Skip to main content
Loading Explore guides…
Multivariable Calculus · Acedra
Home
Explore
Mathematics
Multivariable Calculus
Sign in to save
Chat
If YouTube asks you to sign in, open the video on YouTube.
Watch on YouTube
Reload player
Transcript
76 segments
Copy
Hide video
0:00
[Music] In single variable calculus, you mastered the 2D world of flatland. Think about it. You lived on a curve, a simple predictable line where y equals f ofx. Everything was in a single plane. Your only questions were well pretty straightforward. First, how steep is it? Which you found with dydx. And second, what's the area underneath? which of course was the integral of f ofx dx. That was your entire world. Today we're leaving flatland. We are entering spaceand. This is the world of
0:39
surfaces where z equals f ofxy. A world of three dimensions of mountains and valleys. And here's the most important thing to understand right from the start. This isn't just adding a variable. It's a fundamental shift in perspective. The questions become so much richer. For instance, instead of just asking how steep is the slope, we get to ask which direction is the steepest. If you're standing on a mountain side, you can go down or up or sideways. Instead of finding a maximum
1:15
on a simple curve, we ask where is the peak of this entire mountain? And instead of calculating a flat area, we ask, what is the volume under this surface? [Music] Which brings me to my promise to you. Give me 50 minutes and I will make multivariable calculus. Click. You will see that the tools you already know, they don't break. You don't have to throw anything away. No, no, they evolve. The simple derivative becomes a vector. We call it the gradient or grad
1:54
f. It points the way uphill. The second derivative becomes a matrix. We call it the hessen represented by h. It tells you the shape of the surface you're on. And that brings us to the multivariable calculus toolkit. Now take a look at this. I know it seems like a lot, but by the end of this hour, you will understand every single one of these tools. You are about to master them. Let's begin. All right. The first and most important tool you'll need in these higher
2:29
dimensions isn't a rope or an ice axe. It's a compass. But this is no ordinary compass. It's a compass that doesn't point north, but points straight up the steepest part of the mountain. That compass is called the gradient. Let's make this real. Here's our motivating problem for today. The hill. Imagine you are standing on a hill described by this height function. H ofxy = 100 - x^2 - 2 y^2. You're at the specific point where x= 1 and y = 1. If you plug that in,
3:04
your height h of 111 is 97. Got it? You're standing right there. Here are the two critical questions. First, from this exact spot, which direction is the steepest way up? And second, just how steep is it? What's the slope of that path? Now, here's the challenge. In single variable calculus, the slope was just one number. Easy. Here, it's ambiguous. The slope depends entirely on which way you decide to walk. So, how do we solve this? The core idea is slicing the mountain. Let's
3:42
simplify. Let's forget about all the other directions and only walk in the x direction. Think of it as walking due east. This means we freeze the y variable. We lock it in at y = 1. Our original function was h ofxy = 100 - x^2 - 2 y^2. If we freeze y at 1, it becomes h of x1 = 100 - x^2 - 2 * 1^ 2, which simplifies to h slice of x = 98 - x^2. And just look at that. Our complex 3D surface problem has collapsed into a simple 2D curve problem. And you you already know how to solve this. Let's do
4:25
it. The function of our slice is f slice of x = 98 - x^2. The derivative of that slice frime slice of x is just -2x. So the slope at our point where x = 1 is frime slice of 1, which is -2 * 1 = -2. There it is. The slope of our hill specifically in the x direction is -2. This process, freezing all the other variables to turn a complex function into a simple one and then taking a regular old derivative is called a partial derivative. And the shortcut is actually super simple. When taking the
5:08
derivative with respect to one variable, you just treat all other variables as if they were constants. That's the secret. Let's try it. Let's find the partial derivative with respect to x which we write as that curly dh / dx. Our function is h ofxy = 100 - x^2 - 2 y^2. Okay, let's go term by term. As you can see in this table, the derivative of 100 with respect to x 0. It's a constant. The derivative of - x^2, that's our variable, so it's - 2x. And the derivative of - 2 y^2.
5:50
This is the key idea. We treat y as a constant. So the derivative of that whole term is 0. So partial h over partial x is just 0 - 2x - 0 which is - 2x. At our point 1 1, the slope is - 2 * 1, which is -2. It matches perfectly. Okay. Okay, let's do the same thing for y. Partial h over partial y, we treat x as the constant this time. Looking at the table again, the derivative of 100 zero. The derivative of - x^2. Ah, we're treating x as a constant now, so its derivative is also zero. And the
6:33
derivative of - 2 y^2, that's our variable, so it's -4 y. Combine them and partial h over partial y is simply -4 y. At our point 1, one, the slope in the y direction is -4 * 1, which is -4. All right, it is time for the gradients. Assembling your compass. We did the hard work. We found the two fundamental slopes at our 111. The slope in the x direction is -2. The slope in the y direction is -4. The gradient packages these into a single powerful vector. We write it as Nabla H or DellH. The
7:17
formula, as you can see, is simply a vector of the partial derivatives. Naba F ofXY is the vector containing partial f over partial X and partial f over partial Y. So for our point the gradient of h at 1 one is the vector -2 comma minus4. This one vector this is the satisfying payoff. It instantly and perfectly answers our two original questions. Answer number one. Which direction is steepest? At our point 11 one. The gradient vector -2 -4 points exactly in the direction of
7:58
steepest ascent. It's our compass needle. And answer number two, how steep is it? The magnitude or the length of the gradient vector tells you the slope. So the steepest slope is the magnitude of the vector -2 - 4. We use the Pythagorean theorem. It's the square<unk> of -2 2 + -4 2 which is the<unk> of 4 + 16 which is the square<unk> of 20 which is approximately 4.47. The gradient is your compass and your slope meter allin one. All right, let's step up to the next level. In the last
8:36
chapter, we saw that the gradient is the perfect derivative for a function with multiple inputs but only one output. Think about our mountain example. height equals H of XY, one single number for height. But what if we have multiple inputs and multiple outputs? Think of a robot arm. You send it two inputs, let's say motor angles theta 1 and theta 2, and you get two outputs, the hands x and y coordinates. How in the world do you take the derivative of that? The answer is that the derivative
9:13
must evolve again. It's no longer a number. It's no longer a vector. The derivative becomes a matrix. And that matrix is the Jacobian. Let's look at a motivating problem to make this really clear. The factory. Your inputs are X units of steel and Y units of plastic. Your outputs are U cars and V toys. We have two simple production functions. For cars, u ofxy = 2x + 0.5 y. And for toys, v ofxy = 0.1x + 3 y. This is really just a single vector function f that maps an input vector xy to an
9:57
output vector uv. Now for the critical question. If we get one extra unit of steel X, how does our production of both cars and toys change simultaneously? We can build our intuition by creating an organized table of sensitivities. We actually know how to calculate all the pieces. First, let's look at steel's effect on cars. The partial derivative of u with respect to x is simply two. That means one extra unit of steel gives us two more cars. Simple enough. Next, plastic's effect on cars. The
10:33
partial derivative of U with respect to Y is 0.5. So, one extra plastic gives us half a car. Okay. Now, for the toys, steel's effect on toys. The partial derivative of V with respect to X is just 0.1. So, one extra steel gives us 0.1 more toys. Not very much. And finally, plastic's effect on toys. The partial to V with respect to Y is three. One extra plastic means three more toys. Makes sense. Now for the formalization, assembling the Jacobian matrix. Here's the cool part. The Jacobian is
11:12
just these four values we calculated organized in a standard logical way. Row one is all about the first output, cars. Row two is all about the second output, toys. Column one is all about the first input, steel. and column 2 is all about the second input plastic. So as you can see in the formula the Jacobian J is the matrix with 2 and 0.5 in the first row and 0.1 and 3 in the second row. That's it. This matrix is the derivative of our vector function. We call it the universal translator.
11:48
Why? Because the Jacobian translates a small change in your inputs like delta x and delta y into the resulting change in your outputs delta u and delta v. The relationship is a linear approximation as shown in the formula. The change in outputs vector is approximately the jacobian matrix times the change in inputs vector. Let's put this universal translator into action. Let's say we get a new shipment. 10 extra units of steel. So delta x is 10. And five extra units
12:18
of plastic. So delta y is five. How does our production change? We just multiply our Jacobian matrix by the input change vector 10 5. Let's do the math. The change in cars deltaU is approximately 2 * 10 + 0.5 * 5. That's 20 + 2.5. Giving us 22.5 more cars. And the change in toys, delta v, is 0.1 * 10 + 3 * 5. That's 1 + 15 for 16 more toys. It works perfectly. So, you might be thinking, okay, a neat way to organize derivatives. Why do I really care? Here's why. The true power is revealed
13:06
when we chain functions together. Let's add one more layer to our factory, profit. Let's say profit P is a function of the cars and toys we make. P of UV = 3,00 U + 500V. So now we have a chain of functions. As you can see in this diagram, our inputs X and Y lead to our products U and V, which in turn lead to our final profit P. Following the arrows, the big question now is how does our final profit P change if we get one more unit of steel X. This is the pornoariat chain
13:45
rule. The derivative of the entire chain is the product of the derivative matrices, the Jacobians of each step. The total Jacobian is the Jacobian for profit times the Jacobian for the products. You can see the full formula here. A row vector of the partials of P with respect to X and Y is equal to the Jacobian of the profit function times the Jacobian of the production function. It's matrix multiplication all the way down. Let's calculate just one part of this partial P over partial X. How does
14:22
profit change with steel? The formula shows us it's the partial of P with respect to U time the partial of U with respect to X plus the partial of P with respect to V times the partial of V with respect to X. We already know all these numbers. The partials of P are 3,500 and the partials with respect to X from our factory are 2 and 0.1. So, we get 3,000 * 2 + 500 * 0.1. That's 6,000 + 50, which equals $6,050. $6,050. Now, for the connection that makes it all click. This exact process using the
15:09
chain rule to see how a final output like profit or more importantly error is affected by an initial input by multiplying Jacobians backward through a long chain of functions is the algorithm of back propagation. A deep neural network is a giant composition of functions. We call them layers. When it makes a mistake, back propagation uses the chain rule to compute the gradient with respect to every single weight. It tells every single parameter how to change to make
15:44
the final error smaller. The Jacobian isn't just a mathematical curiosity. It's the engine driving modern artificial intelligence. All right, let's talk about the next piece of the puzzle, the Hessen. So far the gradient told us which way is up and the Jacobian gave us a linear map. These are the multivariable versions of the first derivative. They give you a flat tangent plane approximation of your function. But to really understand the landscape, you need to know more than
16:15
just the slope. You need to know the shape of the land. Is it curving up like a valley floor or is it curving down like the top of a hill? For that we need the second derivative. And in multiple dimensions, the second derivative is not just a number. It's a matrix. It is the hessen. So let's connect this back to what you already know. From concavity to curvature. Remember from single variable calculus, the second derivative fpime of x tells you about a curve's concavity.
16:47
If fprime of x is greater than zero, it's concave up. You know, shaped like a cup. It holds water. If fp prime of x is less than zero, it's concave down, shaped like a frown. It spills the water. Simple enough. But in 3D, a surface can curve differently in different directions. Think of a Pringle chip. It curves up along its short axis, but it curves down along its long axis. A single number just can't describe that. And this brings us to the motivating mystery, the flat spot.
17:21
Imagine you're hiking and you find a perfectly flat spot. It's peaceful. You pull out your gradient compass, the one we built in the first chapter, and it just spins. The gradient is zero, a vector of 0 0. This flat spot, this critical point could be one of three things. One, a local minimum. You're at the bottom of a valley. Two, a local maximum. You're at the peak of a mountain. Or three, a saddle point. You're in a mountain pass. The gradient is useless here. It's zero.
17:51
It gives us no information. To solve this mystery, we can't look at the slope. We have to look at how the surface curves. We need the hessen. Okay, let's formalize this. How do we assemble this matrix of curvature? The hessen matrix H is the matrix of all the second order partial derivatives. You can see the formula here. It's a 2x2 matrix. The element in the top left fxx is the second partial of f with respect to x twice. The bottom right fuby y is the second partial with respect to y
18:23
twice. And the off diagonals f_sub_xy and f_sub_yx are the mixed partials. So fubxx tells you the concavity if you're looking along the x direction. And fub yy tells you the concavity if you're looking along the y direction. And those mixed partials fxy and fyx, they measure the twist of the surface. It's pretty cool. Let's see this in action. Example one, the valley which is a local minimum. Our function is f ofxy = x^2 + y^2 right at the critical point of 0 0. Okay. First partials fubx is 2x
18:58
fuby is 2 y. Simple. Now the second partials fubxx is 2 fuby y is 2 and the mixed partial fubxy is zero. So the hesshen matrix at our critical point is just a matrix with two on the diagonal and zero everywhere else. Both diagonal entries are positive. This matrix screams valley. It's curving up in both directions. All right, next one. Example two. The peak our local maximum. The function is f ofxy =x^2 - y^2. Same process. First partials fubx is -2x. fuby is -2 y. second partials fubxx is
19:40
-2 fuby yy is -2 and the mixed partial is again zero. So the hessen is a diagonal matrix with -2 on the diagonal. Both diagonal entries are negative. This matrix screams peak. It's curving down in both directions. And now for the interesting one. Example three. The saddle. The function is f ofxy = x^2 - y^2. First partials fubx is 2x but fuby is -2 y. Second partials fubxx is 2 but fuby y is -2. The mixed partial is still zero. So what's the hessen at 0 0? It's a diagonal matrix with 2 and -2 on
20:30
the diagonal. The diagonal entries have opposite signs. This is the mathematical signature of a saddle point. It perfectly captures that Pringles chip curvature. Concave up in one direction, concave down in the other. All right, chapter five, unconstrained optimization. It's time to find some peaks and valleys. We've built the tools. We have the gradient knob to find the flat spots. These are our potential candidates for peaks and valleys. And we have the hesshen to tell us what kind of
21:04
flat spot we found. A peak, a valley, something weirder. Now we put them together. Here's the beautiful part. The logic is a direct extension of what you already mastered. Seriously, it's the same two-step dance just with more powerful tools on a much bigger stage. Let's look at the algorithm side by side. on the left for a single variable y= f ofx. Step one, you find critical points where the slope is zero by solving frime ofx equals 0. Step two, you classify each point using the second
21:41
derivative f-prime of x. Simple, right? Now look at the right for multivariable z= f ofxy. Step one, find critical points where the tangent plane is flat by solving for the gradient knob F equals the zero vector. Step two, classify each point using the Hessen matrix H. That's it. That's the entire game plan. Find where it's flat, then check the curvature. So, how do we check the Hessen? This is the second derivative test in 3D. At a critical point, let's call it a comma b. We
22:18
compute the hessen's determinant, which we'll call d. As you can see in this formula, d equals the determinant of h, which is fubxx * fuby y - fubxy 2. Now look at this table. It gives us all the logic we need. First condition, if d is positive and fubxx is positive, that's a local minimum. The intuition is that d being positive means the curvatures agree and fubxxx being positive means it's concave up. It's a valley. Second condition. If d is positive but fubxxx is negative that's a local
22:56
maximum. Again d is positive so the curvatures agree but fubxxx is negative so it's concave down a peak. Third condition. What if d is negative? That's a saddle point. The curvatures are fighting each other. One is bending up, the other is bending down. A Pringle. And finally, if D is zero, the test is inconclusive. You need more advanced methods, but we won't worry about that for now. Okay, enough theory. Let's put this algorithm to work on a real problem. Forget abstract surfaces
23:30
for a moment. Let's talk about money. Here's our motivating problem. Maximizing profit. A company's daily profit P from producing X units of product X and Y units of product Y is given by this function. P of XY = 1500US X - 10^ 2 - 3 * Y - 25^ 2. The question is obvious. To maximize profit, how many units of each product should they make? Let's do it. Here's our function again. Step one, find the critical points. Where is the gradient of P equal to the 0ero vector? First, the partial with
24:10
respect to X. That's -2 * X - 10, which simplifies to -2X + 20. We set that to zero and solve. -2X + 20 = 0. That means X = 10. Easy. Next, the partial with respect to Y. That's -6 * Y - 25, which is -6 Y + 150. Set that to zero. solve and y = 25. So we found only one critical point the point 10, 25. Now for step two, classify the point with the hessen. Is this a peak, a valley or a saddle? We need the second order partials. Okay. P subxx is the derivative of -2x +
24:56
20 with respect to x. That's just -2. P sub YY is the derivative of -6 Y + 150 with respect to Y. That's -6. And P subXY the mixed partial. Well, the derivative of -2X + 20 with respect to Y is just 0. There's no Y in it. So, we can assemble the Hessen matrix. In this case, it's actually constant. It's the matrix with -2 and 0 in the first row and 0 and -6 in the second. We have our critical point 1025. We have our hessen matrix. Let's run the test. First calculate the determinant D.
25:35
That's -2 * -6 - 0^ 2 which is just 12. Okay. Now apply the logic. Is d greater than 0? Yes. 12 is greater than 0. So we know it's either a minimum or a maximum. Now what's the sign of p subx? It's -2 which is less than zero. So since D is positive and P subXX is negative, the point 10, 25 is a local maximum. We found it. So let's write down the final answer. To maximize profit, the company should produce 10 units of product X and 25 units of product Y. And what is the maximum profit at that
26:19
point? We just plug our optimal point 10 and 25 back into the original function. P of 1025 = 1500 - 10 - 10^ 2 - 3 * 25 -25^ 2 is $1,500. So far we've been roaming free on our surface looking for the absolute highest point. But the real world has rules. It has limits. It has constraints. You want to maximize profit, but you only have a $10,000 budget. You want to design the lightest part, but it has to be strong enough not to break. We're no longer looking for the highest point in the entire
27:05
landscape. We're looking for the highest point along a specific path or within a specific boundary. For this, we need lrangee multipliers. Let's start with the motivating problem, the perfect pen. You're building a rectangular pen for your dog. You have a roll of fencing that is exactly 24 m long. Your goal is to build a pen with the maximum possible area. Let's formalize this. You're building a rectangular pen with a 24 m perimeter. There are two parts. First,
27:36
the objective function. This is what we want to maximize. That's f ofxy equals area, which is just x * y. Second, the constraint equation. This is the rule. The perimeter 2x + 2 y = 24. We can simplify that down to g ofxy = x + y = 12. So here's the key. We are trying to find the maximum value of f ofxy but only for points that are on the line x + y = 12. The aha moment the point of tangency. Imagine the contour plot is a topographical map. Your job is to walk
28:16
along the constraint line and find the point that touches the highest possible area contour. If your path is crossing a contour line, you haven't maxed out yet. The maximum happens at the one magic point where your path is perfectly tangent to a contour line. Now, what does tangent mean in the language of gradients? Remember, the gradient naba f is always perpendicular to the contour lines of f. So if the two curves are tangent, their normal vectors, their gradients must be parallel.
28:48
And this gives us the core equation. Two vectors are parallel if one is just a scaled version of the other. We call that scaling factor lambda, the lrangege multiplier. Naba f equals lambda* naba g. Okay, so here's the lrangege multiplier algorithm. Step one, identify your objective function f ofxy and your constraint g ofxy equals c. Step two, calculate the gradients naba f and naba g. Step three, set up the system of equations from naba fals lambda* naba g
29:26
and the constraint. That gives you fubx= lambda * g subx, fuby = lambda * g suby, and g ofxy= c. And finally, step four, solve this system for X, Y, and lambda. The XY solution is your candidate for the maximum or minimum. All right, let's solve the fence problem. Find the max area for a pen with a 24 m perimeter. First, identify functions. Objective f ofxy = xy. Constraint g ofxy = x + y = 12. Second, calculate gradients. Naba f is the vector yx and naba g is the
30:09
vector 1 1. Easy enough. Okay, we have naba f equals the vector yx and naba g equals the vector 1 1. So for step three, we set up the system. The equation na f= lambda * na g becomes the vector yx = lambda * the vector 1 1. This gives us three equations for our three unknowns. X Y and lambda. First, Y equals lambda * 1, so Y equals lambda. Second, X= lambda * 1, so X= lambda. And third, our original constraint, x + y = 12. Now for step four, solve the system. We have our three equations.
30:59
And there's the answer. The optimal dimensions for the pen are 6 m by 6 m, a perfect square. The maximum area is f of 66, which is 6 * 6, 36 square m. But what about lambda? You might be thinking it's just some helper variable we discard. But in fields like economics, lambda has a profound meaning. It's called the shadow price. The value of lambda tells you exactly how much your objective function would increase if you relaxed your constraint by one unit. In our fence problem, we
31:41
found lambda equals 6. This means for every one extra meter of fence we get, the maximum area we can enclose goes up by about 6 square m. We've officially mastered the first half of multivariable calculus, derivatives. We have we mastered the gradients, the Jacobian, the hessen. We use them to find the steepest paths to solve complex optimization problems. It's time to pivot. We're now tackling the other half of calculus. Integration. We are about to calculate the volume
32:17
under a 3D surface, the mass of a physical object, and even the probability of an event. So, let's connect this to what you already know. Let's look at the big picture from area to volume. In single variable calculus, you had the integral from a to b of fx dx. And this answered the question, what's the area under a curve? The logic was simple. First, you slice. You chop the x-axis into tiny pieces called dx. Second, you multiply. The height of each little slice is f ofx. And third, you
32:49
sum. That integral sign just means add up all the little areas. f ofx dx. Now we extend that exact same idea to a double integral for volume. The formula looks like the double integral over a region R of f ofxy da. The logic is identical. First you slice, but now you're chopping the entire xy plane into tiny little patches. DA. Second, you multiply. The height of the column above that little patch is f ofxy. And third, you sum. That double integral sign just means add up the volumes of all those
33:25
tiny columns. It's a beautiful extension of the same core idea. And here's what's so powerful. If f ofxy represents height, you get volume. But what if f ofxy represents density? Then you get mass. It's that simple. The same math, just a different interpretation. Okay, let's dive into our motivating problem. Starting with constant bounds, we need to find the volume under the surface z = 4 - x^2 - y^2 and outside the square defined by x between 0 and 1 and y
33:58
between 0 and 1. Our mission then is to calculate the volume which is the double integral over that region r of 4 - x^2 - y^2 da. So how do we actually do this? We use a technique called iterated integrals. We turn the double integral into two nested single integrals. So the double integral of f ofxy da becomes the integral from c to d of the integral from a to b of f ofxyd. All of that integrated with dy. Now the rule here is nonnegotiable. You work from the inside out. First you
34:36
solve the inner integral. You integrate with respect to x treating y as a constant. Then you take that result and for the second step you solve the outer integral integrating with respect to y. Let's apply this to our problem. We want the volume under z = 4 - x^2 - y^2 over that simple square region. Step one, set up the iterated integral. This is easy since our bounds are constant. V equals the integral from 0 to 1 of the integral from 0 to 1 of 4 - x^2 - y^2 dx dy. Step
35:10
two, solve the inner integral with respect to x. Remember, y is just a constant for this step. Think of it like the number 5. The integral becomes 4x - x cubed over 3 - y^2 x evaluated from x = 0 to x = 1. When you plug in the bounds, you get 4 * 1 - 1 / 3 - y^2 * 1us a big fat 0. This simplifies to 11/3 - y^2. Perfect. So the result of our inner integral is 11/3 - y^2. Now for step three, solve the outer integral with respect to y. We have v = the integral from 0 to 1 of 11/3 - y ^2
35:56
d y. This integrates to 11/3 y - yub / 3 evaluated from y = 0 to y = 1. Plugging in the bounds gives us 11/3 - 1/3 - 0. That gives us 10/3. The exact volume is 10/3 units. That was the warm-up. The real challenge of double integrals isn't the integration itself. It's setting up the bounds. Let's tackle the real challenge. Variable bounds. Find the mass of a triangular plate with vertices at 0 0 2.0 and 2.1. The density is given by the function row of xy = xy. So we need to
36:38
solve mass equals the double integral over the region r of xy da. First we have to describe the triangular region r. Let's look at the x bounds. The triangle clearly spans from x= 0 to x= 2. So 0 is less than or equal to x is less than or equal to 2. These are our constant outer bounds. Now for the tricky part, the y bounds. For any given x in that range, y goes from the bottom edge, which is y= 0, up to the top slanted line. That slanted line connects 0 to 2a 1. So its equation is y = x / 2.
37:16
And here is the critical rule. The bound that contains a variable in this case y = x /2 must go on the inside. So let's set up the calculation with our bounds. x from 0 to 2 and y from 0 to x /2. Calculation step one. The inner integral with respect to y. Mass equals the integral from 0 to 2 of the inner integral from 0 to x / 2 of xy dy all of that dx. Okay, let's solve that inner part. The integral of xy d y is x * y^2 / 2 evaluated from y = 0 to y = x /2. Plugging in the bounds, we get x *
38:02
x / 2^ 2 all / 2. This simplifies nicely. That's x * x^2 / 4 / 2, which equals xub 8. Excellent. The result of the inner integral is x cubed over 8. Now for calculation step two, the outer integral with respect to x. Mass equals the integral from 0 to 2 of x cubed over 8 dx. The integral of that is x 4 over 32 evaluated from 0 to 2. We plug in 2. So we get 2 4 over 32us 0. That's 16 over 32, which is 1/2. The total mass of the plate is 1/2. Now, this isn't just
38:47
some abstract math game. Let's make the connection to reality. These integrals are the workhorse of the physical sciences and beyond. In physics and engineering, you use them to calculate mass, center of mass, and moment of inertia for objects with varying density. In computer graphics, you integrate brightness over areas to find the total light hitting a 3D surface. And in probability and statistics, this is huge. You find the probability of an event happening in a certain region R by
39:15
taking the double integral of a joint probability density function over that region. It's an incredibly powerful and versatile tool. Double integrals are incredibly powerful, but for non-rectangular regions like a circle, setting up the bounds is a nightmare. The solution is to change our perspective, to switch to a coordinate system that is natural for the problem. But this switch isn't free. We need a correction factor. And you've already met the tool that provides it.
39:44
The Jacobian is back. Let's talk about the pain. The Cartisian nightmare. Imagine this problem. Find the volume under the surface z= 9 - x^2 - y^2 and above the xy plane. First, what's the base of this solid? Well, it's where z= 0. So x^2 + y^2 = 9. It's a circle of radius 3. Okay, let's set up the bounds. I want you to feel the pain. The bounds for x are fine. -3 is less than or equal to x is less than or equal to 3. But the bounds for y. Y goes from<unk> 9 - x^2 up to
40:26
positive<unk> 9 - x^ 2. So the integral for the volume looks like this. V equ= the integral from -3 to 3 of the integral from -<unk> 9 - x^ 2 to positive<unk> of 9 - x^2 roo<unk> of 9 - x^ 2 of 9 - x^ 2 - y^2 dydx. This is a truly awful integral. Now for the relief, the polar perspective. Remember x= r cosine theta, y = r sin thet and x^2 + y^2 = r 2. Look what happens to our problem. In cartisian, the region is x^2 + y^2 is less than or equal to 9. A circle with messy square root bounds.
41:10
But in polar, the region is simply 0 is less than or equal to r is less than or equal to 3. And 0 is less than or equal to theta is less than or equal to 2<unk>i. It's a simple rectangle in the r theta plane. And the function z = 9 - x^2 - y^2 just becomes z = 9 - r^2. This seems almost too good to be true. Can we just swap dx dy for dr d theta and call it a day? No. Here's the aha moment. The distortion factor. When we change coordinates, we stretch and distort area. The area element da is not
41:48
constant. If you look at this diagram, you can see what's happening. A patch with the same dr and d theta has a really small area when it's near the origin, but a much larger area when it's far away. We need a correction factor that tells us how much the area is distorted at every single point. And that factor is the absolute value of the determinant of the Jacobian. Let's do the Jacobian calculation. The formula is d * dy equals the absolute value of the determinant of j * dr d
42:17
theta where j is the jacobian of the r theta map to the xy map. It's a matrix of all the partial derivatives. Okay. Step one, compute the four parts. X= r cosine theta. So the partial of x with respect to r is cosine theta. The partial of x with respect to theta is r sin theta. Easy enough. Next y= r sin theta. The partial of y with respect to r is sin theta. The partial of y with respect to theta is r cosine theta. Got them? Step two. Assemble the jacobian
42:49
matrix. J equals the matrix with cosine theta and negative r sin theta in the top row and sin theta and r cosine theta in the bottom row. Step three calculate the determinant. The determinant of J is cosine theta * r cosine theta minus r sin thet * sin theta. Okay, that simplifies. That's r cosine^ 2 theta + r sin^ 2 theta. We can factor out the r. So we get r * cosine^ 2 thet + sin^ 2 theta. And what is that? It's just 1. So the determinant is just r. This gives us
43:29
the golden ticket for integration in polar coordinates. d * d y = r d r d theta. Now for the payoff, solving the problem the easy way. Let's go back to that volume under z= 9 - x^2 - y^2. Step one, convert everything to polar. The region is r from 0 to 3, theta from 0 to 2 pi. The function is 9 - r 2. And the area element da which was d * d y now becomes r d r d theta. Don't forget that r. Step two. Set up the new simple integral. v equals the integral from 0
44:09
to 2 pi of the integral from 0 to 3 of 9 - r 2 * r d r d theta. Distributing that r we get the integral of 9r - r cub r d theta. So much better. Let's solve it. Step three is the inner integral with respect to r. We need the integral from 0 to 3 of 9 r - r cub dr. The anti-derivative is 9 r^ 2 / 2 - r 4 / 4. We evaluate that from 0 to 3. Plugging in 3 we get 9 * 9 / 2 which is 81 /2us 3 4 over 4 which is 81 over4. The result of the inner integral is 81 over4.
44:59
Now for the final step four solve the outer integral with respect to theta. V equals the integral from 0 to 2 pi of just the constant 814 d theta. That's just 814 * theta. evaluated from 0 to 2<unk>i which is 81 over 4 * 2<unk>i. The volume is 81<unk> / 2. We solved it in three simple lines. No square roots, no trig substitution. That is the power of finding the right perspective. And this isn't just a party trick. This is the cornerstone of physics and engineering. Whenever a
45:38
problem has a natural symmetry, you exploit it. For things like pipes or cylinders, you use cylindrical coordinates. The Jacobian gives you dv= rdz d theta. For planets, stars or atoms, you use spherical coordinates. And the jacobian gives you dv= row ^2 sin v d row d theta dv. The jacobian is the key that unlocks these simpler perspectives, turning impossible problems into manageable ones. Can you believe it? You started this video just 50 minutes ago. Think back to that moment. The concepts
46:12
and symbols we were about to cover, they were probably pretty intimidating, a little abstract maybe. But look at them now. They are no longer abstract. They are tools. And not just any tools. They are the powerful toolkit you now have to solve real problems in engineering, physics, computer science, and economics. Let's just take a moment to appreciate the toolbox you have truly mastered. First the gradient del f equals the partial of f with respect to x and the
46:43
partial of f with respect to y. You know this. It's your compass for finding the direction of steepest ascent on any surface. Then the Jacobian, that big matrix of partials, J. You now understand how it translates changes in inputs to outputs across multiple dimensions. It's your universal translator for functions. The hesshen H the matrix of second partials. You use this to determine the very shape of a surface. Whether you're standing on a peak, in a valley, or at a
47:14
saddle point, it's your 3D scanner for functions. Unconstrained optimization. You know the drill. Set delf to zero, then test the hessen. And just like that, you can find and classify the maximums and minimums of a function. Lrangee multipliers, a true gamecher. del f equals lambda del g. This is your secret weapon for finding the max or min of a function when you're stuck on a constrained path. Double and triple integrals. That double integral of f ofxyda. It's not just a formula anymore. It's
47:50
how you calculate the volume under a surface or the mass of an object. And finally, change of variables. dx dy equals the absolute value of the determinant of j du dv. You just saw the power of this. It simplifies impossible integrals by letting you switch to a more natural coordinate system. All of this, every single piece, it all fits together now. Multivariable calculus now clicks.
Here to help you learn
Sign in to keep a private conversation with your course tutor.
Ask me anything about the material…
Sign in to chat