Chapter 60: Linear Preserver Problems (LPP)¶

Prerequisites: Linear Transformations (Ch05) · Determinants (Ch03) · Matrix Groups (Ch55) · Basics of Operator Theory

Chapter Outline: Motivation for Linear Preserver Problems → Preserving the Determinant (Frobenius's Theorem) → Preserving Rank (Dieudonné's Theorem) → Preserving the Spectrum and Eigenvalues → Preserving Invertibility and Singularity → Preserving Norms (Isometries) → Preserving Positive Definiteness and Commutativity → Applications: Operator Evolution in Quantum Information and Symmetry Analysis in Matrix Theory

Extension: Linear Preserver Problems investigate "which transformations leave the soul (essential properties) of a matrix intact"; they are the ultimate tool for understanding symmetries in matrix spaces, revealing the deepest rigid structures in operator algebras.

In studying linear operators $L: M_n \to M_n$ on matrix spaces, a natural question arises: if $L$ preserves a specific matrix property (such as the determinant, rank, or eigenvalues), what form must $L$ take? This is known as the Linear Preserver Problem (LPP). Such studies not only reveal the algebraic rigidity of matrix spaces but also provide the theoretical foundation for symmetry transformations in quantum mechanics and functional analysis.

60.1 Preserving the Determinant: Frobenius's Theorem¶

Theorem 60.1 (Frobenius's Theorem)

Let $L: M_n(\mathbb{C}) \to M_n(\mathbb{C})$ be a linear transformation. If $\det(L(A)) = \det(A)$ for all matrices $A$, then there exist non-singular matrices $M, N$ with $\det(MN) = 1$ such that $L$ has one of the following two forms: 1. Equivalent Form: $L(A) = MAN$ 2. Transpose Equivalent Form: $L(A) = MA^T N$ Significance: This shows that the determinant is not just a value; it strictly constrains the possible forms of transformations on the matrix space.

60.2 Preserving Rank and Singularity¶

Theorem 60.2 (Dieudonné's Theorem)

A linear transformation $L$ preserves the set of all rank-1 matrices (or preserves the set of singular matrices) if and only if it has the form $L(A) = MAN$ or $L(A) = MA^T N$ for non-singular $M, N$. Insight: Rank is the most fundamental topological feature of a matrix; transformations that preserve it are almost inevitably a reorganization of the coordinate system.

60.3 Preserving the Spectrum and Eigenvalues¶

Theorem 60.3 (Preserving the Spectrum)

If $L$ preserves the spectrum of a matrix (all eigenvalues and their multiplicities), then $L$ must be a similarity transformation or its transpose: $$L(A) = P^{-1} A P \quad \text{or} \quad L(A) = P^{-1} A^T P$$ This proves that similarity transformations are the only "legitimate" symmetric operations in eigenvalue theory.

60.4 Preserving Positive Definiteness and Norms¶

Other Preserved Properties

Isometries: Transformations preserving the Frobenius norm must be of the form $L(A) = UAV$ where $U, V$ are unitary.
Positive Definiteness: Transformations preserving the set of symmetric positive definite matrices must have the form $L(A) = P A P^T$ for some non-singular $P$.

Exercises¶

Solution

$\det(P^{-1} A P) = \det(P^{-1})\det(A)\det(P) = \det(A)$.

Solution

$L(A) = MAN$ where $M, N$ are not inverses of each other. For example, $M=2I, N=0.5I$. While $\det(MN)=1$ ensures the determinant is preserved, the individual eigenvalues will change because the transformation is not a similarity transform.

Solution

Not necessarily. It might scale the determinant by a constant factor.

Solution

The form is very broad and not limited to $MAN$. Any linear transformation satisfying $\operatorname{tr}(L(E_{ij})) = \delta_{ij}$ works.

Solution

If a linear transformation maps the set of non-singular matrices to itself, by dimensionality and topological continuity, it must also map the boundary (singular matrices) to itself.

Solution

Yes. Since $\det(A-\lambda I) = \det((A-\lambda I)^T) = \det(A^T - \lambda I)$, the eigenvalues are identical.

Solution

$\|UAV\|_F^2 = \operatorname{tr}((UAV)^*(UAV)) = \operatorname{tr}(V^* A^* U^* U A V) = \operatorname{tr}(V^* A^* A V) = \operatorname{tr}(A^* A) = \|A\|_F^2$.

Solution

Because a physical quantum channel must map a valid density matrix (positive definite with trace 1) to another valid density matrix.

Solution

Usually a combination of scalar multiplication and a similarity transformation.

Solution

Chapter Summary¶

Linear preserver problems establish the algebraic boundaries of matrix properties:

****: Theorems by Frobenius and Dieudonné prove that the core attributes of a matrix (determinant, rank) are extremely "solid," and any linear attempt to preserve them eventually returns to fundamental coordinate rotations and transpositions.

****: From preserving the spectrum (strong constraint) to preserving the trace (weak constraint), LPP demonstrates how different mathematical features exert varying degrees of compression on the transformation space.

****: By investigating preservation properties, we can redefine what constitutes "physically consistent" evolution from a purely algebraic perspective, providing rigorous criteria for modern quantum operator theory and matrix manifold analysis.