/4 + 3/4 & 0 & 0 \\ 0 & 1/4 + 3/4 & 0 \\ 0 & 0 & 1 + 0 \end{pmatrix} = I_3.$$ Consequently, $$\operatorname{tr}(\Phi(X)) = \operatorname{tr}(I_3 X) = \operatorname{tr}(X).$$ Because this holds for all $X \in M_3(\mathbb{C})$, $\Phi$ is trace-preserving.

Lemma 3. The operator trace of $\Phi$ on $M_3(\mathbb{C})$ is $\operatorname{Tr}\Phi = 11/2$.

Proof of Lemma 3. The space $M_3(\mathbb{C})$ is 9-dimensional. Let $\{E_{ij} : 1 \le i, j \le 3\}$ be the standard basis of matrix units. The operator trace $\operatorname{Tr}\Phi$ equals the sum of the diagonal matrix elements of $\Phi$ with respect to this basis: $$\operatorname{Tr}\Phi = \sum_{i=1}^3 \sum_{j=1}^3 [ \Phi(E_{ij}) ]_{ij},$$ where $[ \Phi(E_{ij}) ]_{ij}$ denotes the coefficient of $E_{ij}$ in $\Phi(E_{ij})$. Writing $K = \sum_i k_i E_{ii}$ with $k_1 = 1/2, k_2 = 1/2, k_3 = 1$ and $P = E_{11} + E_{22}$, we have $$K E_{ij} K = k_i k_j E_{ij}, \qquad P E_{ij}^T P = P E_{ji} P.$$ The term $P E_{ji} P$ equals $E_{ji}$ if $i, j \in \{1, 2\}$, and equals $0$ whenever $i = 3$ or $j = 3$. Evaluating $\Phi(E_{ij})$ on each of the 9 basis elements gives:

  1. For $E_{11}$: $\Phi(E_{11}) = (1/4) E_{11} + (3/4) E_{11} = 1 \cdot E_{11}$, contributing 1.
  2. For $E_{22}$: $\Phi(E_{22}) = (1/4) E_{22} + (3/4) E_{22} = 1 \cdot E_{22}$, contributing 1.
  3. For $E_{33}$: $\Phi(E_{33}) = (1) E_{33} + 0 = 1 \cdot E_{33}$, contributing 1.
  4. For $E_{12}$: $\Phi(E_{12}) = (1/4) E_{12} + (3/4) E_{21}$, contributing $1/4$.
  5. For $E_{21}$: $\Phi(E_{21}) = (1/4) E_{21} + (3/4) E_{12}$, contributing $1/4$.
  6. For $E_{13}$: $\Phi(E_{13}) = (1/2) E_{13} + 0 = (1/2) E_{13}$, contributing $1/2$.
  7. For $E_{31}$: $\Phi(E_{31}) = (1/2) E_{31} + 0 = (1/2) E_{31}$, contributing $1/2$.
  8. For $E_{23}$: $\Phi(E_{23}) = (1/2) E_{23} + 0 = (1/2) E_{23}$, contributing $1/2$.
  9. For $E_{32}$: $\Phi(E_{32}) = (1/2) E_{32} + 0 = (1/2) E_{32}$, contributing $1/2$.

Summing these 9 diagonal coefficients gives: $$\operatorname{Tr}\Phi = 1 + 1 + 1 + \frac{1}{4} + \frac{1}{4} + \frac{1}{2} + \frac{1}{2} + \frac{1}{2} + \frac{1}{2} = 3 + \frac{1}{2} + 2 = \frac{11}{2}.$$

Lemma 4. The operator $\Phi$ has eigenvalue $-1/2$ with eigenvector $E_{12} - E_{21}$, and consequently $m(\Phi) = \operatorname{spectral\_min\_re} 3 \Phi \le -1/2$.

Proof of Lemma 4. Consider the non-zero skew-symmetric matrix $V = E_{12} - E_{21} \in M_3(\mathbb{C})$. From the evaluations in Lemma 3: $$\Phi(E_{12}) = \frac{1}{4} E_{12} + \frac{3}{4} E_{21}, \qquad \Phi(E_{21}) = \frac{3}{4} E_{12} + \frac{1}{4} E_{21}.$$ Subtracting these two expressions: $$\Phi(V) = \Phi(E_{12}) - \Phi(E_{21}) = \left(\frac{1}{4} - \frac{3}{4}\right) E_{12} + \left(\frac{3}{4} - \frac{1}{4}\right) E_{21} = -\frac{1}{2} E_{12} + \frac{1}{2} E_{21} = -\frac{1}{2} V.$$ Because $V \ne 0$, $-1/2$ is an eigenvalue of $\Phi$, meaning $-1/2 \in \sigma(\Phi)$. By definition, $\operatorname{spectral\_min\_re} 3 \Phi = \inf \{ \operatorname{Re}(\lambda) : \lambda \in \sigma(\Phi) \}$. Since $-1/2 \in \sigma(\Phi)$ and $\operatorname{Re}(-1/2) = -1/2$, the infimum is bounded above by $-1/2$: $$m(\Phi) \le -\frac{1}{2}.$$

Proof of Theorem not_universal_spectral_bound_positive_tp_dim_three. The map $\Phi$ satisfies is_positive_linear_map 3 Φ by Lemma 1, and is_trace_preserving_linear_map 3 Φ by Lemma 2. By Lemma 3, $\operatorname{Tr}\Phi = 11/2$. By Lemma 4, $m(\Phi) \le -1/2$. Evaluating the expression $3\,m(\Phi) + 3^2 - 3$: $$3\,m(\Phi) + 6 \le 3\left(-\frac{1}{2}\right) + 6 = -\frac{3}{2} + 6 = \frac{9}{2}.$$ Because $\operatorname{Tr}\Phi = 11/2 = 5.5$ and $9/2 = 4.5$, we have $$\operatorname{Tr}\Phi = \frac{11}{2} > \frac{9}{2} \ge 3\,m(\Phi) + 6.$$ This proves not_universal_spectral_bound_positive_tp_dim_three.

Proof of Main Theorem not_universal_spectral_bound_positive_trace_preserving. Assume for contradiction the universal statement: $$h : \forall (d : \mathbb{N}), 3 \le d \to \forall (\Phi : M_d(\mathbb{C}) \to_{\mathbb{C}} M_d(\mathbb{C})),$$ $$\operatorname{is\_positive\_linear\_map} d \Phi \to \operatorname{is\_trace\_preserving\_linear\_map} d \Phi \to \operatorname{Re}(\operatorname{Tr}\Phi) \le (d : \mathbb{R}) \cdot m(\Phi) + d^2 - d.$$ It suffices to prove the universal statement in dimension three, which not_universal_spectral_bound_positive_tp_dim_three refutes. So let $\Phi$ be a positive trace-preserving map on $M_3(\mathbb{C})$; instantiating $h$ at $d = 3$ (with $3 \le 3$) gives $$\operatorname{Re}(\operatorname{Tr}\Phi) \le (3 : \mathbb{R}) \cdot m(\Phi) + 3^2 - 3.$$ Direct numerical evaluation gives $3^2 - 3 = 6$, simplifying the inequality to $\operatorname{Re}(\operatorname{Tr}\Phi) \le 3\,m(\Phi) + 6$. which is the dimension-three statement, and the dimension-three refutation gives the contradiction.

Proof of exists_positive_tp_violating_universal_spectral_bound. The four lemmas carry over. Positivity: $X \mapsto K X K$ and $X \mapsto P X^T P$ are congruences of $X$ and of $X^T$ by real diagonal matrices, and $1 - a^2 \ge 0$ because $0 \le a \le 1$. Trace preservation: $\operatorname{tr}\Phi(X) = \sum_i (k_i^2 + (1 - a^2) p_i^2) X_{ii}$ and $k_i^2 + (1 - a^2) p_i^2 = 1$ for every $i$. Operator trace: in the basis of matrix units, $[\Phi(E_{ij})]_{ij} = k_i k_j + (1 - a^2) p_i p_j [i = j]$, so $\operatorname{Tr}\Phi = (\sum_i k_i)^2 + (1 - a^2) \sum_i p_i^2 = (2a + d - 2)^2 + 2(1 - a^2)$. Eigenvalue: $\Phi(E_{12} - E_{21}) = (a^2 - (1 - a^2))(E_{12} - E_{21})$, so $m(\Phi) \le 2a^2 - 1$; the spectrum is finite, so the set of real parts is bounded below and the infimum is at most this value. Finally, for any $a$, $$(2a + d - 2)^2 + 2(1 - a^2) - \bigl(d(2a^2 - 1) + d^2 - d\bigr) = -2(d-1)a^2 + 4(d-2)a - 2(d-3),$$ and at $a = (d-2)/(d-1)$ this equals $2/(d-1) > 0$; in Lean the identity is checked by clearing the denominator.

5. Discussion

The result answers the question raised in arXiv:2506.02145 and highlighted in Section 9 of arXiv:2610.37187: positive trace-preserving linear maps in every dimension $d \ge 3$ can violate the universal spectral bound $\operatorname{Tr}\Phi \le d\,m(\Phi) + d^2 - d$.

The central significance of this refutation lies in delineating the boundary between 1-positivity and 2-positivity. The paper arXiv:2610.37187 established that every 2-positive map satisfies $\operatorname{Tr}\Phi \le d\,m(\Phi) + d^2 - d$, and moreover that trace preservation is entirely unnecessary for that bound to hold. Prior to the present result, it remained an open possibility that trace preservation might compensate for the lack of 2-positivity, allowing the bound to hold for all positive trace-preserving maps. The counterexample $\Phi(X) = K X K + \frac{3}{4} P X^T P$ shows that trace preservation does not suffice: the universal bound fails even for decomposable positive trace-preserving maps. Consequently, 2-positivity cannot be weakened to positivity in that theorem, in any dimension $d \ge 3$.

Within this one-parameter family (the weight $1 - a^2$ is forced by trace preservation), for fixed $d$ the excess $-2(d-1)a^2 + 4(d-2)a - 2(d-3)$ is maximal at $a = (d-2)/(d-1)$. The excess $2/(d-1)$ tends to $0$ as $d$ grows.

Several questions remain open:

1. Optimal bound for positive trace-preserving maps: the theorems show that $d\,m(\Phi) + d^2 - d$ is not an upper bound for $\operatorname{Tr}\Phi$ over positive trace-preserving maps, but they do not determine the largest possible excess $\operatorname{Tr}\Phi - (d\,m(\Phi) + d^2 - d)$ in dimension $d$, for decomposable maps or for positive maps in general.

2. Spectral distinction between complete positivity and 2-positivity: As noted in Section 9 of arXiv:2610.37187, the other question raised in arXiv:2506.02145 asks whether complete positivity can be spectrally distinguished from 2-positivity. The present result separates positivity from 2-positivity through this bound; it says nothing about 2-positivity versus complete positivity.

Why it is research-level

  • Answers the question left open in arXiv:2610.37187 (Sec. 9) and raised in arXiv:2506.02145: positive trace-preserving maps in dimension three can violate the universal spectral bound Tr Φ ≤ d m(Φ) + d² − d that holds for all 2-positive ones; an explicit decomposable map with trace 11/2 and eigenvalue −1/2 exceeds the bound 9/2
  • Answers the question of arXiv:2610.37187 (Sec. 9, raised in arXiv:2506.02145) in every dimension: for each d ≥ 3 an explicit decomposable positive trace-preserving map violates the bound Tr Φ ≤ d m(Φ) + d² − d, which holds for all 2-positive trace-preserving maps, by 2/(d − 1)

Lean declarations

Each one elaborated against Mathlib, and #print axioms returned nothing outside Classical.choice · Quot.sound · propext.

References

  1. arXiv:2610.37187 — Removing Trace Preservation from a Universal Spectral Bound for 2-Positive Maps
  2. arXiv:2506.02145

Drafted by gemini-3.8-flash from the knowledgebase, checked against it, and reviewed by Claude Code. Only the Lean is verified; the prose is a reading of it.