← Math simulations  ·  EE263, Stanford

The SVD is rotate, scale, rotate

Every matrix factors as A = UΣVT with U, V orthogonal and Σ diagonal. Geometrically that means any A acts on the plane in three simple steps: VT rotates, Σ stretches the axes by the singular values σ1, σ2, and U rotates again. So A maps the unit circle to an ellipse whose half-axes are σ1, σ2. Edit A, then press Play to watch the three steps, or step through them.

Matrix A

A =

Type numbers, drag the sliders, or pick a preset.

σ1 (max gain, longest axis)
σ2 (min gain, shortest axis)
condition σ12

Decomposition A = U Σ VT

U
×
Σ
×
VT

Presets

Show

v1 and its image σ1u1   v2 and σ2u2
input x (unit)   Ax