The domain decides which arguments are available
Begin with the function as defined, including excluded inputs and interval restrictions. Cancellation may expose a removable hole; it does not fill it without a deliberate extension. A pole is different: a vertical asymptote requires an infinite one-sided limit, not merely a zero in a displayed denominator.
Work on each domain interval before comparing different intervals. Derivative signs yield comparisons between two inputs only when the segment joining them lies in a region where the mean value theorem applies. A forbidden input between them breaks that argument. A graph with several decreasing branches need not be decreasing when considered on its whole disconnected domain.
Definition 4.5.1
(Stationary, critical and endpoint candidates)
Keep three lists distinct. Stationary points are interior points with derivative zero. Interior critical points also include places where a derivative fails. Included endpoints are additional candidates for a global extremum. A point outside the domain belongs to none of these lists, even if a formula or a plotted line suggests a feature there.
Why signs give monotonicity and local extrema
Proposition 4.5.2
(Monotonicity from the sign of the derivative)
Let be continuous on an interval and differentiable at the interior points of . If at every interior point , then is strictly increasing on ; if there, is nondecreasing. The statements with reversed signs hold for decreasing functions.
Proof
(Mean value theorem on any two inputs)
Suppose a function is continuous on every closed segment of an interval and differentiable at its interior points. For any two inputs in that interval, the mean value theorem gives
Consequently positive derivative everywhere between them gives strict increase; nonnegative derivative gives nondecrease. The corresponding negative statements follow by negating the function. Derivative continuity is unnecessary. What matters is the sign at the theorem's unknown intermediate point, which is why checking a few sampled derivative values is insufficient.
Proposition 4.5.3
(First-derivative test at an interior point)
Let be continuous at an interior point of its domain interval and differentiable at the nearby points on both sides of , not necessarily at . If immediately to the left of and immediately to the right, then is a strict local minimum. With the signs reversed, it is a strict local maximum. If the signs on the two sides agree, is not a local extremum.
Proof
(Mean value theorem on the two closed segments ending at the point)
For the first-derivative test at an interior point , assume continuity at and differentiability on each side in a neighborhood. If the derivative is negative on the left and positive on the right, apply the mean value theorem separately to the closed segments ending at . At every sufficiently close domain input different from , the function value is strictly larger than , proving a strict local minimum. Reversing the signs proves a strict local maximum. Equal strict signs on both sides give a value below and a value above, so no local extremum occurs.
This argument also works at a corner: it never differentiates at the joining point. Conversely, a zero derivative alone is inconclusive. A differentiable interior extremum must be stationary by the signs of its one-sided difference quotients, but necessity is not sufficiency. Neither condition says whether a local extremum exceeds values on a distant branch.
Counterexample mode
Continuity at the joining point cannot be omitted
Claim: if the derivative is negative on the left of and positive on the right, then is a local minimum, with no continuity needed at . Define for and . For , is negative on the left of and positive on the right. But and , and the same inequality holds at every nonzero input in . So is a strict local maximum, not a minimum. The mean value theorem on is unavailable because is not continuous at , which is exactly the closed-segment hypothesis in the proposed proof. The repair is to assume that is continuous at .
Convexity is a statement about chords
Definition 4.5.4
(Convex and concave functions)
A function is convex on an interval if for all in and every with ,
It is strictly convex if the inequality is always strict. It is concave (strictly concave) if the inequalities are reversed. In words: on a convex function the graph between two points lies on or below the chord.
Proposition 4.5.5
(Convexity from a monotone derivative)
Let be differentiable on an interval . If is nondecreasing on , then is convex on . If is strictly increasing, then is strictly convex. In particular, throughout gives strict convexity, and throughout gives strict concavity.
Proof
(Compare two secant slopes)
Suppose is nondecreasing and take . The mean value theorem gives the left and right secant slopes as derivative values at ordered intermediate points. Therefore
Multiplying by the positive denominators and rearranging yields
Strict increase of the derivative makes the inequality strict. A second derivative positive throughout an interval makes the first derivative strictly increasing by another mean value argument, and hence proves strict convexity. A negative second derivative proves strict concavity by applying this reasoning to the negative function. A single value of the second derivative does not establish an interval of curvature.
Proof X-ray
Why the ordered intermediate points matter
The convexity proof is two applications of the mean value theorem and one rearrangement. Differentiability on the interval gives the left secant slope as with and the right secant slope as with . The key move is that these intermediate points are ordered, , because they lie in disjoint subintervals; that order is what lets monotonicity of compare two slopes taken at different places. The rearrangement then multiplies by the positive denominators and solves for , which writes as a weighted average of and with weights and that sum to . Without monotone the inequality between the slopes fails, and strict increase of turns into a strict inequality.
An inflection requires a point of the continuous curve where convexity changes. In particular, opposite curvature signs on separate sides of a pole do not create an inflection at the absent input. At a nonsmooth corner we will state the one-sided curvature facts explicitly; textbooks that additionally require a tangent for the word “inflection” use a narrower convention. A zero second derivative by itself settles none of these questions.
Concept lensGeometric
Convexity as tangent lines lying below the graph
For a differentiable , convexity on is equivalent to
so every tangent line lies on or below the graph. From the chord inequality, the secant slope is at most for ; letting gives the tangent inequality for , and the symmetric argument with gives it for . Conversely, apply it at to both and and add the results with weights and : the terms in cancel, leaving the chord inequality.
Asymptotes and ranges come from limits
Definition 4.5.6
(Horizontal and oblique asymptote)
A horizontal or oblique asymptote is an affine line for which the difference between the function and the line tends to zero at the relevant infinity. The two directions are checked separately. This definition does not forbid a finite crossing. Nor does it claim that the asymptote is an attained endpoint of the range.
To prove an exact branch range, combine monotonicity, continuity and boundary limits. A bound alone gives only containment. For the reverse inclusion, bracket a desired value between two actual function values and invoke the intermediate value theorem on the segment between their inputs. An infinite endpoint is handled by first choosing a sufficiently distant finite input, not by applying a theorem to an interval with an infinite endpoint.
For global extrema on a closed bounded interval, continuity supplies existence before critical and endpoint values are compared. On open, unbounded or disconnected domains, the existence conclusion needs another argument. A finite limiting height may be a supremum without being a maximum. A local branch maximum may fail even to bound values on another branch.
Worked example: Three decreasing branches and one smooth inflection
Worked example 4.5.7
(Three decreasing branches and one smooth inflection)
Question. Give a complete analytic description of
Find domain, intercepts, asymptotes, monotonicity, extrema, curvature and exact branch ranges.
Solution. The domain excludes and . The function is odd, and its only intercept is the origin. At the left pole, the limits are negative infinity from the left and positive infinity from the right. At the right pole, they are negative infinity from the left and positive infinity from the right. These signs follow by retaining both numerator and denominator signs.
At both infinities the function tends to zero, so is a horizontal asymptote. The graph also crosses that line at the origin. Differentiation gives
The first derivative is strictly negative everywhere in the domain. Thus each of the three domain components is strictly decreasing, and there are no interior local extrema or stationary points. This does not make the whole function decreasing: for example, but . The missing input between them prevents a cross-branch mean value argument.
The second derivative is negative on , positive on , negative on and positive on . These give alternating strict concavity and convexity. The origin is a smooth inflection, with slope , so it is not stationary. Neither pole is a point of inflection.
The left branch decreases from a limit of zero approached through negative values to negative infinity; its range is . The middle branch decreases continuously from positive infinity to negative infinity, so its range is all of by IVT. The right branch has range . The full range is therefore all real numbers, with no global maximum or minimum. Every asserted interval of values is attained by the bracketing argument above.
Worked example: Corners and endpoints belong in the extremum comparison
Worked example 4.5.8
(Corners and endpoints in the extremum comparison)
Question. Analyze on the restricted domain . Find all local and global extrema and describe the curvature on its smooth pieces.
Solution. The quadratic factors as . Thus equals the quadratic outside and its negative inside that interval; it is continuous at both joining points. On the smooth pieces the derivatives are outside and inside. The function decreases to , increases to , decreases to , then increases to the right endpoint.
At input , the left and right derivatives are and ; at input , they are again and . Both are corner minima of value zero. The derivative at neither corner exists, so neither is stationary. Input is a stationary local maximum of value one.
For the global comparison, the complete candidate values are
Continuity on the closed bounded domain guarantees that this candidate comparison finds the global extrema. The global maximum is eight at the left endpoint, and the global minimum is zero at both corners. Relative to this restricted domain, each endpoint is also a one-sided strict local maximum; the interior local maximum is not global.
The second derivative is two on the outer smooth intervals and negative two on the middle interval. Thus the outer pieces are strictly convex and the middle piece strictly concave. Curvature switches across each continuous corner, but there is no tangent there. Under the definition requiring continuity of the function and a change of convexity these are inflections as well as corners; under a tangent-required convention they are not. The actual derivative and curvature facts are independent of that terminology choice.
Worked example: An open domain and an oblique asymptote
Worked example 4.5.9
(An open domain with an oblique asymptote)
Question. Analyze on : global extremum, curvature, asymptotes and exact range.
Solution. The domain is the open interval , so there are no endpoints, and the input is excluded. The derivative is , negative on and positive on . Apply the mean value theorem on each side of : for every . So is a global minimum, although the domain is open and unbounded and the closed-interval existence theorem does not apply. The second derivative is positive on the whole domain, so is strictly convex.
As , , so is a vertical asymptote. As , , so is an oblique asymptote, approached from above and never crossed since . For the range, let . Then , so the intermediate value theorem on gives a point where . The range is therefore , and the minimum value is attained while the asymptote is not an extremal value of any kind.
Common mistakes
Common mistake
(Concluding global monotonicity from monotone pieces)
The function is strictly decreasing on each of its three components, but not on its whole domain: and yet . A derivative sign yields a comparison only on an interval where the mean value theorem applies.
Common mistake
(Treating a sign change across a pole as an inflection)
For the second derivative is negative for and positive for . But is not in the domain, so there is no point of the curve where the convexity changes. The graph has no inflection.
Common mistake
(Reading too much into a zero derivative)
Both and have and . For there is no extremum and changes sign, so is an inflection. For there is a minimum and never changes sign, so is not an inflection. A zero value alone decides neither question.
Assemble the graph only after the proof
Label actual points separately from excluded inputs, and use the proven branch ranges to check the sketch. Do not draw a connector across a pole. At a corner draw the two one-sided slopes, rather than inserting a horizontal tangent because the point is an extremum. Distinguish an approached asymptote from an attained extremal value. These decisions summarize the analysis instead of supplying unproved evidence for it.
Exercises
Quick check
(Find the intervals of convexity and concavity and any inflection point of .)
Solution · Answer
Here . It is negative on and positive on , so is concave on and convex on . Because is continuous and the curvature changes at , the point is an inflection.
Quick check
(Find the global maximum and minimum of on .)
Solution · Answer
The candidates are the endpoints and the corner: , , . The global minimum is , attained at the corner where does not exist. The global maximum is at the endpoint , where .
Quick check
(Find the horizontal asymptote of and decide whether the graph ever crosses it.)
Solution · Answer
The limit at both infinities is , so is the horizontal asymptote. Since , the graph never reaches the line and stays below it.