错位的线性代数 (Sergei Treil, 董耀择)（Z-Library）

# Reading Guide: *Linear Algebra Done Wrong* (错位的线性代数) ## 【One-Line Pitch】 A rigorous, concept-first linear algebra textbook that rebuilds the subject from linear maps rather than matrix computation—ideal for students who have taken a computational course but want to truly understand the structure, and for anyone preparing for functional analysis, quantum mechanics, or machine learning. ## 【Book Arc】 - **Opening (~0%–10%)**: The book opens with a manifesto against the traditional "cookbook" approach. Treil places vector spaces, bases, and linear transformations at the very front, arguing that a basis should be defined by unique representation (existence + uniqueness) rather than by linear independence alone. This reframing sets the tone: concepts before computation. - **Early (~10%–25%)**: Linear transformations are shown to be the true objects of study; matrices emerge as representations of these maps once a basis is chosen. The text then moves into systems of linear equations, row reduction, and the rank theorem—but always with the conceptual machinery of bases and dimension in the foreground. - **Early–Middle (~25%–40%)**: Determinants are introduced not as a computational trick but as a signed volume, derived from a small set of natural properties. The text then pivots to eigenvalues, eigenvectors, and diagonalization, including the complexification of real spaces and the key theorem that distinct eigenvalues guarantee diagonalizability. - **Middle (~40%–60%)**: Inner product spaces, orthogonality, and the Cauchy–Schwarz inequality take center stage. Orthonormal bases lead to the "baby version" of Fourier decomposition, and least-squares fitting is treated as a geometric projection problem rather than a calculus exercise. - **Late (~60%–90%)**: The book covers adjoint operators, unitary/orthogonal transformations, and the spectral theorem for normal operators. Singular value decomposition (SVD), polar decomposition, and the Moore–Penrose pseudoinverse are presented as natural consequences of the spectral theory developed earlier. - **Ending (~90%–100%)**: Jordan canonical form and the Jordan decomposition theorem close the book, with an appendix of solutions to selected exercises. The final chapters tie together generalized eigenspaces, nilpotent operators, and the structure theorem for linear maps. ## 【Key Takeaways】 - **Basis-first pedagogy** (Early): Defining a basis via unique representation (rather than starting with linear independence) makes the concept of dimension and coordinate systems feel inevitable rather than arbitrary. This reframing pays off throughout the book. - **Matrices are representations, not the objects themselves** (Early): A linear transformation is the fundamental object; its matrix depends on a choice of basis. Understanding this distinction clarifies why matrix multiplication is defined the way it is and why similarity transformations matter. - **Determinants as signed volumes** (Early–Middle): Instead of memorizing formulas, Treil derives the determinant from properties like multilinearity and alternating behavior. This makes the determinant's role in invertibility and volume scaling intuitive. - **Distinct eigenvalues ⇒ diagonalizability** (Middle): The proof that eigenvectors for distinct eigenvalues are linearly independent is a cornerstone result. It gives a simple sufficient condition for diagonalization and motivates the deeper question of what happens when eigenvalues repeat. - **Orthonormal bases make coordinates trivial** (Middle): In an orthonormal basis, coordinates are just inner products, and the "baby Fourier decomposition" formula v = Σ(v, vₖ)vₖ shows why orthogonal expansions are so powerful—no linear system solving required. - **Least squares is geometry, not calculus** (Middle): Fitting a line or plane to data is presented as orthogonal projection onto a subspace. This geometric view makes the normal equations and the role of the pseudoinverse transparent. - **The spectral theorem unifies the second half** (Late): Normal operators (those commuting with their adjoint) are exactly the ones that admit an orthonormal eigenbasis. This single result underpins SVD, polar decomposition, and the structure of unitary operators. - **Jordan form as the final structural answer** (Late): When diagonalization fails, the Jordan canonical form provides the closest possible "almost diagonal" representation. The book treats it as a natural extension of the spectral story, not as an isolated computational topic. ## 【Reading Tips】 - **Skim the first chapter if you've seen vector spaces before**, but do read the notes on why basis is defined via unique representation—this choice ripples through the entire book. - **Deep-read Chapter 3 (determinants)** even if you know the formulas. The derivation from properties is where the conceptual payoff lives, and it sets up the spectral theory later. - **The complexification section (Chapter 5, §8) can be skipped on first pass**—the book itself says so. Return to it only if you need the abstract construction for later chapters. - **Work the exercises marked with hints** (e.g., the Fibonacci diagonalization problem, the transpose operator on matrix spaces). These are not busywork; they train the structural thinking the book promises. - **Use the appendix of solutions strategically**: attempt each problem before peeking, and treat the solutions as a check on your reasoning, not a substitute for it. ## 【Coverage Limits】 This guide is based on stratified excerpts covering roughly the first half of the book in detail (through inner product spaces and least squares) plus the table of contents and symbol list for the later chapters. The treatment of SVD, spectral theorems, and Jordan form is synthesized from the book's structure and symbol references rather than from full excerpted proofs. ##