\(\newcommand{\R}{\mathbb R }\)
\(\Leftarrow\) \(\Uparrow\) \(\Rightarrow\)
The integral allows us to compute lengths of curves, as well as the values of functions taken as we move along a curve. In particular, these functions can be vector-valued, so that we can compute the effect of vectors (for example, from a magnetic or gravitational field) on an object moving along a path.
Note that the arclength of \(C\) equals \(\int_C 1\, ds\).
Let \(C\) be the curve parametrized by \(\mathbf x = \mathbf g(t) = \left(t^3 , \frac 32 t^4, \frac 6 5 t^5\right)\), for \(0\le t \le 2\). Compute the arclength of \(C\) and the integral \(\int_C xyz\, ds\)
Solution
We compute \[
\mathbf g'(t) = \left(3t^2, 6t^3, 6t^4\right) = 3t^2\left(1,2t, 2t^2\right)
\] and hence \[
|\mathbf g'(t)| = 3t^2\sqrt{1+ 4t^2 + 4t^4} = 3t^2\left(1+2t^2\right).
\] Thus \[
\text{arclength}(C) = \int_0^2 3t^2+ 6t^4\, dt = 2^3+ \frac 65 2^5
\] and
We have said that the definition is the arclength of the curve \(C\), but computing it depends on having a parameterization for \(C\). We should make sure that the integral of a real-valued function \(f\) over a curve \(C\) is independent of the parametrization. This implies in particular that the arclength of a curve \(C\) is independent of the parametrization.
Suppose that \(\mathbf g:[a,b]\to \R^n\) satisfies the requirements in the definition of arclength, and that \(\mathbf h:[c,d]\to \R^n\) is another parameterization of the same set \(C\), which also satisfies the requirements.
Then for any continuous \(f:\R^n\to \R\), computing \(\int_Cf\, ds\) using either parametrization \(\mathbf g\) or \(\mathbf h\) yields the same answer: \[ \int_C f\ ds = \int_a^b f(\mathbf g(t))|\mathbf g'(t)|\, dt = \int_c^d f(\mathbf h(u))|\mathbf h'(u)|\, du. \] In particular, \[ \text{arclength}(C) = \int_a^b |\mathbf g'(t)|\, dt = \int_c^d|\mathbf h'(u)|\, du. \]The proof uses Inverse Function Theorem, Chain Rule, and Change of Variables. First, we constuct a function \(\phi:[c,d]\to [a,b]\) such that \(\mathbf h=\mathbf g\circ\phi\). Since \(h\left([c,d]\right)= C=g\left([a,b]\right)\), we want to define \(\phi:(c,d)\to(a,b)\) by \(\phi(u)=t\), where \(g(t)=h(u)\). There is a unique such \(t\) if \(t \in(a,b)\), since both functions are injective on those sets. But we run into trouble if \(g(a)=g(b)=h(u^*)\). Even worse, \(\phi\) will not be continuous at this point, since the values less than \(u^*\) will be near one endpoint, while the values greater than \(u^*\) will be near the other. Fortunately, there can only be one such point, and we will be integrating, so it will not effect the result (the set of discontinuities has zero content). By Inverse Function Theorem, \(\phi\) is \(C^1\) on \((a,b)\), or on \((a,u^*)\cup(u^*,b)\). The values of \(\phi(c)\) and \(\phi(d)\) are determined by continuity, so we have a function which will change between any two parameterizations.
Writing \(\mathbf h\) as a function of a variable \(u\in [c,d]\), the Chain Rule implies that \[ \mathbf h'(u) = \mathbf g'(\phi(u)) \phi'(u), \] so \[ \int_c^d f(\mathbf h(u))|\mathbf h'(u)|\, du = \int_c^d f(\mathbf g(\phi(u)))\, |\mathbf g'(\phi(u))| \, |\phi'(u)|\, du. \] We define the new variable \(t = \phi(u)\). Since \(\phi:[c,d]\to [a,b]\) is a bijection, the Change of Variables Theorem implies that \[ \int_c^d f(\mathbf g(\phi(u)))\, |\mathbf g'(\phi(u))| \, |\phi'(u)|\, du. = \int_a^b f(\mathbf g(t)) |\mathbf g'(t)|\, dt, \] and this implies the conclusion of the theorem.This is nice to know, because if our definition of arclength depended on the parametrization, then it would not be a good definition. Arclength should be an intrinsic property of a curve, independent of how we choose to represent the curve.
It also means that it makes sense to describe a curve in a way that does not specifically involve a parametrization, and to compute either its arclength or a line integral \(\int_C f\, ds\). To carry out the computation, we can use any parametrization we like.
Let \(C\) be the ellipse \((x/a)^2+(y/b)^2 = 1\). Write down an integral formula for the arclength of \(C\).
Solution
The fact that arclength is independent of parametrization means that we can choose any paramtrization for \(C\), and we can use it to write down an integral for the arclength. For example, we can choose \[
\mathbf x = \mathbf g(t) = (a\cos t, b\sin t), \qquad 0\le t\le 2\pi.
\] Then \[
\mathbf g'(t) = (-a\sin t, b\cos t), \qquad
\text{ so }|\mathbf g'(t)| = \sqrt{a^2 \sin^2 t + b^2\cos^2 t}.
\] It follows that \[
\text{arclength}(C) = \int_0^{2\pi}\sqrt{a^2 \sin^2 t + b^2\cos^2 t}\, dt.
\] Unfortunately, this integral is impossible to evaluate using elementary functions. The elliptic integral (see the last Problem) was invented for this purpose by Giulio Fagnano (Italy, 1682-1786), also known for showing \(\pi=2i\log\left(\frac{1-i}{1+i}\right)\), and for solving an integration problem posed by Brook Taylor to “non-English mathematicians” as an attempt to discredit European mathematics.
Let \(C\) be a curve in \(\R^n\) parametrized by \(\mathbf x = \mathbf g(t)\) for \(a\le t\le b\).
If \(P =\{t_0\ldots, t_J\}\) is any partition of \([a,b]\), so that \(a=t_0 < t_1 < \ldots < t_J=b\), then the length of \(C\) may be approximated by the sum of the lengths of the line segments connecting \(\mathbf g(t_{j-1})\) to \(\mathbf g(t_j)\) for \(j=1,\ldots, J\), which we will denote \(\mathcal L_P(C)\), so that \[ \mathcal L_P(C) = \sum_{j=1}^J |\mathbf g(t_{j}) - \mathbf g(t_{j-1})|. \]The following theorem guarantees that this definition of arclength is consistant with the one we gave above.
The proof follows by applying the definition of the derivative of \(\mathbf g\) and showing that \(\forall \varepsilon>0,\) there is a partition \(P\) with \(U_P(|\mathbf g'|)-\varepsilon \leq \mathcal L(C) \leq L_P(|\mathbf g'|)+\varepsilon\).
We continue to suppose that \(C\) is a curve in \(\R^n\) parametrized by a function \[ \mathbf x = \mathbf g(t), \quad a\le t\le b, \] with the same assumptions about \(\mathbf g\) as above, e.g. \(\mathbf g'\) exists and is nonzero everywhere in an open interval containing \([a,b]\).
When we use the term “vector field” (rather than, say, “transformation”) we have in mind that for any \(\mathbf x\in S\), \(\mathbf F(\mathbf x)\) is a vector in the strict sense, that is, a quantity with a direction and magnitude, such as a force or a velocity. We want to think of \(\mathbf F\) as assigning an \(n-\)dimensional vector to every point in \(S\). Vector fields arise often in physics, precisely because there are many physical situations where vectors vary from point to point (for example, the force of gravity on an object rotating around the Earth).
The line integral of a vector field arises naturally in a variety of applications. For example, if a particle traverses a curve \(C\) parametrized by \(\mathbf x = \mathbf g(t)\), \(a\le t\le b\), and if there is a force (due for example to gravity, or electromagnetic forces) \(\mathbf F(\mathbf x)\) acting on the particle at position \(\mathbf x\), then \(\int_C \mathbf F\cdot d\mathbf x\) is the work done by the force field on the particle.
In the next example, we will show that the line integral of a vector field is not independent of the parametrization.
We have \(\mathbf g'(t) = (-\sin t, \cos t)\), and \(\mathbf F\circ \mathbf g(t) = (-\sin t, \cos t)\), so
\[ \int_0^{2\pi} \mathbf F(\mathbf g(t))\cdot \mathbf g'(t)\, dt = \int_0^{2\pi}(\sin^2t+\cos^2t)dt = 2\pi. \]
Similar computations show that
\[ \int_0^{2\pi} \mathbf F(\mathbf h(t))\cdot \mathbf h'(t)\, dt = -\int_0^{2\pi}(\sin^2t+\cos^2t)dt = -2\pi. \]
Thus, in this case, we get different answers from the two parametrizations.Line integrals of vector fields do have a certain invariant property - notice that the two integrals above differ only by a negative sign - and it depends on one aspect of the parameterization.
For a parametrized curve, we often speak of the orientation. For example, if \(C\) is a curve with endpoints \(\bf p\) and \(\bf q\) such that \(\bf p \ne q\), and if \(C\) is parametrized by \(\mathbf x = \mathbf g(t), a\le t\le b\), then the orientation depends on whether
Similarly, if \(C\) is a circle in the \(xy\) plane (or an ellipse, or the boundary of a connected set that does not “have any holes”), then the orientation depends on whether \(\mathbf g(t)\) traverses \(C\) in a clockwise or counterclockwise direction, as \(t\) increases from \(a\) to \(b\).
In either case, the curve has only two possible orientations.
In drawing pictures, the orientation is typically represented by an arrow which indicates the direction that \(\mathbf g(t)\) travels along the curve \(C\) as \(t\) increases.
The invariance property of line integral of vector fields is that the line integral of a vector field \(\mathbf F\) over a curve \(C\) depends on the orientation of \(C\) but is otherwise independent of the parametrization.
In fact, changing the orientation of \(C\) can only change the sign of the line integral \(\int_C \mathbf F\cdot d\mathbf x\).
This means that if we want to compute the line integral of a vector field, we need to specify the orientation of the curve. Once we have specified the orientation, we can speak unambiguously about \(\int_C \mathbf F\cdot d\mathbf x\), without having to specify a particular parametrization.
Suppose that \(C\) is a curve with two parameterizations, \(\mathbf g:[a,b]\to \R^n\) and \(h:[c,d]\to \R^n\), that both satisfy all the requirements in the definition of arclength, and such that there is a \(C^1\) function \(\phi:[c,d]\to[a,b]\) which is a bijection.
Then for any vector field \(\mathbf F\) that is continuous everywhere on \(C\), \[ \int_a^b \mathbf F(\mathbf g(t))\cdot \mathbf g'(t)\, dt =\begin{cases} \int_c^d\mathbf F(\mathbf h(u))\cdot \mathbf h'(u)\, du&\text{ if }\mathbf g\text{ and }\mathbf h\text{ have the same orientation}\\ -\int_c^d\mathbf F(\mathbf h(u))\cdot \mathbf h'(u)\, du&\text{ if }\mathbf g\text{ and }\mathbf h\text{ have the opposite orientation}. \end{cases} \]It is often useful to think of \(d\mathbf x\) as a vector \(d\mathbf x = (dx_1,\ldots,dx_n)\). Then given an oriented curve \(C\) and a vector field \(\mathbf F = (F_1,\ldots, F_n)\), we can write \[ \int_C \mathbf F\cdot d\mathbf x = \int_C F_1 \, dx_1+\ldots+ F_n dx_n. \] Here, for each \(j\), we define \[ \int_C F_j\, dx_j = \int_a^b F_j(\mathbf g(t)) \, g_j'(t)\, dt, \] where \(\mathbf x = \mathbf g(t)\), \(a\le t\le b\) is a parametrization of \(C\) with the right orientation. If we write \(\mathbf x(t)\) to indicate that \(\mathbf x = (x_1(t),\ldots, x_n(t))\) is a function of \(t\), then this looks nicer: \[ \int_C F_j\, dx_j = \int_a^b F_j(\mathbf x(t)) \, x_j'(t)\, dt. \]
We also sometimes write \(d\mathbf x = (dx, dy)\) in \(2\) dimensions, or \(d\mathbf x = (dx, dy, dz)\) in \(3\) dimensions.
Sometimes, as in the statement of Green’s Theorem (see the next section), it is traditional to write a vector field \(\mathbf F:\R^2\to \R^2\) with components \((P,Q)\), rather than \((F_1, F_2)\). Then the integrand \(\mathbf F\cdot d\mathbf x\) is rewritten as \(P\,dx+Q\,dy\), and \[ \int_C \mathbf F\cdot d\mathbf x \ = \ \int_C P\, dx + Q\, dy = \int_a^b \left[ P(\mathbf x(t))x'(t) + Q(\mathbf x(t))y'(t) \right] dt \] using the casual but suggestive notation \(\mathbf x(t) = (x(t),y(t))\), \(a\le t\le b\) for a parametrization of the curve with the right orientation.
Please note that with this notation, \(\int_C f \, ds\) and \(\int_C f\, dx\) mean very different things.
Solution
To evaluate the integral, we first find a parametrization that traverses \(C\) in a counterclockwise direction, say \(\mathbf g(t) = (\cos t, \sin t)\) for \(0\le t\le 2\pi\). Then it makes sense to write \(x(t)=\cos t\) and \(y(t)=\sin t\). So convert \(\int_C -y\,dx+x\,dy\) into a concrete integral, we just
Let \(C\) be the ellipse \(x^2+4y^2 = 16\), oriented counterclockwise. Compute \[ \int_C xy\,dx - x\,dy \]
Solution
We can parametrize \(C\) by \[
\mathbf x = (4\cos t, 2\sin t), \qquad 0\le t\le 2\pi.
\] We then make the replacements
Suppose that \(\psi:[a,b]\to \R\) is a \(C^1\) function. Compute \(\int_C y\, dx\), where \(C\) is the set \(\left\{(x, \psi(x)) : a\le x \le b\right\}\), oriented from \((b,\psi(b))\) to \((a,\psi(a))\).
Solution
The natural way to parametrize the curve is by \(\mathbf g(t) = (t, \psi(t))\) for \(a\le t\le b\). Unfortunately, this has the wrong orientation. To fix this, we have two options:
We could choose something like \(\mathbf g(t) = (a -t, \psi(a-t))\) for \(a-b \le t \le 0\). This parametrizes \(C\) and has the right orientation, but it seems unnatural and requires careful substitutions.
We could use the natural parametrization \(\mathbf g(t) = (t,\psi(t))\) for \(a\le t\le b\) with the wrong orientation, and then mutliply the result by \(-1\) to correct for the orientation. Theorem 3 guarantees that this will give the right answer. This leads to \[ \int_C y\, dx = -\int_a^b \psi(t)\, dt. \] (You can check that if you had opted for the more parametrization with the correct orientation, you would have gotten the same result.)
This is sometimes called the Fundamental Theorem of Calculus for Line Integrals of Vector Fields, but we prefer the Fundamental Theorem of Line Integrals.
Compute the arclength of the following curves.
Evaluate the following integrals.
In the integrals below, suppose that \(f\) is an extremely complicated function, for example \(f(x,y) =(2+3x) e^{x^2\cos(xy-3y^3)}\), or a similarly complicated function of more variables.
For each of the following curves, compute \(\int_C ds, \int_C dx\) and \(\int_C dy\).
Let \(C\) be a curve in \(\R^n\), and suppose that \(f:\R^n\to \R\) is a continuous function such that \(f(\mathbf x)=0\) for every \(\mathbf x\in C\).
Let \(C\) be an oriented curve in \(\R^n\), and suppose that \(\mathbf F:\R^n\to \R^n\) is a continuous vector field such that \(\mathbf F(\mathbf x)=\bf 0\) for every \(\mathbf x\in C\).
\(\Leftarrow\) \(\Uparrow\) \(\Rightarrow\)
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 2.0 Canada License.