8
Part Zero · Foundations

Linear algebra I: vectors, matrices, eigenvalues

When a physicist measures the energy of an atom, the spin of an electron, or which way a photon is polarized, the answer is always one of a short list of allowed values. Those values are not arbitrary. Each one is an eigenvalue of a matrix, and the state the system lands in is the matching eigenvector.

So before quantum mechanics we need to see what matrices really are, and why a few special directions survive them untouched.

A matrix is a way of moving space

Forget the grid of numbers for a moment. A 2×2 matrix tells you where the two basic arrows go: the first column is where the arrow along xx lands, the second is where the arrow along yy lands. Everything else follows, because the matrix moves sums to sums and doubles to doubles.

Set the matrix to (abcd)\begin{pmatrix} a & b \\ c & d\end{pmatrix} with aa = 2.0, bb = 1.0, cc = 1.0, dd = 2.0. The plate shows the faint original grid and the amber grid it is carried to.

The determinant is an area

The unit square, shaded on the plate, is carried to a parallelogram. Its area is the determinant, ad−bcad - bc. A determinant of 3 means every region grows three times bigger; a negative one means space was flipped over; zero means space was flattened onto a line and cannot be un-flattened. That last case is why a matrix with zero determinant has no inverse.

Arrows that keep their direction

Most arrows are turned by the matrix. A few special ones are only stretched or shrunk, staying on their own line. Those are the eigenvectors, and the stretch factor is the eigenvalue:

A v=λ vA\,\mathbf v = \lambda\,\mathbf v

The plate draws them as amber lines. With the starting matrix (2112)\begin{pmatrix}2&1\\1&2\end{pmatrix}, the diagonal direction is stretched by 3 and the anti-diagonal by 1. Change the numbers and watch the special lines move. Make bb and cc very different and they can disappear altogether: a rotation turns every arrow, so it has no real eigenvectors.

DeeperAv=λvA\mathbf v = \lambda \mathbf v has a non-zero solution only if A−λIA - \lambda I flattens space, so det⁡(A−λI)=0\det(A-\lambda I) = 0. For 2×2 that is λ2−(a+d)λ+(ad−bc)=0\lambda^2 - (a+d)\lambda + (ad-bc) = 0: the eigenvalues add up to the trace and multiply to the determinant.

Diagonalizing: choosing the natural axes

If you describe space using the eigenvectors as your axes, the matrix does nothing but stretch along each axis. In that frame it is diagonal, with the eigenvalues down the diagonal: A=PDP−1A = P D P^{-1}. Every hard calculation with AA, such as raising it to the hundredth power, becomes easy: raise each eigenvalue to the hundredth power instead.

Physics is full of this. Two equal masses joined by three equal springs move in a tangle, but in the right combinations they move simply: both swinging together at frequency ω0\omega_0, or in opposition at 3 ω0\sqrt3\,\omega_0. Those two motions are the eigenvectors of the system’s spring matrix, and the squared frequencies are its eigenvalues. , , or .

Finding the biggest eigenvector by brute force

Apply a matrix to almost any starting arrow over and over, rescaling each time, and the arrow swings towards the eigenvector with the largest eigenvalue. That is power iteration, and it is how a search engine once ranked the whole web.

Here it finds where a simple weather model settles. If today is sunny, tomorrow is sunny with probability 0.9; if rainy, rainy with probability 0.5. at a time and the probabilities settle at 5/65/6 sunny and 1/61/6 rainy, the eigenvector with eigenvalue 1, whatever the weather was on day one.

A preview of the qubit

Three small matrices will follow us through the whole book, the Pauli matrices:

X=(0110),Y=(0−ii0),Z=(100−1)X = \begin{pmatrix}0&1\\1&0\end{pmatrix},\quad Y = \begin{pmatrix}0&-i\\i&0\end{pmatrix},\quad Z = \begin{pmatrix}1&0\\0&-1\end{pmatrix}

Each has eigenvalues +1+1 and −1-1. Measure an electron’s spin along any axis and you get one of exactly two answers, “up” or “down”, because those are the eigenvalues of the matching matrix. That is the opening claim of this chapter, made concrete. Established

What you now have

Matrices as movements of space, determinants as area, eigenvectors as the directions a matrix leaves alone, diagonalization as choosing the natural axes, and a first look at the matrices of spin.

· · ·

Three questions to carry with you.

What does a zero determinant mean geometrically?tap
The matrix squashes space onto a line or a point, so areas become zero and the move cannot be undone.
Why does a 2D rotation have no real eigenvectors?tap
It turns every arrow, so none stays on its own line. Its eigenvalues are complex: e^(±iθ).
What do the eigenvalues of a matrix add up to?tap
The trace, the sum of the diagonal entries. They multiply to the determinant.

And one to work out. The matrix (4123)\begin{pmatrix}4&1\\2&3\end{pmatrix} has eigenvalues 5 and one other. What is the other?

λ =

Sources: G. Strang, Introduction to Linear Algebra. S. Axler, Linear Algebra Done Right. Checked by code: Pauli eigenvalues ±1; Markov stationary state (5/6, 1/6); coupled-mass frequencies ω₀ and √3 ω₀.

Next · Chapter 9
Linear algebra II: inner products and Hilbert spaces

A function is a vector with infinitely many parts. That idea is half of quantum mechanics.

Turn the page