Differentiation Rules
Complicated functions are simple ones stacked together, and each rule undoes one kind of stacking.
Before this Lesson
Here's a tempting shortcut for the slope of a product: find each factor's slope and multiply. Try it on . Each factor has slope 1, so the shortcut says has slope 1 everywhere. You already know it's .
Slopes don't multiply just because functions do. Multiplying, dividing, and nesting each change a slope in their own way, and one rule covers each. Those rules bring exponentials, logs, and trig within reach, and then slopes settle a stranger question: what a fraction is worth when it reads 0/0.
Multiplying: the product rule
Think of as the area of a rectangle with width and height , both changing with . Nudge . The width grows by a sliver , the height by a sliver , and the rectangle gains area in three places.
The corner, , is a sliver of a sliver, and it vanishes as the nudge shrinks, as it did for the growing square in Derivatives and Slope. Divide the two strips by and you have the product rule, as the Handbook's Derivatives list prints it:
Each factor takes a turn changing while the other holds still. On , that gives , which is right this time.
Dividing: the quotient rule
A fraction rises when its top rises and falls when its bottom rises. That's why the Handbook's quotient rule has a minus sign, and why the order of its numerator matters:
The term with , the top's change, is the positive one. Write the numerator in the other order and every answer comes out with the wrong sign. If you doubt the order, test it on a fraction you know: with and , the rule gives , which is what the power rule says for .
Nesting: the chain rule
Now one function sits inside another: . Expanding that would be misery. Instead, name the inside, , so that . A nudge now travels down a line, like motion through a train of gears: moving moves , moving moves , and each link multiplies the nudge by its own rate.
The rates multiply, and that's the chain rule:
The Handbook's list prints this in function notation, without calling it the chain rule. In words: the derivative of the outside, times the derivative of the inside. The outside's derivative leaves the inside whole:
Most mistakes with these rules happen here, and nearly all of them are the same mistake: doing the outside and stopping. At , that gives when the slope is 160. The Handbook's power rule, , has the chain rule built in, and its is simply 1 when is just .
Exponentials, logs, and trig
The Handbook's Derivatives list has the rest of the pieces you need. It writes each one in terms of , with the chain rule's already attached. For plain that factor is 1, and for it's 3.
| Function of u | Derivative with respect to x |
|---|---|
Adapted from the FE Reference Handbook, Derivatives
is its own slope. That's what a continuous rate has meant all along: grows at , or times whatever is there right now.
has slope . It climbs ever more lazily: slope 1 at , and only 0.1 at .
A radian is arc length over radius, the in the Handbook's Circular Sector entry. A full turn is radians, so 180° is radians and one radian is about 57.3°. For more, see Areas of Plane Figures.
has slope . Walk along a sine wave reading slopes: climbing at slope 1 at the start, flat at the crest, and falling at slope −1 as it crosses the axis at . Those readings trace out the cosine wave. The cosine itself starts flat and then falls, so its slope is .
All of this has one condition, stated at the top of the Handbook's list: "All arguments of the trigonometric functions are in radians." A small angle in radians is nearly its own sine, , and that's what makes the sine's slope at zero exactly 1. In degrees, every trig derivative would carry a stray factor of . So means 1 radian. Keep the calculator in radians for calculus, and switch back for geometry.
The table's is , and your calculator probably has no sec key, so enter it that way.
When y won't stand alone
Sometimes and come tangled together, as in . There's no single function to solve for, because this curve is a closed loop: at it has two heights, and .
A loop still has a tangent line at each point, so you need a way to differentiate without solving for . Here it is: near any one point, a short piece of the loop is a function of , whether or not you have its formula. That makes a nested function with on the inside:
Differentiate each term as though were , then attach for the inside. When and multiply, that term is a product, so the product rule joins in. The Handbook has no entry for this, and it doesn't need one: it's the chain rule again.
Fractions that read 0/0
What does do as closes in on 2? Put in and you get 0/0. That isn't a number, and it isn't an error either. It's an open question: the top and bottom are both racing to zero, and the fraction's fate depends on how fast each one is going. A limit asks what value the fraction closes in on.
For a polynomial fraction, factoring settles it. The top is , so the cancels, and everywhere except at 2 itself the fraction is plain :
But as has nothing to factor, so use slopes. Suppose and both reach zero at . Zoom in until both look straight. A short step away, each one's height is its slope times :
The cancels, so as it shrinks, the ratio of heights settles on the ratio of slopes. For at , the top has slope 2 and the bottom has slope 1, and sure enough, the fraction is 2.22 at and 2.02 at .
That's L'Hôpital's rule, which the Handbook prints with the older spelling, L'Hospital's Rule. For a fraction in the form 0/0:
Differentiate the top by itself and the bottom by itself. This isn't the quotient rule: you aren't after the slope of the fraction, you're comparing two slopes. So becomes , which goes to 1 (in radians only), and becomes , also 1.
If the new fraction still reads 0/0, both slopes are zero there, so go on to the next derivatives. The Handbook builds this in: the limit equals the first ratio of derivatives that isn't indeterminate.
The same race out to infinity
Now let grow without bound in . The top and bottom both grow forever, and ∞/∞ is the same kind of open question: what matters is how fast each one is growing. The Handbook's rule covers this form too, and its limit point may be infinite:
The bottom climbs steadily at 1 per unit while the top's climb fades away, so the log loses. In the same way, becomes , which goes to 0.
Rational functions at infinity
A ratio of polynomials doesn't need L'Hôpital at infinity, and a few numbers show why.
| x | Top | Bottom | Top ÷ bottom |
|---|---|---|---|
| 10 | 257 | 209 | 1.230 |
| 100 | 29,507 | 20,009 | 1.475 |
| 1,000 | 2,995,007 | 2,000,009 | 1.497 |
Out there, the terms do nearly all the work and the rest barely register. So keep the highest-power term on top and on the bottom, and cancel: . The Handbook doesn't print this shortcut, so know its three outcomes:
- Same highest power on top and bottom: the limit is the ratio of their coefficients.
- Higher power on the bottom: the limit is 0.
- Higher power on top: the fraction grows without bound.
Check the form first
The Handbook's rule is only for 0/0 and ∞/∞. Use it on anything else and it hands you a wrong answer with a straight face. As , the fraction is , and that's the answer. Differentiating top and bottom would give . So substitute first, and look at what you get.
| Form | What it means | L'Hôpital's rule? |
|---|---|---|
| 0/0 | An open race | Yes, or factor |
| ∞/∞ | An open race | Yes |
| A number over a nonzero number | That's the limit | No |
| A number over ∞ | The limit is 0 | No |
| A nonzero number over 0 | No finite limit | No |
| ∞ over a number | No finite limit | No |