Why the Surface Is a Choice
Stokes’ theorem relates the circulation of a vector field around a closed curve to the flux of its curl through a surface bounded by that curve. The boundary curve is fixed by the problem, but the spanning surface often is not. A useful choice can turn a difficult surface integral into a simple one; a careless choice can make the calculation harder or invalidate the theorem.
The best candidate is not automatically the flattest or most familiar surface. Compare the curl integrand, the region of integration, the surface’s smoothness, and the orientation it induces on the boundary. A surface is useful only if all the theorem’s hypotheses are satisfied.
What to Compare
For each plausible surface, assess the actual integral Stokes’ theorem would require. If a surface is the graph \(z=g(x,y)\), oriented upward, its oriented area element is $$ \mathbf{n}\,dS=(-g_x,-g_y,1)\,dx\,dy. $$ This often makes a graph convenient: the flux integral is over a planar region, and the dot product with the curl can be inspected before carrying out any integration. For a horizontal plane oriented upward, the area element is simply \((0,0,1)\,dA\).
A practical comparison includes four questions:
- Integrand: Does the curl have a simple dot product with the candidate’s oriented normal?
- Bounds: Is the projected region easy to describe and integrate over?
- Validity: Is the surface smooth or piecewise smooth where needed, and is \(\mathbf{F}\) continuously differentiable on an open set containing it?
- Orientation: Does the chosen normal induce the direction of traversal specified for \(C\)?
Once a candidate passes these checks, it is legitimate to use it. If a second candidate also satisfies the hypotheses and has the same oriented boundary, Stokes’ theorem says both surface integrals equal the same circulation. This does not mean their integrands or calculations must look alike.
A Graph Surface Formula
The graph formula follows directly from a parameterization, and it explains why graph surfaces are often easy to compare. Parameterize \(z=g(x,y)\) by \(\mathbf{r}(x,y)=(x,y,g(x,y))\). Its tangent vectors are \(\mathbf{r}_x=(1,0,g_x)\) and \(\mathbf{r}_y=(0,1,g_y)\). Their cross product points upward because its third component is positive.
For a vector field \(\mathbf{G}\), the oriented flux through this graph is therefore $$ \iint_S \mathbf{G}\cdot\mathbf{n}\,dS = \iint_D \mathbf{G}(x,y,g(x,y))\cdot(-g_x,-g_y,1)\,dx\,dy, $$ where \(D\) is the graph’s projection into the \(xy\)-plane. The formula requires a differentiable graph over the region \(D\); its upward orientation is fixed by the positive third component. For Stokes’ theorem, this flux formula is applied to \(\mathbf{G}=\nabla\times\mathbf{F}\).
Worked Example: A Circular Boundary and Two Caps
Worked Example: Choose Between a Disk and a Paraboloid Cap
Let \(C\) be the circle \(x^2+y^2=4,\ z=0\), traversed counterclockwise when viewed from above, and let \(\mathbf{F}=(-y/2,x/2,0)\). Compare the flat disk with the paraboloid cap \(z=4-x^2-y^2\).
First compute the curl: $$ \nabla\times\mathbf{F} = \left( \frac{\partial 0}{\partial y}-\frac{\partial (x/2)}{\partial z}, \frac{\partial (-y/2)}{\partial z}-\frac{\partial 0}{\partial x}, \frac{\partial (x/2)}{\partial x}-\frac{\partial (-y/2)}{\partial y} \right) =(0,0,1). $$
Flat disk. Its upward unit normal is \((0,0,1)\), so the integrand is \(1\). The disk has radius \(2\), hence its area is \(\pi(2)^2=4\pi\), and $$ \iint_S(\nabla\times\mathbf{F})\cdot\mathbf{n}\,dS = \iint_S 1\,dS = 4\pi. $$ The plane disk is smooth and \(\mathbf{F}\) is smooth everywhere, so this is a valid and particularly simple choice.
Paraboloid cap. Its projection is the same disk \(D=\{(x,y):x^2+y^2\leq 4\}\). Here \(g_x=-2x\) and \(g_y=-2y\), so the upward oriented area element is \((2x,2y,1)\,dx\,dy\). Its dot product with the curl is \(1\), giving $$ \iint_D 1\,dx\,dy = \int_0^{2\pi}\int_0^2 r\,dr\,d\theta = \int_0^{2\pi}2\,d\theta = 4\pi. $$ The cap is smooth, and at \(x^2+y^2=4\) it has \(z=0\), so its boundary is \(C\). The upward normal gives the required counterclockwise boundary orientation.
As a check independent of the surface integral, parameterize the boundary by \(\mathbf{r}(t)=(2\cos t,2\sin t,0)\) for \(0\leq t\leq 2\pi\). Then \(\mathbf{F}(\mathbf{r}(t))=(-\sin t,\cos t,0)\), \(\mathbf{r}'(t)=(-2\sin t,2\cos t,0)\), and their dot product is \(2\). Thus \(\int_0^{2\pi}2\,dt=4\pi\), as required. Both surfaces work, but the flat disk makes the calculation immediate.
Worked Example: Check the Curl Before Comparing Caps
Worked Example: A Square and a Curved Graph
Let \(C\) be the boundary of the square \(-1\leq x\leq1,\ -1\leq y\leq1,\ z=0\), traversed counterclockwise from above. Let \(\mathbf{F}=(-y/2,x/2,x)\). Compare the flat square with the graph \(z=g(x,y)=(1-x^2)(1-y^2)\), which meets the plane \(z=0\) along the square’s boundary.
Computing each curl component from its definition gives $$ \nabla\times\mathbf{F} = \left( \frac{\partial x}{\partial y}-\frac{\partial (x/2)}{\partial z}, \frac{\partial (-y/2)}{\partial z}-\frac{\partial x}{\partial x}, \frac{\partial (x/2)}{\partial x}-\frac{\partial (-y/2)}{\partial y} \right) =(0,-1,1). $$ In particular, the second component is \(0-1=-1\).
Flat square. The upward normal is \((0,0,1)\), so the curl-flux integrand is \(1\). The square has area \(2\cdot2=4\), and the surface integral is \(4\).
Curved graph. Its partial derivative with respect to \(y\) is \(g_y=-2y(1-x^2)\). The upward oriented area element is \((-g_x,-g_y,1)\,dx\,dy\), so the curl dot this vector is $$ (0,-1,1)\cdot(-g_x,-g_y,1) = g_y+1 = 1-2y(1-x^2). $$ The graph is smooth over the square. Its height vanishes on each edge because either \(1-x^2=0\) or \(1-y^2=0\), so its boundary is exactly \(C\). Its upward orientation induces the required traversal. Thus $$ \begin{aligned} \iint_S(\nabla\times\mathbf{F})\cdot\mathbf{n}\,dS &=\int_{-1}^1\int_{-1}^1\left(1-2y(1-x^2)\right)\,dy\,dx\\ &=4-2\int_{-1}^1(1-x^2)\,dx\int_{-1}^1y\,dy\\ &=4-2\left(\frac{4}{3}\right)(0)=4. \end{aligned} $$ The \(y\)-dependent term integrates to zero because it is odd in \(y\) over symmetric bounds.
Check against the boundary integral, where \(z=0\) and \(dz=0\), so only \((-y/2)\,dx+(x/2)\,dy\) contributes. On the bottom edge, \(y=-1\), \(x\) goes from \(-1\) to \(1\), and the contribution is \(1\). On the right edge, \(x=1\), \(y\) goes from \(-1\) to \(1\), and the contribution is \(1\). On the top edge, \(y=1\), \(x\) goes from \(1\) to \(-1\), and the contribution is \(1\). On the left edge, \(x=-1\), \(y\) goes from \(1\) to \(-1\), and the contribution is \(1\). The total is \(1+1+1+1=4\), agreeing with both surface integrals.
A Sign Pitfall: Use the Correct Curl Component
The square example also shows why the curl must be checked before substituting into a surface formula. A common error is to compute its second component as \(+1\). The definition is $$ (\nabla\times\mathbf{F})_2 = \frac{\partial F_1}{\partial z} - \frac{\partial F_3}{\partial x}. $$ Here that is \(0-1=-1\), not \(+1\). With the incorrect curl \((0,1,1)\), the graph dot product would be written \(1-g_y\). The correct curl \((0,-1,1)\) gives \(1+g_y=1-2y(1-x^2)\). The total happens to remain \(4\) because the \(g_y\) term integrates to zero on this symmetric square, but the incorrect pointwise integrand is still wrong. A coincidentally unchanged final value is not a valid check of intermediate work.
Worked Example: A Singular Field Rules Out the Flat Disk
Worked Example: Use a Hemisphere to Avoid a Singularity
Let \(C\) be the unit circle \(x^2+y^2=1,\ z=0\), traversed counterclockwise from above. Consider $$ \mathbf{F}(x,y,z)=\left(-\frac{y}{s},\frac{x}{s},0\right), \qquad s=x^2+y^2+z^2. $$ This field is defined and continuously differentiable away from the origin. The flat disk would be an easy surface geometrically, but it contains the singularity at the origin. It cannot be used in Stokes’ theorem for this field.
Instead, take the upper unit hemisphere \(S\), whose outward normal gives the specified boundary orientation. Differentiating the field where \(s\ne0\) gives $$ \nabla\times\mathbf{F} = \left(\frac{2xz}{s^2},\frac{2yz}{s^2},\frac{2z^2}{s^2}\right) = \frac{2z}{s^2}(x,y,z). $$ The hemisphere does not contain the origin, and the field is smooth on an open neighborhood of it, so this candidate is valid.
On the unit sphere, \(s=1\) and the outward unit normal is \((x,y,z)\). Consequently, the curl dot the unit normal is \(2z(x^2+y^2+z^2)=2z\). Use spherical coordinates \(x=\sin\theta\cos\phi,\ y=\sin\theta\sin\phi,\ z=\cos\theta\), with \(0\leq\theta\leq\pi/2\) and \(0\leq\phi\leq2\pi\). The sphere’s area element is \(dS=\sin\theta\,d\theta\,d\phi\). Therefore $$ \iint_S(\nabla\times\mathbf{F})\cdot\mathbf{n}\,dS = \int_0^{2\pi}\int_0^{\pi/2} 2\cos\theta\sin\theta\,d\theta\,d\phi = \int_0^{2\pi}1\,d\phi = 2\pi. $$
Check the result directly on \(C\). With \(\mathbf{r}(t)=(\cos t,\sin t,0)\), \(0\leq t\leq2\pi\), the field is \((-\sin t,\cos t,0)\) and \(\mathbf{r}'(t)=(-\sin t,\cos t,0)\). Their dot product is \(1\), so the circulation is \(\int_0^{2\pi}1\,dt=2\pi\). The hemisphere succeeds not because it is simpler than the disk in shape, but because it avoids the singularity that makes the disk invalid.
A Reliable Selection Process
When several surfaces appear possible, make the decision in a fixed order. Testing validity before comparing computational convenience avoids investing effort in a surface Stokes’ theorem cannot use.
Record the curve and its direction of travel. Any candidate must have that boundary orientation.
Confirm that the surface is suitably smooth and that the field has continuous first partial derivatives on an open set containing the whole surface.
Compute or simplify the curl, form its dot product with each candidate’s oriented normal, and inspect the region of integration.
Use a valid candidate with manageable calculations. When practical, compare the result with a boundary integral or another valid spanning surface.
Why Surface Selection Matters
Stokes’ theorem permits a change of surface, not a change of the boundary curve or its orientation. A good choice can simplify the normal, reduce the integrand, or replace awkward bounds with familiar ones. But convenience comes after validity: a flat disk through a singularity is not rescued by being easy to parameterize, and a surface with the wrong induced boundary direction gives the negative of the desired circulation.
Keep the curl calculation, the normal, and the orientation linked. In particular, for an upward graph the oriented area element is \((-g_x,-g_y,1)\,dx\,dy\), not \((g_x,g_y,1)\,dx\,dy\). Writing this vector explicitly makes sign errors easier to catch and makes candidate surfaces easier to compare.
Check Your Understanding
For each question, consider both whether a candidate surface is valid and whether it is computationally convenient.
- What must be true about the field on a neighborhood of a candidate spanning surface before Stokes’ theorem can be applied?
- For an upward-oriented graph \(z=g(x,y)\), what is the oriented area element, and why is its third component positive?
- Why can a flat disk be invalid for a field that is smooth everywhere except at the disk’s center, even when its boundary avoids that point?
- In the square example, compute the second component of the curl from its defining partial derivatives. What is the resulting graph integrand?
- If two valid oriented surfaces share the same oriented boundary, what does Stokes’ theorem say about their curl-flux integrals?