Multivariable calculus and vector analysis
Introduced| English | Meaning |
|---|---|
| gradient/ˈɡreɪdɪənt/ | The vector of first partial derivatives of a scalar-valued function. |
| directional derivative/daɪˈrekʃənl dɪˈrɪvətɪv/ | The rate of change at a point per unit distance in a specified direction, when that limit exists. |
| Jacobian/dʒæˈkəʊbɪən/ | The determinant of the coordinate derivative matrix. |
One point, several ways to move
- On a hillside, the change in height depends on both where you stand and how you move. Different directions can share the same slope, and a flat surface gives zero slope in every direction. A directional rate also depends on whether movement is measured per unit distance or per unit time.
- GRE multivariable calculus connects the gradient 梯度, directional derivatives 方向导数, linear approximation, path derivatives and area changes. Keep the function, point, direction and coordinate region explicit before applying a formula.
Partials build the gradient
- A partial derivative varies one input while holding the others fixed. For $f(x,y)=x^2+3xy$, the results are $f_x=2x+3y$ and $f_y=3x$. Differentiating the other input as well would instead describe a change along a path.
- The gradient is the vector of first partial derivatives. For a two-variable function, $\nabla f=(f_x,f_y)$. At a point where $f$ is differentiable, this vector determines the first-order change in every direction; the partial derivatives alone do not guarantee differentiability.
$\dfrac{\partial}{\partial x}(x^2 y)$ equals
Difficulty 1/5 · 100 coins + a small bonus on first completion today
Treat $y$ as fixed: differentiating $x^2y$ with respect to $x$ gives $2xy$. Differentiating y as well would not be this partial derivative.
A direction needs a unit vector
- A directional derivative is the rate of change at a point per unit distance in a specified direction, when that limit exists. For a differentiable function and a unit vector $\mathbf u$, $D_{\mathbf u}f=\nabla f\cdot\mathbf u$. A nonunit velocity instead gives a rate with respect to its path parameter.
- If $\nabla f\ne0$, the greatest unit directional rate is $\|\nabla f\|$, achieved in direction $\nabla f/\|\nabla f\|$. The gradient is normal to a regular level curve there. If $\nabla f=0$, every unit directional derivative is zero; that does not make the function locally constant or select a unique steepest direction.
For $f=x^2+3y^2$ at $(1,1)$, $\nabla f=(2,6)$. The direction $(3,4)$ has length $5$, so use $\mathbf u=(3/5,4/5)$. The unit directional rate is $2(3/5)+6(4/5)=6$. The value $30$ would instead be the rate along a path whose velocity is $(3,4)$.
The gradient arrow is shortened for display; its direction represents the gradient.
For $f = x^2 + y^2$, find the directional derivative at $(1, 2)$ in the direction of $(3, 4)$.
Difficulty 2/5 · 140 coins + a small bonus on first completion today
$\nabla f = (2, 4)$, $\mathbf{u} = (0.6, 0.8)$: $1.2 + 3.2 = 4.4$.
If f is differentiable at a point and its gradient there is zero, what follows?
Difficulty 1/5 · 100 coins + a small bonus on first completion today
The dot product with the zero gradient is zero for every unit direction. Higher-order changes may still occur, so local constancy does not follow.
Differentiability is one linear approximation
- At an interior point $\mathbf a$, differentiability means one linear map $L$ approximates all small changes. The remainder $R$ must be small compared with the size of the change:
- For a differentiable scalar function, $L(\mathbf h)=\nabla f(\mathbf a)\cdot\mathbf h$. Existing partial derivatives, or even existing directional derivatives in every direction, do not by themselves establish this common linear approximation. Continuous first partial derivatives in a neighbourhood are a useful sufficient condition.
Work through a partial-derivative trap
- Define $g(0,0)=0$ and $g(x,y)=xy/(x^2+y^2)$ elsewhere. Along either coordinate axis, $g$ is zero, so both partial derivatives at the origin exist and equal zero.
- Along the diagonal $(x,y)=(t,t)$ with $t\ne0$, $g(t,t)=1/2$. The function therefore fails to approach its assigned value zero and is not continuous at the origin. Differentiability implies continuity, so these partial derivatives cannot establish differentiability.
What does this evidence establish?
At an interior point of an open domain in R², distinguish sufficient conditions from weaker information.
For $g(x,y)=xy/(x^2+y^2)$ away from the origin, what is $g(t,t)$ for nonzero t?
Difficulty 2/5 · 140 coins + a small bonus on first completion today
The numerator is t² and the denominator is 2t², giving 1/2. This conflicts with an assigned origin value of zero when checking continuity.
Let g be zero at the origin and $xy/(x^2+y^2)$ elsewhere. Explain why both partials at the origin are zero but differentiability fails.
Along either axis g is identically zero, so both coordinate difference quotients are zero. Along the nonzero diagonal g(t,t)=1/2, which does not approach g(0,0)=0. Thus g is not continuous at the origin; differentiability would imply continuity. The two partials alone are insufficient.
A path changes both inputs
- If $f$ is differentiable at the path point and $x(t),y(t)$ are differentiable at the parameter value, put $F(t)=f(x(t),y(t))$. Evaluate the partials at that path point; then
- For $f=x^2+3y^2$, $x=t^2$ and $y=t$, substitute into the partials first: $f_x=2t^2$ and $f_y=6t$. The path derivative is $(2t^2)(2t)+(6t)(1)=4t^3+6t$, giving $10$ at $t=1$. Directly differentiating $t^4+3t^2$ confirms it.
For $f=x^2+3y^2$, $x=t^2$ and $y=t$, find the path derivative at $t=1$.
Difficulty 2/5 · 140 coins + a small bonus on first completion today
The composite is $t^4+3t^2$; its derivative is $4t^3+6t$, giving 10.
A tangent plane is a local model
- For differentiable $f$, the tangent plane at $(a,b)$ describes a local first-order approximation, not automatically the whole graph:
- For $f=x^2+3y^2$ at $(1,1)$, this plane is $z=4+2(x-1)+6(y-1)$. The actual graph remains curved: its first partial derivatives vary from point to point, and its second partials include $f_{xx}=2$ and $f_{yy}=6$.
What forces the whole graph to be a plane?
- If $f_x=a$ and $f_y=b$ everywhere on $\mathbb R^2$, integrating along coordinate lines gives $f=ax+by+c$, an affine function. The domain matters: on disconnected open pieces, the additive constant could differ between pieces.
- Weaker conditions need counterexamples. The surface $x^2+y^2$ has $f_{xy}=0$ but is curved. For $f=e^x$, every nonempty level set is a vertical line, yet the graph is not a plane. Even $f_{xx}=f_{yy}=0$ is insufficient: $f=xy$ has a nonzero mixed partial.
Does it force a plane?
Assume f is smooth on all of R². Distinguish conditions forcing an affine graph from weaker conditions.
If $f_{xy} = 0$ everywhere on $\mathbb{R}^2$, the graph of $f$ is a plane.
Difficulty 1/5 · 100 coins + a small bonus on first completion today
$x^2 + y^2$ has zero mixed partial.
Explain why neither $f_{xy}=0$ nor $f_{xx}=f_{yy}=0$ on all of R² forces an affine graph. Give a smooth counterexample for each.
For the first claim, f=x²+y² has mixed partial zero but is curved. For the second, f=xy has both pure second partials zero but mixed partial one and is not affine. Vanishing of the stated subset of derivatives is weaker than constancy of the whole first derivative.
Describe the integration region twice
- For the triangle $D=\{(x,y):0\le y\le x\le1\}$, a vertical slice fixes $x$ and lets $y$ run from $0$ to $x$. A horizontal slice fixes $y$ and lets $x$ run from $y$ to $1$. For a continuous integrand on this triangle, both iterated descriptions give the same double integral.
- Using the constant integrand $1$ shows the area in both orders:
Which reverses the order of $\int_0^1\int_0^x1\,dy\,dx$ over the same triangle?
Difficulty 1/5 · 100 coins + a small bonus on first completion today
For a fixed y between 0 and 1, the region has x between y and 1. Both orders describe $0\le y\le x\le1$.
Coordinate changes need area and coverage
- Here Jacobian 雅可比行列式 means the determinant of the coordinate derivative matrix. The matrix itself is often called the Jacobian matrix. For the map $T:(u,v)\mapsto(x,y)$, the area scale is the absolute value $|\det DT|$, while the signed determinant also records orientation.
- For continuous integrands on bounded regions with piecewise smooth boundaries, a standard sufficient setup uses a continuously differentiable coordinate map that is one-to-one with nonzero determinant on the interior. Boundary exceptions need checking separately. Transform the integrand and limits as well as the area factor. For $x=2u$, $y=-3v$, the determinant is $-6$, but the area factor is $6$.
The map $x=2u$, $y=-3v$ has determinant -6. What positive area factor belongs in a double-integral substitution?
Difficulty 2/5 · 140 coins + a small bonus on first completion today
The absolute determinant is 6. The negative sign concerns orientation rather than a negative area.
Try $x=u^2$, $y=v$ on $-1\le u\le1$, $0\le v\le1$. Its image is the unit square. Integrating $|2u|$ over the parameter rectangle gives 2, whereas the square has area 1. Explain the mismatch and repair the parameter region.
Every image point with x>0 has two preimages, one positive and one negative u. The full rectangle counts the square twice; taking an absolute determinant does not repair noninjectivity. Restrict to 0<=u<=1, 0<=v<=1. The interior map is one-to-one with positive determinant; the zero determinant at its boundary u=0 is an area-zero exception. The integral of 2u over this branch is 1.
Polar coordinates: transform both factors
- With $x=r\cos\theta$, $y=r\sin\theta$ and $r\ge0$, the determinant is $r$, so $dA=r\,dr\,d\theta$. A full angular turn covers the disc once apart from its boundary seam and the origin; these area-zero exceptions do not change the usual area integral. The disc of radius $2$ has area $4\pi$.
- On the unit disc, use $x^2+y^2=r^2$. The area element supplies another factor $r$, making the full integrand $r^3$. Thus
The area element in polar coordinates is
Difficulty 1/5 · 100 coins + a small bonus on first completion today
With radius r nonnegative, the determinant and absolute area factor of the standard polar map are r.
Evaluate $\iint_D (x^2 + y^2)\,dA$ over the unit disc $D$, to three decimal places.
Difficulty 2/5 · 140 coins + a small bonus on first completion today
$\int_0^{2\pi}\int_0^1 r^3\,dr\,d\theta = \pi/2$.
Check the hypothesis before the calculation
- Use unit directions for rates per unit distance, and distinguish partials from a full linear approximation. State the differentiability and path conditions before applying the chain rule; distinguish a local tangent plane from an affine graph everywhere.
- Draw the region before changing integration order. Under a valid coordinate change, transform the function, the bounds and the absolute area factor; check coverage so that the image is not counted more than once.
For $f=x^2+3y^2$ at (1,1), a student calls 30 the unit directional rate along (3,4), the maximum directional rate, and the slope of a plane equal to f everywhere. Correct all three claims.
The gradient is (2,6). Normalising (3,4) gives (3/5,4/5), with directional rate 6; 30 instead uses velocity (3,4). The maximum unit directional rate is sqrt(40), achieved along (2,6)/sqrt(40), rather than necessarily along (3,4). The plane z=4+2(x-1)+6(y-1) is the local tangent model, while f remains quadratic and curved.