What does a matrix do, and which outputs can disappear?

What do the numbers in a vector describe?

Translate between an arrow and a list of coordinates. Each representation makes a different aspect of the same object easier to inspect.

What does multiplying a vector mean?

Scaling one vector and combining two vectors are different operations. Check the operands and the type of result before interpreting the notation.

What is the same in a diagram and a column calculation?

Switch between geometric motion and coordinate arithmetic to see one linear operation from two angles. Strang’s column combination is a worked example; neither representation is best for every task.

How can columns explain a matrix equation?

Translate coordinates into a geometric picture. The columns of a matrix then show which combinations of vectors can reach the desired output.

Which equations survive a collapse in dimension?

Once a transformation flattens space, reachable outputs lie in its image. A target inside that image and one outside it lead to different answers.

Which outputs disappear when a transformation collapses space?

Some target vectors fall outside the image after a collapse. In the eigenvalue example it is A minus lambda times the identity that becomes singular; this does not make every system unsolvable.

Why shift a matrix before looking for an eigenvector?

A zero determinant signals collapse into a lower dimension. Applying that test to the shifted matrix explains how a nonzero vector can lie in its null space.

How do rotations change in four dimensions?

Compare a fixed axis in three dimensions with planes of rotation in four. The familiar axis description does not carry over unchanged.

Why does an exponential trace a circle?

Return to the plane and separate position from the rule for moving. If velocity equals position multiplied by i, it is perpendicular to the radius and has the radius’s length. Starting at one gives motion around the unit circle at unit angular speed. This account explains how a local rule determines a trajectory.

Now ask whether moving through angle a and then b is the same as moving through a+b. The group-theoretic account starts with composition: adding inputs corresponds to multiplying outputs. One account gives the velocity; the other lets you check how two actions combine. A single angle works beautifully here. It does not follow that one angle specifies every rotation in four dimensions.

Which distance survives when the higher-dimensional picture fails?

Write each corner center as coordinates ±1. Its distance from the origin is √d. With corner spheres of radius one, a central sphere tangent to them has radius √d−1. The nearest outer face stays two units from the origin. In ten dimensions those two numbers explain how the central sphere reaches outside the outer box. A three-dimensional drawing suggests a guess; the length formula and tangency condition carry the argument. Check which distances scale before taking a picture into another dimension.

When do two reflections rotate a figure, and when do they translate it?

In the example described at SIGGRAPH 2019, reflecting successively across two intersecting lines produces a rotation about their intersection. Reflecting across two distinct parallel lines produces a translation. The talk also considers those parallel lines in the projective model, where their intersection is a point at infinity representing their shared direction. Start by identifying the two lines used in the operation. Then ask which point stays fixed or how far the resulting motion moves every point in the same direction.

A circular representation looks right; does the model actually use it for addition?

This reading begins by translating between vector coordinates and geometric actions. The addition paper brings that reading habit inside a model: integer representations contain both linear variation and periodic structure, fitted as a generalized helix. Projecting numbers into circles is not enough. The authors replace internal activations to test whether the fitted representation restores the tendency toward the originally correct answer, and use ablations to investigate the components’ role.

Distinguish explaining a representation geometrically from explaining every computational step. The paper supplies intervention evidence for its tested models and two-digit, single-token addition regime, but does not isolate the exact construction of the answer helix from the two inputs. Drawing structure proposes a question; predicting and testing how a change affects the output makes the picture carry more explanatory weight.