Visual walkthrough — Second derivative test — concavity, inflection points

Before symbol one: a function $f$ is a rule that takes a number $x$ and gives back a height $f(x)$. Plot the point $(x,f(x))$ for every $x$ and you get a curve — a road drawn above a number line. That road is all we will look at.

---

Step 1 — The slope of the road: what $f^{\prime }$ measures

WHAT. Pick a point on the curve and lay a straight ruler so it just grazes the curve there, touching but not crossing. That ruler is the tangent line. Its steepness — how much height you gain per step to the right — is the slope. We name this slope $f^{\prime }(x)$ (read "$f$-prime of $x$"). The little prime mark ${}^{\prime }$ is just notation for "the slope function of $f$."

WHY this tool. We want to talk about bending, but bending is a change in direction, and direction is exactly what slope captures. So the honest starting point is slope. Nothing about "second" yet — first we must be fluent in the first.

PICTURE. Three tangent rulers on one curve.

---

Step 2 — Watch the slope change as you walk the cup

WHAT. Take a bowl-shaped ($\cup$) curve and walk left-to-right, reading off the tangent slope at each stop. On the left the ruler points steeply down (very negative slope). Near the bottom it goes flat (zero). On the right it tilts up (positive). So the slope-number marches $-3\to -1\to 0\to +1\to +3$: it is increasing.

WHY. This is the key observable. "Concave up" is not a new mysterious thing — it is literally the slope getting larger as you move right. We are turning a shape-word ("cup") into a number-behaviour ("slope increasing").

PICTURE. The same bowl with tangent rulers drawn at five stops; slope values labelled.

---

Step 3 — Measure "slope increasing" — apply the derivative again

WHAT. We have a number that changes as $x$ moves: the slope $f^{\prime }(x)$. To ask "is this number increasing?" we do the exact same thing we did in Step 1 — take the slope of it. The slope of $f^{\prime }$ is written $f^{\prime \prime }(x)$ (two primes = "slope of the slope").

WHY this tool and not another. We already have one perfect device for "is a quantity increasing?": the derivative, because $g^{\prime }>0$ means "$g$ goes up as $x$ goes up." We do not need to invent anything new — we just feed $f^{\prime }$ into that same machine. Choosing the derivative here is the only natural choice because increase/decrease of any function is what the derivative reports.

PICTURE. Two stacked plots — top is $f^{\prime }(x)$ (a rising line for our cup); bottom is its own slope $f^{\prime \prime }(x)$, sitting above zero.

---

Step 4 — The cap ($\cap$): every sign flips

WHAT. Now walk a dome-shaped ($\cap$) curve. Tangents go from steeply up on the left, to flat at the peak, to steeply down on the right: slope marches $+3\to +1\to 0\to -1\to -3$ — decreasing. Slope decreasing means its slope is negative: $f^{\prime \prime }<0$.

WHY show this separately. The contract is to cover every case. Cup and cap are the two concavities; we must see the sign genuinely reverse, not assume it.

PICTURE. Dome with five tangent rulers, slopes labelled decreasing; a small inset shows $f^{\prime }$ as a falling line.

---

Step 5 — Where cup becomes cap: the inflection point

WHAT. Slide a single curve from a $\cup$ region into a $\cap$ region. Somewhere the bending must reverse. At that switch the slope $f^{\prime }$ stops increasing and starts decreasing — so $f^{\prime \prime }$ passes from $+$ through $0$ to $-$. The crossing point $(c,f(c))$ is the inflection point.

WHY $f^{\prime \prime }=0$ is only a clue. For $f^{\prime \prime }$ to go from $+$ to $-$ it must cross zero if it is continuous. But hitting zero is not the same as crossing it — a value can touch $0$ and bounce back to the same sign. That is why the definition demands a genuine sign change, not merely $f^{\prime \prime }(c)=0$.

PICTURE. One S-shaped curve; left half shaded as $\cup$, right as $\cap$, the join marked as inflection, with the $f^{\prime \prime }$ sign strip $+0-$ underneath.

---

Step 6 — The degenerate trap: $x^4$ touches zero but never turns

WHAT. Take $f(x)=x^4$. Its slope-of-slope is $f^{\prime \prime }(x)=12x^2$. At $x=0$ this equals $0$ — the "clue" fires. But $12x^2\ge 0$ for every $x$ (a square is never negative). So $f^{\prime \prime }$ dips to $0$ and rises again without changing sign: concave up on both sides. No inflection, despite $f^{\prime \prime }(0)=0$.

WHY it matters. This is the case the naive rule "$f^{\prime \prime }=0\Rightarrow$ inflection" gets wrong. Seeing $12x^2$ kiss the axis and stay non-negative is the whole lesson.

PICTURE. Top: the flat-bottomed $x^4$ bowl. Bottom: $f^{\prime \prime }=12x^2$ touching the axis at $0$ but staying above it.

---

Step 7 — The other degenerate case: inflection where $f^{\prime \prime }$ does not exist

WHAT. The definition allowed a second door: an inflection where $f^{\prime \prime }(c)$ fails to exist. Here is the honest example — the cube root $f(x)=x^{1/3}$. Differentiate twice: $f^{\prime }(x)=\frac{1}{3}{x}^{-2/3},f^{\prime \prime }(x)=\underset{\text{constant}}{\underbrace{-\frac{2}{9}}}\underset{=1/x^{5/3}}{\underbrace{x^{-5/3}}}.$ At $x=0$ the denominator $x^{5/3}$ is zero, so $f^{\prime \prime }(0)$ is undefined — division by zero, the curve is vertical there. But look at the sign on each side: for $x<0$, $x^{5/3}<0$ so $f^{\prime \prime }=-\frac{2}{9}\cdot (\text{neg})>0$ (concave up, $\cup$); for $x>0$, $x^{5/3}>0$ so $f^{\prime \prime }<0$ (concave down, $\cap$).

WHY this matters. The concavity genuinely flips $\cup \to \cap$ across $x=0$, and $f$ is continuous there, so $(0,0)$ is a true inflection point — even though $f^{\prime \prime }$ was never zero and never even defined at the switch. This closes the definition's second door that Step 6 left open.

PICTURE. Top: the vertical-tangent cube-root curve, up-bending then down-bending. Bottom: $f^{\prime \prime }$ diving to $+\infty$ from the left and to $-\infty$ from the right — a genuine sign change through a break, not through zero.

---

Step 8 — The payoff: why concavity classifies a flat point (Taylor)

WHAT. At a critical point $c$ the tangent is flat: $f^{\prime }(c)=0$ (this is where Critical points live). Near $c$, Taylor's expansion writes the height difference exactly as the bending term plus a leftover: $f(x)-f(c)=\underset{\text{constant}}{\underbrace{\frac{1}{2}}}\underset{\text{concavity sign}}{\underbrace{f^{\prime \prime }(c)}}\underset{\ge 0\text{always}}{\underbrace{(x-c{)}^2}}+\underset{\text{remainder}}{\underbrace{R(x)}},\frac{R(x)}{(x-c{)}^2}\underset{x\to c}{\to }0.$

WHY this tool. We want to compare the height $f(x)$ to the flat-point height $f(c)$ for nearby $x$. Taylor's formula is the microscope that writes that difference in terms of $f^{\prime \prime }$. The linear term died because $f^{\prime }(c)=0$, leaving the square term in charge.

Read the sign. Since $(x-c{)}^2\ge 0$ always and $R$ is negligible near $c$ (the limit above), the difference $f(x)-f(c)$ carries the sign of $f^{\prime \prime }(c)$:

• $f^{\prime \prime }(c)>0\Rightarrow f(x)>f(c)$ nearby ⇒ sitting in a valley ⇒ local minimum.
• $f^{\prime \prime }(c)<0\Rightarrow f(x)<f(c)$ nearby ⇒ sitting on a peak ⇒ local maximum.
• $f^{\prime \prime }(c)=0\Rightarrow$ square term vanishes ⇒ $R$ now decides ⇒ inconclusive, defer to the First derivative test.

PICTURE. A cup with a flat tangent at its base (min) beside a cap with a flat tangent at its crest (max); the parabola $\frac{1}{2}{f}^{\prime \prime }(c)(x-c{)}^2$ drawn over each.

---

The one-picture summary

Everything compressed: the curve on top, its slope $f^{\prime }$ in the middle, its concavity $f^{\prime \prime }$ at the bottom — three plots sharing one $x$-axis so you can read the whole story vertically.

Recall Feynman retelling — the whole walk in plain words

Draw a hilly road. The slope tells you if you're going up or down (that's $f^{\prime }$). Now don't watch the height — watch the slope itself as you walk. If the slope keeps getting bigger (from steep-down, to flat, to steep-up), the road is curving like a smile, a cup: that's $f^{\prime \prime }>0$. If the slope keeps shrinking, the road frowns, a cap: $f^{\prime \prime }<0$. Measuring "is the slope growing?" is the same job as measuring slope, so we just take the slope of the slope — that's the two little primes. Where the smile turns into a frown, the slope stops growing and starts shrinking, so $f^{\prime \prime }$ flips sign — that's the inflection point — either passing through zero (like $x^3$) or through a break where $f^{\prime \prime }$ blows up (like the cube root). It is not an inflection if the road merely kisses flatness and keeps the same curve (like $x^4$). Finally, if you stop on a flat spot: a cup-flat is the bottom of a valley (a min), a cap-flat is the top of a bump (a max). That's the whole second-derivative test, seen not memorised.

Recall Quick self-check (open after attempting)

Q1. What does $f^{\prime \prime }>0$ mean geometrically? → Concave up ($\cup$); the slope $f^{\prime }$ is increasing. Q2. Why is taking the derivative twice the right tool for concavity? → Concavity asks "is the slope increasing?", and the derivative is exactly the "is it increasing?" detector — applied to $f^{\prime }$. Q3. Is $f^{\prime \prime }(c)=0$ enough to guarantee an inflection? → No; $f^{\prime \prime }$ must actually change sign. $x^4$ has $f^{\prime \prime }=12x^2\ge 0$, no flip. Q4. Can an inflection happen where $f^{\prime \prime }$ does not exist? → Yes; $x^{1/3}$ at $0$ flips $\cup \to \cap$ with $f^{\prime \prime }(0)$ undefined. Q5. $f^{\prime }(c)=0$ and $f^{\prime \prime }(c)<0$ — classify. → Local maximum (cap-flat = peak).

Connections

• First derivative test
• Critical points
• Taylor series
• Curve sketching
• Maxima and minima — optimization
• Concavity and convex functions