Let $A$ be an invertible $n \times n$ matrix. Suppose we interpret each row of $A$ as a point in $\mathbb{R}^n$; then these $n$ points define a unique hyperplane in $\mathbb{R}^n$ that passes through each point (this hyperplane does not intersect the origin).

Under this geometric interpretation, $A^{-1}$ has an interesting property: the normal vector to the hyperplane is given by the row sums of $A^{-1}$ (i.e. $A^{-1} \cdot 1$, where $1 = \langle 1, \dots, 1 \rangle^T$).

Within this geometric interpretation of $A$, what other interesting properties does $A^{-1}$ have? Do the individual entries of $A^{-1}$ have geometric meaning? How about the column sums?


Here is a visual answer for the $2\times 2$ case.

  • Plot the row (or column) vectors, $a_1, a_2$ of $A$ in $\mathcal{R}^2$ to visualize $A$. The area of the parallelogram they form is of course $\det(A)$.
  • In the same space, plot the row (or column) vectors $a^1, a^2$ of $A^{-1}$, and the area of their parallelogram is then $\det(A^{-1}) = 1/ \det(A) $.
  • The relationship between the two illustrates various properties of the matrix inverse.

An example is shown in the picture below, which comes from the matrix (in R notation)

A <- matrix(c(2, 1, 
              1, 2), nrow=2, byrow=TRUE)

In the R package matlib I recently added a vignette illustrating this with the following diagram for this matrix.

inverse diagram

Thus, we can see:

  • The shape of $A^{-1}$ is a $90^o$ rotation of the shape of $A$.

  • $A^{-1}$ is small in the directions where $A$ is large

  • The vector $a^2$ is at right angles to $a_1$ and $a^1$ is at right angles to $a_2$

  • If we multiplied $A$ by a constant $k$ to make its determinant larger (by a factor of $k^2$), the inverse would have to be divided by the same factor to preserve $A A^{-1} = I$.

I wondered whether these properties depend on symmetry of $A$, so here is another example, for the matrix A <- matrix(c(2, 1, 1, 1), nrow=2), where $\det(A)=1$.

enter image description here

It would be interesting to extend this to other properties and to the $3 \times 3$ case, which I leave to others.

| cite | improve this answer | |
  • 2
    $\begingroup$ Not all your observations are true in the general case. The 90° thing for example. You just got lucky with your numbers. For example $A = [[-1, 0], [-3, 2]]$, $A^{-1} = [[-1, -0. ], [-1.5, 0.5]]$ $A_1 \cdot A^{-1}_2 = -1.5 \neq 0 $. Note: internal brackets are rows. $\endgroup$ – user3578468 May 25 '19 at 2:42
  • $\begingroup$ Please provide statements of which of my observations are not true in the general case and why. $\endgroup$ – user101089 May 25 '19 at 22:13
  • $\begingroup$ But I did give you an example. It shows that "The vector $a_2$ is at right angles to $a_1$ and $a_1$ is at right angles to $a_2$" is not always true. $\endgroup$ – user3578468 May 27 '19 at 10:50

It turns out that an answer for the $3 \times 3$ case has similar properties and is also illuminating.

  • Start with a unit cube, representing the identity matrix. Show its transformation by a matrix $A$ as the corresponding transformation of the cube.

  • This also illustrates the determinant, det(A), as the volume of the transformed cube, and the relationship between $A$ and $A^{-1}$.

In R, using the matlib and rgl package, the unit cube is specified as

# cube, with each face colored differently
colors <- rep(2:7, each=4)
c3d <- cube3d()
# make it a unit cube at the origin
c3d <- scale3d(translate3d(c3d, 1, 1, 1),
               .5, .5, .5)

A $3 \times 3$ matrix $A$ with $\det(A)=2$ is

A <- matrix(c( 1, 0, 1, 
               0, 2, 0,  
               1, 0, 2), nrow=3, ncol=3)

Extending the 2D idea from the answer above of drawing the images of $A$ and $A^{-1}$ together to 3D, we get the following, best viewed as an animated graphic. The faces of the parallelpiped representing $A^{1}$ are colored identically to those of $A$, so you can see the mapping from one to the other.


| cite | improve this answer | |
  • $\begingroup$ Great visualization! $\endgroup$ – Vincent Oct 17 '18 at 13:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.