Matrices: Tables That Compute

The spreadsheet answered before she finished asking. My daughter needed to know which of two phone plans was cheaper over a year: different monthly fees, different per-gigabyte rates, and a usage number that kept changing. Instead of recomputing every scenario, we laid the numbers in a grid, plans as rows and cost parts as columns, and multiplied the grid by each usage in one move. Change the usage, repeat the same motion, read the new totals. The table had stopped being a record and started being a machine, and the machine answered faster than the question could be rephrased. Two plans, one grid, and the grid answered both before the question finished changing.

A matrix is exactly that: a table that computes. Rows and columns hold numbers with addresses, addition works entry by entry, and multiplication mixes rows with columns on purpose, which turns whole systems of equations into single operations. Five sections below build the machine from its parts: reading the grid, adding tables, the peculiar multiplication, the determinant shortcut, and the one habit that keeps it oiled.

Tables with addresses

Every entry in a matrix lives at a street address: row first, column second. The number in row 2, column 3 is a-two-three, and the address system is the whole point, because position carries meaning the way seats carry meaning in a classroom. Two tables with the same numbers in different seats are different matrices, the way two classrooms with swapped students are different rooms.

Size matters first: a 2-by-3 grid behaves differently from a 3-by-2 one, and operations demand matching dimensions the way plugs demand matching sockets. Adding requires identical shapes, entry by entry, top-left with top-left. Our counting guide builds tables that list possibilities; matrices build tables that then compute with them, and the address system is what upgrades a ledger into a machine. Accountants knew tables could hold numbers; mathematicians discovered tables could move them.

Read one slowly before computing anything: top row is the first equation's ingredients, left column is the first unknown's appearances, and every cell answers what-stands-where. Students who read tables fluently before operating on them make half the errors, because most matrix mistakes are seating mistakes, numbers in the right arithmetic but the wrong address.

Adding tables, scaling tables

Addition is the honest operation: same shape, add matching entries, done. Two monthly budgets as two rows add into a combined row, and the meaning survives because every address kept its job. Subtraction works identically, and scaling multiplies every entry by one number, doubling a whole table the way you'd double a recipe. If matrices only did this, they would be tidy spreadsheets and nothing more.

The phone-plan grid shows why even this much matters. Monthly fee plus per-gig rate times usage, for two plans side by side, is one addition of two scaled rows, and changing the usage replays the same two motions. Repetition with variation is the signature of matrix thinking: set up the structure once, then pour new numbers through it forever. The setup is the work; the reruns are free. Set the grid up in September and every bill through June answers itself on request.

Identity tables are the quiet heroes: ones down the diagonal, zeros elsewhere, the grid equivalent of multiplying by one. Multiplying by the identity changes nothing, adding zero changes nothing, and these do-nothing elements anchor every fancier operation the way zero and one anchor arithmetic. Learn their look cold; they appear as checkpoints in every longer computation.

Multiplying means mixing on purpose

Matrix multiplication ignores every instinct addition built: rows of the first meet columns of the second, multiply pairwise, and add. A 2-by-3 times a 3-by-2 yields a 2-by-2, because the middle dimensions must match and then vanish from the answer's shape. The rule looks arbitrary until you see what it computes: each answer cell blends one whole row with one whole column, which is exactly how a total price blends quantities with rates.

Work one by hand, slowly, because the fingers must learn the pattern. Top-left answer equals row one dot column one: first times first, plus second times second, plus third times third. Three products, one sum, one cell, and the same ritual fills every other cell. Slow is smooth and smooth is fast: careful first passes beat three hasty retries. Order matters here, unlike everything before it: A times B is generally not B times A, and the asymmetry is the point, since mixing in one direction means something different from mixing in reverse.

This is where systems of equations disappear into single lines. Two equations in two unknowns become one matrix times one mystery column equals one answer column, and the whole apparatus of elimination hides inside the multiplication (Wikipedia, Matrix). Our systems guide teaches the line-by-line version. Matrices are that version compressed to a gesture, and the gesture scales to a hundred equations where hand methods drown. What took a page now takes a line, and the line scales to any size the problem demands.

One number that decides everything

The determinant squeezes a square matrix down to a single number with outsized authority. For a 2-by-2, it is top-left times bottom-right minus top-right times bottom-left, ten seconds of arithmetic that answers the biggest question in advance: nonzero means the system has exactly one solution, zero means trouble, parallel lines or stacked ones. Compute it first, and half the dead ends announce themselves before you enter.

Geometrically it measures stretch: a transformation matrix with determinant 2 doubles every area it touches, negative determinant flips orientation like a mirror, and zero determinant flattens the plane into a line, destroying information. Our coordinate plane guide provides the map these stretches redraw. The number is small. Its testimony is total. Compute it before solving anything, the way you check the weather before dressing.

Game designers live by it: every rotation, reflection, and scaling of a sprite is a small matrix applied to coordinates, thousands of times per second, and the determinant constantly reports whether the picture stays honest. The same table your child multiplies by hand runs every game they play, which is a respectable afterlife for homework. A game engine multiplies thousands of these tables per frame without blinking, which is why the math homework and the hobby share one engine.

Undoing the mix

Undoing a table is harder than building one. Just as subtraction undoes addition, every square matrix with a nonzero determinant has an inverse that unmixes exactly what it mixed: multiply by the matrix, then by its inverse, and you land back where you started. For two-by-two grids the recipe fits in one line: swap the diagonal, negate the off-diagonal, divide everything by the determinant. Run it on a doubling grid and watch halves appear; the machine runs backward as willingly as forward.

Singular tables, the zero-determinant kind, have no undo, and the failure is informative: parallel lines never meet, stacked equations repeat each other, and no inverse can manufacture an answer from a question that never had one. Division by zero wears a grid costume here, and recognizing the costume saves the hours that blind computing would bill. Check the determinant before inverting, always, because the number tells you in advance whether the trip back exists.

Solving systems is where inverses earn their keep. Write the coefficients as a matrix, the unknowns as a column, the answers as another column, multiply both sides by the inverse, and the mystery column stands revealed. Classrooms teach elimination because hands learn it; professionals use the inverse because machines do. Same answers, different century, and the century that computes wins.

The one habit: Friday table

Every Friday, translate one weekly computation into a table: the grocery list as prices times quantities, the chore schedule as people times tasks, the savings plan as months times deposits. Write the grid, label rows and columns, compute one cell by hand, then admire the machine you built. My daughter's fridge holds a points table for our card games now, and disputes end where the arithmetic says they end. Fifty-two tables a year is fifty-two problems reorganized from lists into grids, and grids that compute beat lists that merely record. A year of tables is a year of seeing structure first and numbers second, which is the whole game.