Jacobian Determinants and Local Volume Scaling

Why a Determinant Measures Local Area Change

A smooth transformation of two variables can stretch, compress, shear, or reverse a small region. The Jacobian determinant summarizes the transformation’s first-order effect on area at a point. Its magnitude measures the local area multiplier; its sign records orientation. Keeping those roles separate explains why a density transformation uses the absolute value of the determinant.

The key idea is local. A nonlinear map may stretch one small region more than another, but sufficiently close to a point it behaves approximately like its derivative, which is a linear map. The determinant of that linear map gives the area scaling to first order.

The Bivariate Jacobian

Definition: Let $T$ map an open set in $\mathbb{R}^2$ to $\mathbb{R}^2$, with $T(u,v)=(T_1(u,v),T_2(u,v))$. If its first partial derivatives exist, its Jacobian matrix at $(u,v)$ is $$ D T(u,v)= \begin{pmatrix} \frac{\partial T_1}{\partial u} & \frac{\partial T_1}{\partial v}\\ \frac{\partial T_2}{\partial u} & \frac{\partial T_2}{\partial v} \end{pmatrix}. $$ The Jacobian determinant is $$ J_T(u,v)=\det D T(u,v) =\frac{\partial T_1}{\partial u}\frac{\partial T_2}{\partial v} -\frac{\partial T_1}{\partial v}\frac{\partial T_2}{\partial u}. $$

The columns of $D T(u,v)$ describe the images, to first order, of small movements in the $u$- and $v$-directions. If those columns are $a=(a_1,a_2)$ and $b=(b_1,b_2)$, they span a parallelogram whose signed area is $a_1b_2-a_2b_1$. Thus its ordinary, nonnegative area is $|a_1b_2-a_2b_1|$.

Geometric interpretation: At a point where $T$ is differentiable, $|J_T(u,v)|$ is the first-order local area multiplier. A sufficiently small region near $(u,v)$ has image area approximately $|J_T(u,v)|$ times its original area. A negative determinant reverses orientation; it does not produce negative area.

From Linear Maps to Local Scaling

For a linear map with matrix $A$, the unit square is mapped to the parallelogram spanned by the columns of $A$. The determinant is the signed area of that parallelogram, so the area multiplier is $|\det A|$. A square with side length $r$ has area $r^2$, and its image under the same linear map has area $r^2|\det A|$. The multiplier is therefore independent of the square’s size.

For a differentiable nonlinear map, the derivative supplies the corresponding linear approximation. If $h=(h_1,h_2)$ is a small displacement from $x=(u,v)$, differentiability gives

T(x+h)=T(x)+D T(x)h+o(\lVert h\rVert).

The error term is small relative to the displacement as $h$ approaches zero. Consequently, after shrinking a small region around $x$, its image is approximated by the parallelogram produced by $D T(x)$. The area ratio approaches $|\det D T(x)|$. This is a local statement: a varying Jacobian determinant means that equal-sized regions at different locations can have different image areas.

Worked Example: A Nonlinear Map at Two Points

Consider $T(u,v)=(u+v,e^u+v)$. Find its Jacobian determinant and interpret the local area scaling at $(\ln 3,0)$.

The derivative matrix and determinant are

D T(u,v)= \begin{pmatrix} 1&1\\ e^u&1 \end{pmatrix}, \qquad J_T(u,v)=1-e^u.

At $(\ln 3,0)$, the determinant is $1-e^{\ln 3}=1-3=-2$. The map sends this point to $T(\ln 3,0)=(\ln 3,3)$. A small square of area $\varepsilon^2$ around the input point is therefore mapped to a region with area approximately $2\varepsilon^2$. The negative sign indicates reversed orientation, not negative area. For comparison, at $(\ln 2,0)$ the determinant is $-1$, so the first-order area multiplier there is $1$, again with reversed orientation.

Why Density Formulas Use the Absolute Value

Suppose $X=(U,V)$ has density $f_X$, and $Y=T(X)$. If $T$ is a continuously differentiable bijection between open regions, with a continuously differentiable inverse and nonzero Jacobian determinant, the density transformation is

f_Y(y)=f_X\!\left(T^{-1}(y)\right) \left|\det D T^{-1}(y)\right|.

The inverse determinant appears because this formula asks how much area in the $y$-space corresponds to a given small area in the $x$-space. If $T$ expands area by a factor of $c>0$, then the corresponding $y$-region is $c$ times as large as the $x$-region. Probability mass is unchanged, so density per unit area must be divided by $c$. Algebraically, the chain rule gives

D T^{-1}(T(x))=[D T(x)]^{-1}, \qquad \left|\det D T^{-1}(T(x))\right| =\frac{1}{|\det D T(x)|}.

More formally, for a measurable set $B$ in the image region, the multivariate change-of-variables formula gives

P(Y\in B) =\int_{T^{-1}(B)} f_X(x)\,dx =\int_B f_X(T^{-1}(y)) \left|\det D T^{-1}(y)\right|\,dy.

The absolute value is essential because probability mass and area are nonnegative. Orientation reversal changes the sign of a signed determinant but cannot make either quantity negative.

Worked Example: An Affine Map and Its Transformed Density

Let $X=(U,V)$ be uniform on the unit square $0\leq u\leq1,\ 0\leq v\leq1$, so $f_X(u,v)=1$ there. Apply $T(u,v)=(2u+v,u+3v)$. Find the density of $Y=T(X)$.

The derivative matrix is constant, and its determinant is

D T= \begin{pmatrix} 2&1\\ 1&3 \end{pmatrix}, \qquad \det D T=2(3)-1(1)=5.

The inverse matrix is

(D T)^{-1} =\frac{1}{5} \begin{pmatrix} 3&-1\\ -1&2 \end{pmatrix}.

Writing $y=(x,y_2)$ to distinguish the two output coordinates, the inverse transformation is $u=(3x-y_2)/5$ and $v=(-x+2y_2)/5$. Its determinant has absolute value $1/5$. Therefore the transformed density is $1/5$ wherever the inverse point lies in the original unit square, and is zero elsewhere:

f_Y(x,y_2)= \begin{cases} \frac15, & 0\leq \frac{3x-y_2}{5}\leq1 \ \text{and}\ 0\leq \frac{-x+2y_2}{5}\leq1,\\ 0, & \text{otherwise}. \end{cases}

The image of the unit square is a parallelogram with area $5$, because the area multiplier is $5$. The density integrates to $(1/5)(5)=1$, as required. This checks both the determinant and the transformed density.

Worked Example: Reflection and Expansion

Let $U$ and $V$ be independent uniform random variables on $[0,1]$, and define $T(u,v)=(u,-2v)$. Determine the local area multiplier and the density of $Y=T(U,V)$.

Here

D T= \begin{pmatrix} 1&0\\ 0&-2 \end{pmatrix}, \qquad \det D T=-2, \qquad |\det D T|=2.

The negative sign records reflection across the horizontal axis, while the magnitude says areas double. The image of the unit square is the rectangle $0\leq x\leq1,\ -2\leq y\leq0$, with area $2$. The inverse is $u=x,\ v=-y/2$, and its Jacobian determinant has absolute value $1/2$. Since the original joint density is $1$ on the square,

f_Y(x,y)= \begin{cases} \frac12, & 0\leq x\leq1,\ -2\leq y\leq0,\\ 0, & \text{otherwise}. \end{cases}

The density integrates to $(1/2)(1)(2)=1$. Using the signed value $-2$ as a density multiplier would incorrectly give a negative density.

Common Mistake: Treating the Sign as an Area

A frequent error is to say that a negative Jacobian determinant means a region has negative area, or to insert a negative determinant into a density formula. The determinant is signed because it records orientation. The actual area multiplier is its absolute value, and the density formula uses the absolute determinant of the inverse transformation. Check the direction of the map as well: the density of $Y$ uses $\left|\det D T^{-1}(y)\right|$, not $\left|\det D T(x)\right|$.

Key takeaway: The Jacobian determinant combines two pieces of local geometry. Its magnitude gives the factor by which area changes; its sign indicates whether orientation is preserved or reversed. Probability densities use the magnitude of the inverse map’s determinant because probability mass stays nonnegative and unchanged under a transformation.

Check Your Understanding

Answer each question using the distinction between signed orientation and nonnegative area.

For a bivariate derivative matrix with columns $a$ and $b$, what geometric object do those columns span, and what is its area?
A transformation has determinant $-4$ at a point. State its local area multiplier and interpret the negative sign.
If a one-to-one transformation expands local area by a factor of $3$, what factor appears in the transformed density from the inverse Jacobian?
For $T(u,v)=(3u+v,u+v)$, calculate the Jacobian determinant and the local area multiplier.
Why would using a signed negative Jacobian as a density multiplier violate a basic property of densities?