The Gram determinant of the vectors $\mathbf x_1,\mathbf x_2,\dots,\mathbf x_k \in \mathbb R^n$ is defined as:

$$\Gamma (\mathbf{x}_1, \ldots ., \mathbf{x}_k) = \left|\begin{array}{ccc} \mathbf{x}_1^{\top} \mathbf{x}_1 & \cdots & \mathbf{x}_1^{\top} \mathbf{x}_k\\ \vdots & \ddots & \vdots\\ \mathbf{x}_k^{\top} \mathbf{x}_1 & \cdots & \mathbf{x}_k^{\top} \mathbf{x}_k \end{array}\right| = \det (\mathbf{X}^{\top} \mathbf{X}) \qquad\qquad (1)$$

where $\mathbf{X}$ is the $n \times k$ rectangular matrix with columns $\mathbf{x}_1, \ldots, \mathbf{x}_k$. It gives the "sub-dimensional" squared volume of the parallelepiped spanned by these vectors.

If I apply a linear transformation $\mathbf A\in\mathbb{R}^{n\times n}$ on the vectors $\mathbf x_i$, the volume changes to

$$\Gamma (\mathbf{A}\mathbf{x}_1, \ldots, \mathbf{A}\mathbf{x}_k) = \det (\mathbf{X}^{\top} \mathbf{A}^{\top} \mathbf{A}\mathbf{X}) \qquad\qquad (2)$$

although I see no obvious way to simplify this. I'd expect the volume to change in a simple manner after a linear transformation, like a constant volume dilation, but

$$\Gamma (\mathbf{A}\mathbf{x}_1, \ldots ., \mathbf{A}\mathbf{x}_k) \ne [\det (\mathbf{A})]^2 \Gamma(\mathbf{x}_1, \ldots, \mathbf{x}_k)$$

in general (unless $k=n$ or $\mathbf B$ is proportional to the identity matrix).

I could apply a $k\times k$ matrix $\mathbf B$ "on the right" and use the identity,

$$\det (\mathbf{B}^{\top} \mathbf{X}^{\top} \mathbf{X}\mathbf{B}) = \det (\mathbf{X}^{\top} \mathbf{X}) [\det (\mathbf{B})]^2 \qquad\qquad (3)$$

i.e., a volume dilation. But of course this is not the same as (2) above. Here the matrix $\mathbf B$ is acting on the rows of $\mathbf X$ and the geometrical meaning of this operation is not obvious to me.

Why my intuition is fails here? Why a linear transformation does not result in a constant volume dilation of sub-dimensional volumes? And what is the geometrical meaning of (3)?

I am looking for answers to these questions that offer some intuition, and particularly geometrical insight.


First, the equation for the volume should be $(\det(X^{\top} X))^{1/2}$.

The reason why equation (2) holds, is because it really depends on the orientation of the lower-dimensional parallelopided.

Consider for instance the transformation of $\mathbb{R}^4=\mathbb{R}^2\times\mathbb{R}^2$ given by $\mathrm{diag}(2,2,3,3)$. If you consider a parallelogram in the first $\mathbb{R}^2$, the transformation multiplies the 2-volume (area) of the parallelogram with a factor $2\cdot 2=4$. A parallelogram in the second $\mathbb{R}^2$ would increase with a factor $3\cdot 3=9$.

If the parallelopipedum is a $n$-volume (so $k=n$), then a linear transformation just multiplies the volume with a factor $\det A$. To put it in a very non-mathematical, but intuitive way: it doesn't matter how you orient it, a volume will always stretch/shrink in all directions.

I can't give an interpretation of equation (3), I don't find a realistic interpretation of $\mathbf{X}\mathbf{B}$ that's related to the parallelopipedum.

  • $\begingroup$ Thanks, now I understand why sub-dimensional volumes are not all scaled in the same way under generic linear transforms. I would still like to understand (3) better. $\endgroup$ – becko Sep 25 '18 at 12:31

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.