Bounding Surface Integrals by Comparison

Why Bound a Surface Integral?

A scalar surface integral combines the values of a function across a surface, weighted by surface area. Sometimes an exact calculation is difficult, but it is still possible to establish a useful range for the answer. Even when an integral has been computed exactly, an independent estimate can reveal whether its size is plausible.

The basic idea is simple: if the integrand never falls below a known value or rises above another, its integral is constrained by those values and by the surface area. The details matter, especially when the integrand can be negative or when the area itself is known only approximately.

Comparison for Scalar Surface Integrals

Let $S$ be a piecewise smooth surface with finite area, and let $f:S\to\mathbb R$ be a continuous scalar function. The scalar surface integral $\int_S f\,dS$ weights each small area element by the value of $f$ there. Surface area is nonnegative, so pointwise inequalities for $f$ can be integrated while preserving their direction.

Theorem: If $m\leq f(\mathbf{x})\leq M$ at every point $\mathbf{x}\in S$, then $$ m\,\operatorname{Area}(S) \leq \int_S f\,dS \leq M\,\operatorname{Area}(S). $$ Here $m$ and $M$ may be negative. The theorem requires finite surface area and valid bounds everywhere on $S$, apart from sets of surface area zero.

To see why the theorem holds, subtract $m$ from $f$. Since $f-m\geq0$, its integral is nonnegative. Thus $\int_S f\,dS-m\operatorname{Area}(S)\geq0$. Applying the same reasoning to $M-f\geq0$ gives $M\operatorname{Area}(S)-\int_S f\,dS\geq0$. Together these are the stated inequalities.

If $f\geq0$, the comparison has an especially useful form when the area is itself bounded. Suppose $0\leq m\leq f\leq M$, and suppose $A_-\leq\operatorname{Area}(S)\leq A_+$, where $0\leq A_-\leq A_+$. Then

mA_- \leq \int_S f\,dS \leq MA_+.

The lower estimate uses the smallest allowed function value and the smallest allowed area. The upper estimate uses the largest of each. If $m=0$, the lower estimate is simply zero; a more informative lower bound requires a strictly positive lower bound for $f$ and a positive lower bound for the area.

Combining Bounds When the Integrand Can Be Negative

For signed integrands, it is not generally correct to pair the lower function bound with the lower area bound and the upper function bound with the upper area bound. Multiplying inequalities by negative quantities can reverse their direction. Instead, consider all endpoint products.

Theorem: Suppose $m\leq f\leq M$ on $S$, and $0\leq A_-\leq\operatorname{Area}(S)\leq A_+$. Then $$ \min\{mA_-,mA_+,MA_-,MA_+\} \leq \int_S f\,dS \leq \max\{mA_-,mA_+,MA_-,MA_+\}. $$

For a surface of positive area $a$, the average value $q=(1/a)\int_S f\,dS$ satisfies $m\leq q\leq M$, by the comparison theorem. Hence the integral equals $qa$, where $q\in[m,M]$ and $a\in[A_-,A_+]$. The smallest and largest possible products of numbers in these two intervals occur among the four endpoint products. This proves the bound. If the surface has zero area, its scalar surface integral is zero, and the same conclusion follows because the area interval then includes zero.

When $f$ is known to be nonnegative, the simpler bounds $mA_-$ and $MA_+$ are convenient. For a signed function, use the endpoint-product rule or retain a known exact area and apply the first comparison theorem directly.

Getting an Area Bound from a Parametrization

A regular parametrization $\mathbf{r}(u,v)$ has area density $J(u,v)=\lVert\mathbf{r}_u\times\mathbf{r}_v\rVert$. For a parameter domain $D$, the area is $\iint_D J(u,v)\,du\,dv$, provided the parametrization covers the surface exactly once except possibly on boundary sets of area zero. Otherwise, the parameter integral counts area with multiplicity, which may exceed the geometric area.

If $D$ has finite area and $J_-\leq J(u,v)\leq J_+$ throughout $D$, then integrating these inequalities gives $$ J_-\,\operatorname{Area}(D) \leq \operatorname{Area}(S) \leq J_+\,\operatorname{Area}(D). $$ For example, continuous $J$ on a compact parameter domain has a minimum and maximum, so those extrema provide such bounds. They need not be easy to calculate sharply: even coarse valid bounds can help assess the scale of a scalar integral.

Worked Comparisons

Worked Example: A Scalar Integral on a Tilted Plane

Parametrize the patch by $\mathbf{r}(u,v)=(u,v,u+2v)$ for $0\leq u\leq1$ and $0\leq v\leq2$. Estimate and then calculate the integral of $f(x,y,z)=2+x$.

The parameter domain is a rectangle of area $1\cdot2=2$, and this parametrization is one-to-one. Its tangent vectors and area density are $$ \mathbf{r}_u=(1,0,1),\qquad \mathbf{r}_v=(0,1,2),\qquad \mathbf{r}_u\times\mathbf{r}_v=(-1,-2,1),\qquad J=\sqrt{1+4+1}=\sqrt{6}. $$ Thus $\operatorname{Area}(S)=2\sqrt{6}$. On the patch, $x=u$, so $2\leq f\leq3$. Comparison gives $$ 2(2\sqrt{6})\leq \int_S f\,dS\leq3(2\sqrt{6}), \qquad 4\sqrt{6}\leq \int_S f\,dS\leq6\sqrt{6}. $$

For a direct calculation, substitute $f=2+u$ and $dS=\sqrt{6}\,du\,dv$: $$ \begin{aligned} \int_S f\,dS &=\int_0^2\int_0^1(2+u)\sqrt{6}\,du\,dv\\ &=\int_0^2\sqrt{6}\left[2u+\frac{u^2}{2}\right]_0^1\,dv\\ &=\int_0^2\frac{5\sqrt{6}}{2}\,dv =5\sqrt{6}. \end{aligned} $$ This lies between $4\sqrt{6}$ and $6\sqrt{6}$. As a check, the inner integral is $2+\tfrac12=\tfrac52$, and multiplying by the $v$-interval length $2$ gives $5\sqrt6$.

Worked Example: A Nonnegative Integral on a Paraboloid

Consider the graph $z=(x^2+y^2)/2$ above the unit disk $x^2+y^2\leq1$, and let $f(x,y,z)=1+x^2+y^2$. Find comparison bounds, then evaluate the integral to check them.

The graph is covered once by its disk domain. Its area density is $J=\sqrt{1+x^2+y^2}$. Since $0\leq x^2+y^2\leq1$, the density satisfies $1\leq J\leq\sqrt2$, and the disk has area $\pi$. Therefore $$ \pi\leq\operatorname{Area}(S)\leq\pi\sqrt2. $$ The integrand satisfies $1\leq f\leq2$. The nonnegative comparison bounds are consequently $$ \pi\leq\int_S f\,dS\leq2\pi\sqrt2. $$

To check, use polar coordinates, with radius $r$ from $0$ to $1$ and angle $\theta$ from $0$ to $2\pi$: $$ \begin{aligned} \int_S f\,dS &=\int_0^{2\pi}\int_0^1(1+r^2)^{3/2}r\,dr\,d\theta\\ &=2\pi\cdot\frac12\int_1^2 t^{3/2}\,dt \qquad (t=1+r^2,\;dt=2r\,dr)\\ &=\frac{2\pi}{5}\left(2^{5/2}-1\right) =\frac{2\pi}{5}(4\sqrt2-1). \end{aligned} $$ Indeed, $\frac{d}{dt}\bigl(\frac25t^{5/2}\bigr)=t^{3/2}$, so the substituted antiderivative gives the displayed value. Dividing by $\pi$, the exact result is $\frac25(4\sqrt2-1)$. It exceeds $1$ because $8\sqrt2-2\geq5$, equivalent to $\sqrt2\geq7/8$. It is at most $2\sqrt2$ because $8\sqrt2-2\leq10\sqrt2$. Thus the computed result satisfies both comparison bounds.

Worked Example: A Signed Integrand on a Cylinder Patch

Use $\mathbf{r}(u,v)=(\cos u,\sin u,v)$, $0\leq u\leq\pi/2$, $0\leq v\leq1$, to estimate and calculate the integral of $f(x,y,z)=x-2$.

This parametrization covers the quarter-cylinder patch once. Its tangent vectors are $(-\sin u,\cos u,0)$ and $(0,0,1)$; their cross product has length $1$. Hence $dS=du\,dv$ and the patch area is $\pi/2$. Since $x=\cos u$ ranges from $0$ to $1$, we have $-2\leq f\leq-1$.

For a valid comparison using an area interval, take $A_-=1$ and $A_+=\pi/2$, so $1\leq\operatorname{Area}(S)\leq\pi/2$. The four endpoint products are $-2$, $-\pi$, $-1$, and $-\pi/2$. Therefore the signed-integrand rule gives $$ -\pi\leq\int_S f\,dS\leq-1. $$

Direct integration gives $$ \begin{aligned} \int_S f\,dS &=\int_0^1\int_0^{\pi/2}(\cos u-2)\,du\,dv\\ &=\int_0^1\left[\sin u-2u\right]_0^{\pi/2}\,dv\\ &=\int_0^1(1-\pi)\,dv =1-\pi. \end{aligned} $$ The antiderivative checks because $\frac{d}{du}(\sin u-2u)=\cos u-2$. Also, $-\pi\leq1-\pi$ since $0\leq1$, and $1-\pi\leq-1$ since $2\leq\pi$. The exact value lies inside the estimated interval.

A Common Mistake with Negative Values

In the cylinder example, a tempting but invalid shortcut is to pair the lower integrand bound with the lower area bound and the upper integrand bound with the upper area bound. It would give $-2(1)=-2$ as a proposed lower bound and $-1(\pi/2)=-\pi/2$ as a proposed upper bound. That would claim $$ -2\leq\int_S f\,dS\leq-\frac{\pi}{2}. $$ But the actual value $1-\pi$ is less than $-2$, since $\pi>3$. The estimate fails. The correction is to check all four products and take their minimum and maximum, giving the valid interval $[-\pi,-1]$.

Using Estimates as a Scale Check

A comparison estimate is not usually a replacement for a calculation. Its strength is that it can be obtained from simpler information: an area estimate and reliable extrema or bounds for the integrand. If an exact or numerical answer falls outside the resulting interval, at least one assumption, bound, area computation, or integral calculation must be wrong.

A useful estimate should be both valid and interpretable. Check the following before relying on it:

The bounds for $f$ hold on the entire surface, not just at a few sample points.
The area bounds apply to the geometric surface being integrated over.
A parameter-domain area formula is used only when the parametrization covers the surface once, except possibly on boundary sets of area zero.
If $f$ may be negative, all endpoint products are considered before identifying the lower and upper estimates.

Bounds can be loose and still be useful: they can catch a result with the wrong sign or an implausible magnitude. Sharper bounds give a narrower interval, but are only helpful if they remain rigorously justified.

Check Your Understanding

Use the comparison principles to reason through these questions.

If $3\leq f\leq5$ on a surface whose area is between $2$ and $4$, what bounds follow for $\int_S f\,dS$?
Why does integrating $f-m\geq0$ establish the lower comparison bound?
If $m\leq f\leq M$ includes negative values and area is uncertain, which products should be checked?
What can go wrong if a regular parametrization covers a surface twice but its area density is integrated over the full parameter domain?
Can a comparison interval verify that an integral is plausible without determining its exact value? Explain.