Linear Algebra Through Pictures Before Symbols
A matrix is a transformation of space before it is a table of numbers. Seeing that first makes the algebra easier to trust.
On this page
A matrix is easiest to understand as a motion before it is understood as a table of numbers. Draw a grid, choose a rule that moves its points, and the rule becomes something you can see. Once you can predict the motion, the symbols stop looking arbitrary, and the algebra can be checked against a picture rather than taken on trust.
Vectors as arrows first
Draw a vector as an arrow from the origin. The pair of numbers describing it, such as 3 and 1, says how far to travel along each axis: three steps across and one up. Adding two vectors means placing one arrow at the tip of the other and drawing the arrow from the start to the end. Scaling a vector by two doubles its length and keeps its direction.
The numbers attached to a vector depend on the axes chosen to describe it. Change the axes and the same arrow receives different numbers, though the arrow has not moved. That distinction between the thing and its coordinates is what makes change of basis, and many later methods, make sense.
These pictures carry most of the subject. Later ideas, from projection to similarity, are statements about arrows: how long they are, which way they point, and how they relate. If a sketch of two arrows does not make a claim obvious, a symbolic argument probably will not either.
A matrix is a map of the basis
A two-by-two matrix is easiest to read by its columns. Each column is where one of the two unit arrows lands after the map. The arrow along the first axis goes to the first column, and the arrow along the second axis goes to the second column. Every other point follows, because any arrow is a combination of the unit arrows.
Take columns (2, 0) and (0, 1). The first unit arrow doubles in length and the second stays put, so the whole plane stretches horizontally by a factor of two. Now take columns (0, 1) and (-1, 0). The first arrow moves straight up and the second moves to the left, which is a quarter turn. Both results are visible before any arithmetic, and the arithmetic should then agree with the sketch.
Maps compose. Applying one matrix and then another gives the same result as applying a single matrix built from both, and the picture makes the order visible. Stretch first and then rotate, and the outcome differs from rotating first and then stretching. Drawing both orders on the same grid is a quick way to see why the order of matrix multiplication matters.
The determinant measures change in area
The determinant of a two-by-two matrix tells how much the map scales areas. Apply the map to a unit square and measure the parallelogram that results. A value of two means areas double. A value of one half means they halve. A negative value means the map flips orientation, as a mirror does. A value of zero means the map squashes the plane onto a line, and information is lost.
That last case matters in practice. A map that collapses a dimension cannot be undone, because different inputs now produce the same output. A zero determinant means a system of equations has either no solution or infinitely many, depending on the right-hand side. A sketch of the flattened plane makes that result feel like a fact rather than a rule to memorize.
Projections and similarity
Projecting a vector onto a direction gives the part of the vector that lies along that direction, like the shadow an arrow casts on a line. The dot product of two vectors measures how far they point the same way. Aligned vectors give a large value, and perpendicular vectors give zero.
This is why similarity search works. A text or an image can be represented as a vector, and items pointing in similar directions are judged alike. The trade-off is that a projection discards whatever lies perpendicular to the chosen direction. Dimension reduction keeps the directions with the most spread and drops the rest, which is useful and also lossy. Models work in spaces with far more dimensions than a page can show, so the sketch builds intuition that the algebra then has to confirm.
A visual routine before the symbols
Use a short routine on each new problem.
- Sketch the input vectors on grid paper.
- Mark where the two unit arrows should go under the map, and draw the result.
- Predict the sign and rough size of the determinant from the sketch.
- Apply the matrix to one input and compare the answer with the drawing.
- Only then carry out the full calculation, and check it against the prediction.
The routine catches errors that pure computation hides. A sign mistake or a transposed column shows up as a picture that disagrees with the prediction, even when the arithmetic looks plausible.
The same routine works in three dimensions, though the sketch is harder to draw. Use a simple set of axes and mark the three unit arrows. Most of the intuition carries over, and the places where a flat picture misleads are worth noting as you go.
Practice with a single transformation
Choose a two-by-two matrix with whole-number entries. Draw the unit square and its two arrows, then sketch where the square lands. Compute the determinant and check whether the sketched area agrees. Next, choose a matrix whose columns point in the same direction. Predict that the image will be a line, then confirm with the determinant. Finally, take three vectors in the plane and project them onto one direction, noting which parts survive. Write one sentence for each sketch stating in words what the map does. If those sentences come easily, the symbolic work that follows will be easier to trust.
