# Chapter ᚈ(T): The caboose *Teclam Áirme, the Mathematical Compendium — a copy for AI readers of https://www.teclamairme.com/#ch-T10, as of 2026-09-30. Links are relative to https://www.teclamairme.com/fortheaireaders/.* [Guide for AI readers](index.md) · [Its glossary cards](ch10-cards.md) · [Its references](ch10-refs.md) · [Previous chapter](ch09.md) · [Next chapter](ch11.md) *A footnote such as [r-10.3] is reference entry 10.3 and links to it. Section 10 of 28; chapters are named by their Ogham letter, never by number.* ᚈ – [Tinne](ch10-cards.md#g-og-tinne): associated with the holly tree, Section 10 ## Chapter ᚈ(T): The caboose > “I have in previous papers defined a ‘Matrix’ as a rectangular array of terms, out of which different systems of determinants may be engendered, as from the womb of a common parent.” > — James Joseph Sylvester · On the relation between the minor determinants of linearly equivalent quadratic functions, 1851 [r-10.21](ch10-refs.md#r-10.21) [r-10.26](ch10-refs.md#r-10.26) Formula: ([det [[1, 2], [2, 4]]](ch10-cards.md#g-detA) + [δ₁₂](ch10-cards.md#g-kronecker) + [(1, 2) · (2, −1)](ch10-cards.md#g-dot0) + [N²](ch10-cards.md#g-nilp) + [A² − 5A](ch10-cards.md#g-cayham))/([tr A](ch10-cards.md#g-trA) + [‖A‖_F](ch10-cards.md#g-frob) + [λ_max(A)](ch10-cards.md#g-lmax) + [rank A + null A](ch10-cards.md#g-ranknull)) Key concepts: Linear algebra. It includes symbols you may not have seen before, such as: [det](ch10-cards.md#g-detA), [tr](ch10-cards.md#g-trA), [‖A‖_F](ch10-cards.md#g-frob), [λ](ch10-cards.md#g-lmax), and [δ](ch10-cards.md#g-kronecker). Zero here is a lost dimension: a matrix that flattens the plane onto a line leaves no area behind. Here in the caboose, we have plenty of tools to guide the engineer, including this fabulous map on the wall. We have a whole landscape on the page and the country itself moving past the window. We can turn the map, change its scale, or imagine squeezing it until the entire sheet lands on one line. From the back of the train the whole route behind us has collapsed onto a single line of track, which is what the matrix on that map does to the plane. Linear algebra makes such transformations exact. The matrix chosen here sends a whole plane of possible starting points onto a line: many addresses acquire the same destination, a direction survives, and an area disappears. Linear algebra grew partly from solving simultaneous equations. Arrange their coefficients in a grid and [systematic elimination](ch10-cards.md#g-gausselim) becomes possible. Treat the grid itself as an object, as Cayley did in the nineteenth century, and another interpretation appears: the matrix describes a transformation. [r-10.1](ch10-refs.md#r-10.1) [r-10.2](ch10-refs.md#r-10.2) Field notes Almost everything here comes from one matrix, A = [[1, 2], [2, 4]]. The one exception is N = [[0, 1], [0, 0]], which appears once, squared. A sends (x, y) to (x + 2y, 2x + 4y). The second output is always twice the first, so every point lands on the line y = 2x: (1, 0) and (−1, 1) both land on (1, 2). Two different addresses on our map share a destination, their starting points cannot be recovered, and the matrix has no [inverse](ch10-cards.md#g-identity). [r-10.3](ch10-refs.md#r-10.3) One fact explains everything that follows. A/5 drops every point straight onto the line y = 2x, along the perpendicular: it is the [orthogonal projection](ch10-cards.md#g-orthproj) onto that line. A does the same, then stretches the result by five. So the direction along the line survives, the perpendicular one dies, and the area goes with it. Dropping a point that is already on the line leaves it where it is, so applying A twice only stretches by five once more: A² = 5A. And, as we will see, three readings of A below all come out as 5. #### Numerator [det [[1, 2], [2, 4]]](ch10-cards.md#g-detA) = 0; the headline fact about A: it squashes the plane onto a line. The [determinant](ch10-cards.md#g-detA) measures how much a transformation of the plane scales area. Here [det A](ch10-cards.md#g-detA) = 1 × 4 − 2 × 2 = 0, because the second row is twice the first. A square becomes a segment: the map still has an image, but its image has no area. [r-10.4](ch10-refs.md#r-10.4) [Cramer’s rule](ch10-cards.md#g-cramer), which solves a system of equations by dividing by this [determinant](ch10-cards.md#g-detA), would have to divide by zero, so here it fails. [r-10.14](ch10-refs.md#r-10.14) [δ₁₂](ch10-cards.md#g-kronecker) = 0; the Kronecker delta, the bookkeeping symbol behind [the identity matrix](ch10-cards.md#g-identity). δᵢⱼ is 1 when i = j and 0 otherwise, so δ₁₂ = 0 because its indices differ. Its values form [the identity matrix](ch10-cards.md#g-identity), the transformation that leaves every vector alone. This δ is not Feigenbaum’s. [r-10.5](ch10-refs.md#r-10.5) [(1, 2) · (2, −1)](ch10-cards.md#g-dot0) = 0; the two directions A treats most simply, and they are perpendicular. 1·2 + 2·(−1) = 0, and for two nonzero vectors a zero dot product means a right angle. A sends (1, 2) to five times itself and sends (2, −1) to zero: one direction survives and is stretched, the perpendicular one is lost. Directions like these two, which a map only stretches or shrinks, even to nothing, are its eigenvectors, and the multipliers, here 5 and 0, are its eigenvalues. A is a [real symmetric matrix](ch10-cards.md#g-spectralthm): its entries are real and it reads the same across its diagonal. In such a matrix, eigenvectors with different eigenvalues always meet at right angles. [r-10.8](ch10-refs.md#r-10.8) [N²](ch10-cards.md#g-nilp) = 0; a second matrix, chosen to show another way of losing: its image survives one application and dies on the next. N sends (x, y) to (y, 0). Apply it again and both coordinates are zero, so N is nonzero but [N²](ch10-cards.md#g-nilp) is the zero matrix. A matrix with a vanishing power is called nilpotent. [r-10.9](ch10-refs.md#r-10.9) [A² − 5A](ch10-cards.md#g-cayham) = 0; the Cayley–Hamilton theorem: every square matrix satisfies its own equation. Multiplying gives A² = [[5, 10], [10, 20]] = 5A, so [A² − 5A](ch10-cards.md#g-cayham) is the zero matrix. In general a square matrix satisfies its own [characteristic polynomial](ch10-cards.md#g-charpoly), the polynomial whose roots are its eigenvalues; for A, [that polynomial is λ² − 5λ](ch10-cards.md#g-charpoly), with roots 0 and 5. Put A in place of λ and you get our identity, A² − 5A = 0. [r-10.11](ch10-refs.md#r-10.11) #### Denominator Now ask what the same transformation keeps. Three readings of A all say 5, for a common reason: its only [nonzero eigenvalue](ch10-cards.md#g-lmax) is 5. [tr A](ch10-cards.md#g-trA) = 5; the trace, the simplest summary of a matrix. It adds the diagonal entries, 1 + 4, and it always equals the sum of the eigenvalues: 5 + 0. [r-10.6](ch10-refs.md#r-10.6) [‖A‖_F](ch10-cards.md#g-frob) = 5; the Frobenius norm, the size of all entries together. √(1² + 2² + 2² + 4²) = √25 = 5. For a symmetric matrix it is also the square root of the sum of the squared eigenvalues, √(25 + 0). [r-10.7](ch10-refs.md#r-10.7) [λ_max(A)](ch10-cards.md#g-lmax) = 5; the stretch along the surviving direction. It is the multiplier on the surviving direction: A stretches (1, 2) by 5. The other eigenvalue, on the lost direction (2, −1), is 0. [rank A + null A](ch10-cards.md#g-ranknull) = 1 + 1; the [rank–nullity theorem](ch10-cards.md#g-ranknull), counting what the map keeps and what it loses. *Rank* is the dimension of the output: one surviving line. *Nullity* is the dimension of the vectors sent to zero: one lost line. Their sum is [the input dimension](ch10-cards.md#g-vspace), 2, as the [rank–nullity theorem](ch10-cards.md#g-ranknull) promises. [r-10.10](ch10-refs.md#r-10.10) The only fraction is (0 + 0 + 0 + 0 + 0)/17: 5 + 5 + 5 + 1 + 1 = 17. **Fine print.** The [determinant](ch10-cards.md#g-detA), delta, and dot product are scalars; [N²](ch10-cards.md#g-nilp) and [A² − 5A](ch10-cards.md#g-cayham) are zero matrices. Each is evaluated in its own setting and its zero is then entered as the integer 0; the formula is not adding a scalar to a matrix. [r-10.19](ch10-refs.md#r-10.19) What this chapter adds to the formula: (0 + 0 + 0 + 0 + 0)/17 = 0 The map has shown us what a transformation can keep and what it can lose. Its remaining line still has direction, even though the area is gone. That’s our station coming into view—the museum is just beside the platform. The first gallery has a beautiful fountain that I think you’ll enjoy. A matrix moves the whole map at once, by the same stretch everywhere; the first gallery asks what changes from point to point, and how a rate adds up to a total. Let’s go watch numbers flow! Watch and read - [The determinant | Essence of linear algebra, chapter 6](https://www.youtube.com/watch?v=Ip3X9LOh2dk) 3Blue1Brown: the determinant as the factor by which area is scaled, so zero means collapse - [Eigenvectors and eigenvalues | Chapter 14, Essence of linear algebra](https://www.youtube.com/watch?v=PFDu9oVAE-g) 3Blue1Brown: what the eigenvalues in the denominator actually measure - [Determinant](https://en.wikipedia.org/wiki/Determinant) Wikipedia - [Eigenvalues and eigenvectors](https://en.wikipedia.org/wiki/Eigenvalues_and_eigenvectors) Wikipedia