Maxima, Minima, and Inflection Points
At the top of a hill the slope is zero, and that's how you find the top.
Before this Lesson
Walk a hilly road in fog so thick you can't see the landscape. All you have is the tilt of the ground under your feet and the way that tilt is changing. That's enough to notice the top of a hill, where the ground goes level. But level ground also comes at the bottom of a dip and on a shelf partway up a climb. And the highest spot of the whole walk may not be level at all, if the road simply ends while it's still climbing.
Tilt and change of tilt are the first and second derivatives, and between them they sort out the level spots. The road's end is one more place you'll have to check.
Flat spots are the candidates
The sign of says whether a curve is rising or falling, as you saw in Derivatives and Slope. A smooth curve can't go from rising to falling without passing through a slope of zero, any more than you can walk from uphill to downhill without one level step. So at every peak and valley of a smooth curve, the tangent is flat and . A point where that happens is a critical point.
That turns a hunt along the whole curve into algebra. Take and set its derivative to zero:
Those are the only two places on the entire curve that can be a peak or a valley.
The peak is a local maximum: higher than everything near it, though not higher than everything, since the curve passes that height beyond . The valley is a local minimum.
Curving down or curving up
To , the peak and the valley look alike: it's zero at both. To tell them apart, walk through each one and read the slope as you go. Over a peak it reads something like +2, +1, 0, −1, −2, dropping the whole way. Through a valley it reads −2, −1, 0, +1, +2, climbing the whole way.
How fast the slope is changing is the second derivative, , so its sign gives the verdict. When is negative, the slope is dropping and the curve bends downward like a dome, which at a flat spot means a peak. When is positive, the slope is climbing and the curve bends upward like a bowl, which at a flat spot means a valley. That bend is the curve's concavity: concave down or concave up.
The Handbook's Differential Calculus section prints this as its Test for a Maximum and Test for a Minimum. It says only "maximum" and "minimum," with no "local" or "concave," but these are the same ideas.
This is the second-derivative test, and its minus sign trips people up, because sounds as though it ought to mean a low point. But it's the slope that's dropping, not the height, and a slope can only drop through zero by going over a top.
For the cubic, . At that's , so it's a peak, with height . At it's , so it's a valley, with height .
Beyond polynomials
The method doesn't care what kind of function you feed it. Differentiate with the rules from Differentiation Rules, set , and test with .
From the Handbook's Derivatives list, where and are functions of :
Two habits start to matter more here:
- Check each root against the domain. The function exists only for , and a problem may fence off the negatives itself by saying "for ." A root of outside the domain isn't a candidate.
- Answer what was asked. Where a maximum occurs is an . The maximum value is at that , so it takes one more step: put the back into .
Where the bend switches
Look at the cubic again between its peak and its valley. Left of it bends downward, and right of it bends upward. Drive along it from left to right and you'd be steering right, then left, with the wheel passing through center for just one instant. That spot is an inflection point, where the concavity switches.
The sign of is the concavity, so has to change sign there. The Handbook's Test for a Point of Inflection asks for exactly that: , and changing sign as increases through .
So solve , then check the sign on each side. For the cubic, gives , and is at and at , so the sign does change. The height comes from : , which puts the inflection point at .
An inflection point usually isn't flat. The slope here is , the steepest the curve gets between peak and valley, and that's no accident: has a valley at , and a valley of is where its own derivative, , is zero. It's the same trick as before, one derivative up.
A cubic's is a straight line, which crosses zero exactly once, so every cubic has exactly one inflection point. A higher polynomial can have more.
When the test has no verdict
When is zero at a flat spot, it's tempting to announce an inflection point. Look at , , and first. At , all three report and , identical readings. Yet the first has a valley there, the second a peak, and the third a shelf on a steady climb.
So when and are both zero, the second-derivative test is undetermined. The Handbook's maximum and minimum tests don't cover this case at all. Nor is proof of an inflection point, which is why the Handbook's inflection test also demands the sign change. On , is zero at but positive on both sides, so the curve bends upward the whole way, and there's no inflection point.
When the test goes quiet, go back to what your feet can tell you, and check a little to each side of the point. The Handbook doesn't print this check, but it's only the sign of again:
- Positive, then negative: uphill, then downhill, so a local maximum.
- Negative, then positive: downhill, then uphill, so a local minimum.
- The same sign on both sides: neither, just a pause on the way up or down. On a polynomial, that shelf is an inflection point with a flat tangent.
Here's how the readings at a point add up:
| At x = a | What the point is |
|---|---|
| Local maximum | |
| Local minimum | |
| Undetermined: check f′ on each side | |
| Inflection point, if f″ changes sign | |
| None of these: just rising or falling |
When the road ends
Everything so far has been local: a point compared with its neighbors. Now a different question. On a closed interval, a stretch such as that includes its ends, where are the very highest and very lowest points? Those are the absolute maximum and absolute minimum.
Back on the foggy road, if it ends while you're still climbing, the end is the high point, and the ground there isn't level. Setting can never find it, and the Handbook's tests say nothing about ends. So add them to the candidates by hand:
- Solve , and keep only the roots inside the interval.
- Compute at each of those and at both ends.
- The largest value is the absolute maximum, and the smallest is the absolute minimum.
You don't need the second-derivative test here. You're comparing heights, so it doesn't matter what kind of point each candidate is. On the same cubic, the winners change with the interval: