Why Scalar Surface Integrals Ignore Reparametrization

One Surface, Different Parameter Coordinates

A surface patch can be described using many different pairs of parameters. Changing those parameters may stretch, compress, or reverse the parameter domain, even though the geometric points on the surface remain the same. A scalar surface integral must give the same value for every valid description of the same patch.

The reason is not that the area factor stays fixed. It generally changes. Rather, the change in the surface area factor exactly compensates for the change in area in the parameter domain. The chain rule describes the first change, and the two-dimensional change-of-variables formula accounts for the second.

Central idea: A regular, one-to-one change of parameters changes the local area factor by the absolute value of its two-dimensional Jacobian determinant. That is exactly the factor required by change of variables in a double integral.

Parameterizations and Reparametrizations

Let $D$ be a parameter domain in the $st$-plane, and let $\mathbf X:D\to\mathbb R^3$ be a continuously differentiable parameterization of a surface patch. Assume that $\mathbf X$ is one-to-one and regular: its tangent vectors $\mathbf X_s$ and $\mathbf X_t$ are linearly independent at each interior point of $D$. Equivalently, $\mathbf X_s\times\mathbf X_t\ne\mathbf 0$.

Let $E$ be another parameter domain, now in the $uv$-plane. A change of parameters is a map $\Phi:E\to D$, written as $\Phi(u,v)=(s(u,v),t(u,v))$. We use it to define a second description of the surface by composing the maps:

\mathbf Y(u,v)=\mathbf X(\Phi(u,v)) =\mathbf X(s(u,v),t(u,v)).

For this to be a regular, one-to-one reparametrization, assume that $\Phi$ is a bijection from $E$ to $D$, is continuously differentiable, has a continuously differentiable inverse in the interiors, and has nonzero Jacobian determinant there. The two parameterizations then describe the same patch, with each point represented once in each parameter domain. As usual, boundaries can be included when they have area zero and the integrands extend appropriately.

Definition: A regular, one-to-one reparametrization of $\mathbf X$ is a map $\mathbf Y=\mathbf X\circ\Phi$, where $\Phi:E\to D$ is a continuously differentiable bijection with continuously differentiable inverse and nonzero Jacobian determinant in the interiors of the domains.

If $f$ is a scalar-valued function on the surface, its scalar surface integral using $\mathbf X$ is

\iint_D f(\mathbf X(s,t))\,\lVert\mathbf X_s\times\mathbf X_t\rVert\,ds\,dt.

The cross-product norm is the surface area factor: a small parameter rectangle is mapped approximately to a parallelogram whose area is its parameter area multiplied by this factor. The corresponding integral using $\mathbf Y$ is defined in the same way, over $E$.

How the Area Factor Changes

Write the derivative matrix of $\Phi$ as

D\Phi(u,v)= \begin{pmatrix} s_u&s_v\\ t_u&t_v \end{pmatrix}, \qquad \det D\Phi=s_ut_v-s_vt_u.

The chain rule gives the two tangent vectors of the reparametrized surface:

\mathbf Y_u =s_u\mathbf X_s+t_u\mathbf X_t, \qquad \mathbf Y_v =s_v\mathbf X_s+t_v\mathbf X_t,

where the derivatives of $\mathbf X$ are evaluated at $(s(u,v),t(u,v))$. Taking the cross product and using its bilinearity gives

\begin{aligned} \mathbf Y_u\times\mathbf Y_v &=(s_u\mathbf X_s+t_u\mathbf X_t) \times(s_v\mathbf X_s+t_v\mathbf X_t)\\ &=s_ut_v(\mathbf X_s\times\mathbf X_t) +t_us_v(\mathbf X_t\times\mathbf X_s)\\ &=(s_ut_v-s_vt_u)(\mathbf X_s\times\mathbf X_t). \end{aligned}

In the second line, the terms involving a vector crossed with itself vanish. In the last line, $\mathbf X_t\times\mathbf X_s=-(\mathbf X_s\times\mathbf X_t)$. Taking norms therefore yields

\lVert\mathbf Y_u\times\mathbf Y_v\rVert = \left|\det D\Phi(u,v)\right| \lVert\mathbf X_s\times\mathbf X_t\rVert.

This identity is the key calculation. The absolute value appears because a surface area factor is nonnegative. If $\Phi$ reverses orientation, its determinant is negative, but that reversal does not make area negative.

The Reparametrization Theorem

Theorem: Suppose $\mathbf X:D\to\mathbb R^3$ is a regular, one-to-one parameterization and $\Phi:E\to D$ is a regular, one-to-one reparametrization satisfying the conditions above. If $f$ is integrable on the surface patch, then $$ \iint_E f(\mathbf Y(u,v))\,\lVert\mathbf Y_u\times\mathbf Y_v\rVert\,du\,dv = \iint_D f(\mathbf X(s,t))\,\lVert\mathbf X_s\times\mathbf X_t\rVert\,ds\,dt. $$

To justify the result, first use $\mathbf Y=\mathbf X\circ\Phi$ and the area-factor identity. The integrand in the $uv$-domain becomes the original integrand, evaluated at $\Phi(u,v)$, multiplied by the absolute Jacobian:

\begin{aligned} f(\mathbf Y(u,v))\lVert\mathbf Y_u\times\mathbf Y_v\rVert &=f(\mathbf X(\Phi(u,v))) \lVert\mathbf X_s\times\mathbf X_t\rVert_{\Phi(u,v)} \left|\det D\Phi(u,v)\right|. \end{aligned}

Now apply the two-dimensional change-of-variables formula. For an integrable function $g$ on $D$ and a continuously differentiable bijection $\Phi:E\to D$ with nonzero Jacobian, that formula states that the integral of $g$ over $D$ equals the integral of $g\circ\Phi$ multiplied by $\lvert\det D\Phi\rvert$ over $E$. Take $g(s,t)=f(\mathbf X(s,t))\lVert\mathbf X_s\times\mathbf X_t\rVert$. This gives precisely the theorem’s equality.

The argument shows why both ingredients matter. The chain rule supplies the Jacobian in the transformed area factor; change of variables uses that same Jacobian to compensate for the altered parameter domain. Neither factor can be omitted.

Worked Examples

Worked Example: A Nonlinear Change on a Plane Patch

Let $\mathbf X(s,t)=(s,t,0)$, where $D$ is the region $0\le s\le1$, $0\le t\le1+s$. Let $\Phi(u,v)=(u,(1+u)v)$ for $0\le u\le1$, $0\le v\le1$, and take $f(x,y,z)=1+x+y$. Compute the scalar surface integral in both parameterizations.

First, $\mathbf X_s=(1,0,0)$ and $\mathbf X_t=(0,1,0)$, so $\mathbf X_s\times\mathbf X_t=(0,0,1)$ and its norm is $1$. Since $x=s$ and $y=t$, the integral in $D$ is

\begin{aligned} \iint_D f\,dS &=\int_0^1\int_0^{1+s}(1+s+t)\,dt\,ds\\ &=\int_0^1\left[(1+s)t+\frac{t^2}{2}\right]_{0}^{1+s}ds\\ &=\int_0^1\left((1+s)^2+\frac{(1+s)^2}{2}\right)ds\\ &=\frac32\int_0^1(1+s)^2\,ds\\ &=\frac32\left[\frac{(1+s)^3}{3}\right]_0^1 =\frac32\cdot\frac73=\frac72. \end{aligned}

The map $\Phi$ has Jacobian matrix $\begin{pmatrix}1&0\\v&1+u\end{pmatrix}$, so $\det D\Phi=1+u>0$. The reparameterized surface is $\mathbf Y(u,v)=(u,(1+u)v,0)$. Its scalar value is $f(\mathbf Y)=1+u+(1+u)v$, and the area factor is $1+u$. Thus

\begin{aligned} \iint_E f\,dS &=\int_0^1\int_0^1 \bigl(1+u+(1+u)v\bigr)(1+u)\,dv\,du\\ &=\int_0^1(1+u)^2 \left[v+\frac{v^2}{2}\right]_0^1du\\ &=\frac32\int_0^1(1+u)^2\,du\\ &=\frac32\cdot\frac73=\frac72. \end{aligned}

The two calculations agree. The factor $1+u$ accounts for the stretching of the parameter domain under $\Phi$.

Worked Example: Reversing and Rescaling Parameters

Consider the plane patch $\mathbf X(s,t)=(s,t,s)$ over $D=[0,1]\times[0,1]$, and let $f(x,y,z)=x$. Use the parameter transformation $\Phi(u,v)=(1-2u,v)$ on $E=[0,\tfrac12]\times[0,1]$. This maps $E$ bijectively onto $D$, reversing the first parameter direction.

For the original parameterization, $\mathbf X_s=(1,0,1)$, $\mathbf X_t=(0,1,0)$, and $\mathbf X_s\times\mathbf X_t=(-1,0,1)$, whose norm is $\sqrt2$. Therefore

\begin{aligned} \iint_D f\,dS &=\int_0^1\int_0^1 s\sqrt2\,dt\,ds\\ &=\sqrt2\left(\int_0^1s\,ds\right)\left(\int_0^1dt\right)\\ &=\sqrt2\cdot\frac12\cdot1=\frac{\sqrt2}{2}. \end{aligned}

Here $D\Phi=\begin{pmatrix}-2&0\\0&1\end{pmatrix}$, so $\det D\Phi=-2$ and $\lvert\det D\Phi\rvert=2$. The new parameterization is $\mathbf Y(u,v)=(1-2u,v,1-2u)$, with area factor $2\sqrt2$. Also $f(\mathbf Y)=1-2u$. Hence

\begin{aligned} \iint_E f\,dS &=\int_0^{1/2}\int_0^1(1-2u)2\sqrt2\,dv\,du\\ &=2\sqrt2\int_0^{1/2}(1-2u)\,du\\ &=2\sqrt2\left[u-u^2\right]_0^{1/2}\\ &=2\sqrt2\left(\frac12-\frac14\right) =\frac{\sqrt2}{2}. \end{aligned}

The negative determinant records orientation reversal, not negative area. Using $\det D\Phi$ instead of its absolute value would incorrectly give a negative area factor.

Worked Example: A Curved Patch with a Convenient Scalar Function

Let $\mathbf X(s,t)=(s,t,st)$ on the region $D$ given by $0\le s\le1$, $0\le t\le1+s$. Take $f(x,y,z)=1/\sqrt{1+x^2+y^2}$. Reparametrize using $\Phi(u,v)=(u,(1+u)v)$ on the unit square $E$.

The tangent vectors are $\mathbf X_s=(1,0,t)$ and $\mathbf X_t=(0,1,s)$. Their cross product is $(-t,-s,1)$, so its norm is $\sqrt{1+s^2+t^2}$. On the surface, $x=s$ and $y=t$, so $f(\mathbf X(s,t))=1/\sqrt{1+s^2+t^2}$. Their product is $1$, and therefore

\begin{aligned} \iint_D f\,dS &=\int_0^1\int_0^{1+s}1\,dt\,ds\\ &=\int_0^1(1+s)\,ds\\ &=\left[s+\frac{s^2}{2}\right]_0^1 =\frac32. \end{aligned}

For the new parameters, $s=u$ and $t=(1+u)v$, with Jacobian determinant $1+u$. The area-factor identity gives $\lVert\mathbf Y_u\times\mathbf Y_v\rVert=(1+u)\sqrt{1+u^2+(1+u)^2v^2}$. The scalar function at $\mathbf Y$ is the reciprocal of the square root in this expression. Thus

\begin{aligned} \iint_E f\,dS &=\int_0^1\int_0^1(1+u)\,dv\,du\\ &=\int_0^1(1+u)\,du\\ &=\left[u+\frac{u^2}{2}\right]_0^1 =\frac32. \end{aligned}

Again the values match. The curved surface changes the area factor, but the chosen scalar function cancels its original geometric part, leaving the Jacobian as the only extra factor after reparametrization.

When the Argument Applies—and a Common Mistake

The theorem requires a one-to-one change of parameters with a nonzero Jacobian in the interior. These conditions ensure that the surface patch is covered once and that the two-dimensional change-of-variables formula applies in the stated form. If a parameter map covers the same points multiple times, simply applying this formula as though it were one-to-one can count some of the surface more than once. If the Jacobian vanishes, the map is not a regular change of coordinates at those points, so the theorem cannot be invoked there without additional analysis.

A frequent error is to change the formula inside the integral but forget the parameter Jacobian. In the first example, this would replace $s=u$, $t=(1+u)v$ in $f$, but incorrectly keep the area factor equal to $1$. The incorrect calculation would be

\begin{aligned} \int_0^1\int_0^1\bigl(1+u+(1+u)v\bigr)\,dv\,du &=\int_0^1\frac32(1+u)\,du\\ &=\frac32\cdot\frac32=\frac94, \end{aligned}

which disagrees with the correct value $7/2$. The error is that the new area factor is not $1$: it is $1+u$. Multiplying by that factor gives the correct transformed integrand and restores the value $7/2$.

Another version of the same mistake is to use the signed determinant in an area integral. As the second example shows, a negative determinant means the parameter orientation has reversed. It does not mean that the surface has negative area. Scalar surface integrals use the norm of a cross product, and therefore the absolute value of the Jacobian.

Why Reparametrization Invariance Matters

A parameterization is a coordinate description, not the surface itself. The ability to change coordinates without changing a scalar surface integral lets us choose a description that makes the domain or the calculation more manageable. The result also provides a useful consistency check: if two valid parameterizations describe exactly the same patch and produce different scalar surface integrals, then some factor, domain, or substitution has been handled incorrectly.

The essential calculation can be remembered in three linked steps: compose the surface parameterization with the coordinate change, apply the chain rule to the tangent vectors, and use change of variables with the absolute Jacobian. The area factor changes, but the scalar surface integral does not.

Check Your Understanding

Answer each question by referring to the area-factor identity and the change-of-variables argument.

What condition on the two tangent vectors makes a surface parameterization regular?
If $\det D\Phi=-3$, what factor multiplies the original surface area factor?
Why does the transformed area factor contain the absolute value of the Jacobian determinant?
Which chain-rule identities relate $\mathbf Y_u,\mathbf Y_v$ to $\mathbf X_s,\mathbf X_t$?
Why can a parameter map that covers a patch more than once not be treated as a one-to-one reparametrization?