Understanding Concavity Without the Textbook Fluff
Concave up means the derivative is increasing. Concave down means the derivative is decreasing. That's basically the whole thing, but in practice most people mess it up because they've only seen it presented as "holds water" vs "spills water," which is a mnemonic that doesn't actually help you compute anything. I spent years tutoring calculus students who could identify concavity on a neatly factored polynomial but fell apart the second they hit a rational function with a discontinuity or a piecewise definition. The pattern repeats every semester.
How to Actually Determine Concave Up Concave Down
Take the second derivative. Set it equal to zero and find where it's undefined. Those are your inflection point candidates. Test intervals around those points by plugging in values. Positive second derivative equals concave up on that interval. Negative equals concave down. Period. The faster method uses the first derivative directly if you already have it. When f'(x) is going up, you're concave up. When f'(x) is going down, concave down. This matters because sometimes computing the second derivative symbolically is a nightmare and the first derivative is clean. I've seen people spend twenty minutes differentiating a quotient twice when they could have just looked at the slope behavior of f'(x) over a test interval.
The Edge Case That Wastes Everyone's Time
Here's the one nobody warns you about. Consider f(x) = x^(4/3). The second derivative is f''(x) = (8/9)x^(-2/3). At x = 0, the second derivative is undefined. The function is continuous there. The first derivative exists there and equals zero. So is x = 0 an inflection point? It's not. The sign of f''(x) is positive on both sides. Concavity doesn't change. Students automatically flag any point where f'' is undefined as an inflection point and lose points for it. My workaround was simple: I made them write out a sign chart for f'' on both sides of every candidate point before they'd even consider calling it an inflection point. No exceptions. Cuts the error rate dramatically after a couple weeks of practice. Another situation that trips people up involves absolute value functions inside the concavity test. Take f(x) = x|x|. You can rewrite this as x² for x 0 and -x² for x < 0. The first derivative is 2|x|, which is differentiable everywhere including at zero. The second derivative is 2 for x > 0 and -2 for x
0. At x = 0 the second derivative jumps from -2 to 2, so there is an inflection point and the concavity switches. But if you try to differentiate x|x| using standard rules without handling the piecewise structure first, you'll get garbage at x = 0 because the product rule gives you something involving |x|/x which is undefined there.
Get the Full Details
When This Method Breaks Down
The second derivative test for concavity requires f to be twice differentiable on the interval you're testing. If f'' doesn't exist at a point, you can't use it to determine concavity on an open interval containing that point. You have to split the interval and test each side separately. This is obvious to people who've done it enough times but beginners will often write something like "f''(0) doesn't exist so the function has no concavity at 0" as if that's a valid mathematical statement. Apart from that, the method is reliable for any function you're likely to encounter in a standard course. For functions defined parametrically or implicitly, you need to compute the second derivative using the appropriate chain rule extensions, which adds a layer of algebra that tends to introduce sign errors. I'd recommend sticking to the first derivative slope observation method for those cases instead, since it's less mechanically fragile. Also worth noting: concavity tells you about the shape of the graph locally but it doesn't replace the first derivative test for classification of critical points. A function can be concave down everywhere and still have no local maxima if the critical point is an endpoint or a boundary issue. Don't conflate the two concepts.
