Eigenvectors Are Directions a Linear Operator Only Scales

A linear operator $T:V\to V$ takes vectors in a space and sends them back into the same space. Most vectors are moved to a different direction. An eigenvector is the special case where the direction survives.

The equation is small:

$$Tx=\lambda x,x\ne 0.$$

The operator $T$ acts on $x$, but the result stays on the same line through $x$. The scalar $\lambda$ records what happened along that line: stretch, shrink, sign flip, or collapse.

Direction is the invariant part

A general vector can be turned, sheared, or mixed with other directions. An eigenvector is not protected from scaling, but it is protected from being turned away from its own line.

The non-zero condition matters. The zero vector always satisfies $T0=\lambda 0$, but it carries no direction. Eigenvectors are direction carriers, so $x=0$ is excluded.

The eigenvalue is the scale factor

Once the direction survives, the only remaining question is how the vector changed along that direction. That number is the eigenvalue.

For an eigenvector $x$,

$$Tx=\lambda x.$$

Different signs and sizes of $\lambda$ mean different one-dimensional behaviours:

value of $\lambda$	effect on an eigenvector $x$
$\lambda >1$	stretches along the same direction
$0<\lambda <1$	shrinks along the same direction
$\lambda =1$	leaves $x$ unchanged
$\lambda =0$	sends $x$ to $0$
$\lambda <0$	flips direction and scales

The eigenvalue belongs to the direction, not just to one vector. If $x$ is an eigenvector for $\lambda$, then any non-zero scalar multiple $\alpha x$ is also an eigenvector for the same $\lambda$:

$$T(\alpha x)=\alpha Tx=\alpha \lambda x=\lambda (\alpha x).$$

So the real object is often the whole line, or more generally the whole subspace, of vectors sharing the same scale factor.

Eigenspaces collect all directions with the same scale

Fix a scalar $\lambda$. The eigenspace for $\lambda$ is the set of vectors on which $T$ behaves exactly like scalar multiplication by $\lambda$:

$$E_{\lambda }=\{x\in V:Tx=\lambda x\}.$$

This set includes $0$, even though $0$ is not an eigenvector. That is useful: with $0$ included, $E_{\lambda }$ becomes a subspace.

If $x,y\in E_{\lambda }$, then

$$\begin{array}{rl}T(\alpha x+\beta y)& =\alpha Tx+\beta Ty\\ & =\alpha \lambda x+\beta \lambda y\\ & =\lambda (\alpha x+\beta y).\end{array}$$

So any linear combination of vectors in $E_{\lambda }$ stays in $E_{\lambda }$.

An eigenspace is therefore not a pile of unrelated eigenvectors. It is the whole subspace where the operator acts like the simple operation $x\mapsto \lambda x$.

Subtract the simple part

The equation

$$Tx=\lambda x$$

says that two actions agree on $x$: applying $T$ and scaling by $\lambda$. Move them to one side:

$$Tx-\lambda x=0.$$

Using the identity operator $I$, this becomes

$$(T-\lambda I)x=0.$$

So the eigenspace is a kernel:

$$E_{\lambda }=\ker (T-\lambda I).$$

This is the practical test. A scalar $\lambda$ is an eigenvalue exactly when $T-\lambda I$ has a non-zero vector in its kernel. For a matrix $A$, this becomes

$$(A-\lambda I)x=0,x\ne 0.$$

A non-zero solution means that $A-\lambda I$ collapses some direction to zero. That direction is an eigenvector of $A$.

The geometric picture

Eigenvectors are the directions where the operator becomes one-dimensional. On an eigenspace, the complicated operator $T$ behaves like the simple scalar operator $\lambda I$.

That is the core compression:

$$\text{complicated linear action}⇝\text{pure scaling on special subspaces}.$$

Eigenvalues name the scaling. Eigenvectors carry the directions. Eigenspaces collect all vectors that share the same scaling behaviour.