You can find local maxima and minima from a graph by locating points where the curve changes direction — a peak where it stops rising and starts falling, or a valley where it stops falling and starts rising — then checking whether that point beats its immediate neighbors rather than the entire graph. The boundary matters: a peak is only "local" if it is the highest point in some small neighborhood around it, so a graph can hold several local maxima while having just one global maximum.
Why the Definition of "Local" Decides What Counts as a Peak
Local in calculus means neighborhood-bound, not graph-wide. A point is a local maximum when its y-value is greater than or equal to the y-values of every nearby point on the graph, and a local minimum when it is less than or equal to its neighbors' values. The qualifier is strict about proximity: a hilltop that is only the third-highest peak on the whole picture still qualifies as a local maximum, provided nothing immediately around it rises higher.
That is why the same curve can have three local maxima and still have one global maximum — the one with the largest y-value across the entire domain. Textbooks including the widely used Stewart Calculus state the definition in exactly this neighborhood form, which is the version exam questions test. When a problem asks for "all local maxima," it wants every peak, not just the tallest.
The phrase "from a graph" changes the method, not the meaning. On a graph you identify candidate peaks visually by hunting for turning points, then confirm them with a sign-change check: the slope goes positive to negative at a maximum, negative to positive at a minimum. An equation gives you the derivative; a picture gives you the shape.
One boundary worth settling early: local extrema can only occur at interior turning points or at endpoints, and endpoints are treated differently by different textbooks. Most first-year courses exclude endpoints from "local maximum" definitions because the neighborhood has to exist on both sides. If your course includes them, an endpoint counts whenever it beats its one-sided neighbor. Check your syllabus before an exam.
The Four-Point Visual Check That Separates Peaks from False Alarms
A usable screening method takes about ten seconds per candidate and catches the errors that cost marks. Work left to right across the graph, mark every place where the curve flattens or reverses, then run the same four tests on each mark. The order matters because the cheapest check eliminates the most candidates.
1: Find the turning points. Scan the curve for spots where the tangent is horizontal — the moment the curve stops climbing and starts dropping, or stops dropping and starts climbing. On paper these look like rounded tops and bottoms. 2: Check the side on the left. Confirm the curve was genuinely rising before the point (for a maximum) or genuinely falling (for a minimum). A flat shelf that never rose is not a peak. 3: Check the side on the right. Confirm the curve falls after a maximum or rises after a minimum. If both sides rise, you are at a valley, not a hilltop. 4: Compare with immediate neighbors. Read the y-value of the candidate against points just left and right at the same scale. Strictly greater than both means a maximum; equal to a neighbor on a plateau needs the plateau rule below.
If a candidate fails step two or three, you are looking at an inflection or a shelf, not an extremum. Redo the scan on whichever side failed — the actual turning point is usually one grid square away, and misreading a steep curve at low resolution is the single most common graph-reading error. Sharp corners rather than smooth arcs signal a cusp, where the derivative may not exist but a local extremum can still occur.
Sharp Corners, Cusps, and Flat Shelves: Where Graph Reading Breaks
Smooth curves are the easy case. The three failure scenarios below are where visual inspection produces wrong answers, and each has a rule that resolves it. If your graph shows any of these shapes, apply the specific rule rather than the smooth-curve habit.
| Graph feature | What it looks like | Correct classification | Rule that settles it |
|---|---|---|---|
| Sharp corner (cusp) | Two straight-ish segments meeting at a point | Local extremum if it is the highest or lowest point nearby | Derivative is undefined at the corner; the value test decides |
| Flat shelf (plateau) | Horizontal segment of non-zero length | Every point on the shelf is a local extremum | The whole segment attains the same extreme value; some textbooks count the endpoints of the shelf as the named extrema |
| Vertical asymptote | Curve shoots up or down and never turns | No extremum | The function is unbounded; a "peak" at the asymptote is not a finite value |
| Discontinuity jump | Curve breaks and restarts higher or lower | Endpoint-style judgment on each fragment | Each fragment is assessed against its own one-sided neighborhood |
Ap Calculus rubrics penalize naming a cusp as a maximum only when the derivative is required; the value-based definition still allows it, which is why the cusp row above says "value test decides." Watch for this in exam settings, where the justification language matters as much as the answer.
Flat shelves trip up more students than cusps. A horizontal segment of finite length means every point on that segment reaches the same y-value, so every point is technically a local maximum or minimum depending on which side it sits. Most exam keys accept the endpoints of the shelf, some accept any point on it, and a few accept "the entire interval." Read the marking scheme if you can; if you cannot, state both the interval and its endpoints.
Asymptotes produce a distinct error: a curve that climbs toward a vertical asymptote looks like it has a peak just off-screen. It does not — there is no finite y-value to call a maximum, and the neighborhood test fails because the function is unbounded as it approaches. Discontinuities behave similarly, with each continuous fragment needing its own assessment rather than a judgment across the break.
Numbered Walkthrough: Reading Every Extremum off a Sample Curve
Worked examples make the method stick, so here is a complete read of a representative graph: a cubic-style curve that rises from the lower left, peaks at x = −2 with y = 5, drops through the origin area to a valley at x = 1 with y = −4, and climbs past x = 4. The numbers are illustrative and match the shape, not a specific published problem. The walkthrough shows the exact order of operations, including where to write each answer.
1: Mark the domain. Note the x-range shown on the axis. If the domain is stated as all real numbers, no endpoint checks are needed; if a closed interval like [−4, 5] is drawn, mark both ends for later. 2: Scan left to right and flag turning points. On this curve, flag x = −2 and x = 1. Ignore everything else — the rise past x = 4 has no turning point within the frame. 3: Classify each flagged point. At x = −2 the curve rises to the left and falls to the right, so it is a local maximum. At x = 1 it falls to the left and rises to the right, so it is a local minimum. 4: Read the y-values at each extreme. The peak sits at y = 5; the valley sits at y = −4. Record both coordinates, not just the y-values — exam answers usually want the pair. 5: Compare for global status. The highest value anywhere on the curve is 5, so x = −2 is also the global maximum on the shown domain. The lowest is −4, making x = 1 the global minimum, unless an endpoint beats it. 6: Check endpoints when the domain is closed. At x = −4 the curve reads roughly y = −1, well below the peak, so it is neither a global nor a local maximum under most conventions. At x = 5 the value is roughly y = 4, again below the peak but above the valley. 7: Write the answer as coordinates plus classification. Format like "local maximum at (−2, 5); local minimum at (1, −4); global maximum at (−2, 5) on the shown domain." Add endpoint values separately if your course includes them.
Failure handling matters more than the happy path. If step three leaves you unsure whether the curve rises on the left, trace backward two grid squares and compare heights with the candidate. If a turning point falls exactly on a grid line that is hard to read, cite a range and flag it — "local maximum near x = 5" with the y-value read to the nearest half-unit beats a confident wrong decimal. And if the graph has no grid at all, so that you cannot read coordinates, the answer is the location description only; never invent coordinates from an unmarked picture.
Reading Extrema When the Graph Has No Gridlabels or Scales
Graphs without labeled axes are common in testing, and they demand a different answer format. When no scale is printed, you cannot produce coordinates, so the correct response names the turning points by position and character instead of by number. That is not a dodge — it is the accurate answer the graph supports.
Order the turning points left to right and label them with a description plus a count. "A local maximum at the first marked peak, then a local minimum at the next turning point, then a second local maximum before the right endpoint" is a complete, defensible read. If the question asks how many local minima exist, the count is the answer.
The gray-scale or sketch setting adds a second problem: relative height comparisons get harder as resolution drops. Two peaks that differ by half a grid square can look identical at print size, so the global maximum becomes ambiguous. State the ambiguity rather than picking one — exam rubrics generally award the setup credit for identifying that the comparison is not resolvable.
A useful heuristic applies when the graph has no scale but the question implies one exists: assume uniform spacing between grid lines unless told otherwise, read each turning point to the nearest line, and carry the uncertainty forward into the answer as a stated range. That gives the grader something to credit even when the exact value is unreadable.
Checking Your Answer Against the Slope-Sign Test
Once you have a candidate list, the slope-sign test confirms or rejects each entry in one pass and doubles as a cross-check against derivative-based answers. The test reads the curve's local direction: moving left to right, note the sign of the vertical change just before and just after each candidate.
| Sign pattern left → right | Verdict | Typical textbook label |
|---|---|---|
| + then − | Local maximum | Rising to falling |
| − then + | Local minimum | Falling to rising |
| + then + | Neither; curve kept rising | No extremum |
| − then − | Neither; curve kept falling | No extremum |
| + then 0 then + | Terrace, not a peak in the strict sense | Check the shelf rule |
Apply the test to the sample curve from the walkthrough: at x = −2 the direction is positive then negative, confirming a maximum; at x = 1 it is negative then positive, confirming a minimum. The test takes under a minute for a five-point graph and catches every smooth-curve error.
Run it in both directions when you have time. Mentally walking right to left gives the same verdict reversed, and disagreement between the two directions means you misread one of the neighborhoods. The most frequent cause is a grid square that looked flat but actually slopes gently.
A second independent check uses the second derivative concept even though you have no equation: a peak that curves downward (concave down) is a maximum candidate, one that curves upward is a minimum candidate. If the slope-sign test says maximum but the curvature clearly opens upward across the whole region, re-examine the drawing or the reading — one of them is off.
What Top Tutorials Leave Out About Domain and Endpoint Conventions
Standard tutorials demonstrate the turning-point scan and stop there, which leaves four practical gaps that show up in real coursework and testing. Each gap below is a rule the top-ranked explanations typically omit or mention only in passing.
The first is domain ambiguity. If the graph is drawn as a figure without an explicit domain statement, ask what the drawn curve implies. A curve that visibly levels off at the edge of the frame is usually treated as continuing; a curve cut off mid-rise has an endpoint whose status depends on the course. The American convention in most single-variable texts treats endpoints as valid extrema when they dominate their one-sided neighborhood, while rigorous analysis texts exclude them. Name the convention you are using.
The second is the global-versus-local ranking when a graph has multiple equal peaks. If two hills reach the identical y-value and nothing exceeds them, both are global maxima, and the global maximum is the value, not a single point. The third is that labels like "absolute maximum" and "global maximum" are interchangeable in most textbooks but not all — read which term your course uses. The fourth is that the count of extrema can change with the viewing window; zooming out can hide a small local valley entirely.
These gaps matter because they change answer format, not the underlying method. A student who has memorized the scan but never met the endpoint convention loses credit on closed-interval problems, and one who has only seen single-peak examples misreads the multi-peak case. The fix is to state the convention explicitly in your answer and to rank every extremum by value, not just name the turning points.
Graph-Reading Versus Derivative Methods: Which to Use, and When
Graphs and derivatives answer the same question through different evidence, and knowing when each is the right tool saves time on tests and in applied work. The table compares the two methods on the dimensions that actually decide which one to reach for.
| Criterion | Reading the graph | Using the derivative |
|---|---|---|
| Data needed | A drawn curve with readable axes | An equation you can differentiate |
| Best for | Piecewise, empirical, or hand-sketched functions | Clean algebraic functions |
| Endpoint handling | Visible; judgment needed | Requires domain restriction |
| Cusps and corners | Detected directly | Fails where derivative is undefined; needs the value test |
| Confidence | Limited by grid resolution | Exact within arithmetic |
| Typical error | Misreading a steep section | Sign errors in the derivative |
| Time for a five-point graph | About one minute | Depends on the algebra; usually longer for messy functions |
Derivative work gives exact locations, which matters when the question wants precise coordinates or when the graph is only a sketch. Graph reading wins when no equation exists — experimental data, piecewise definitions, or a word problem that hands you a picture and nothing else. Keep both skills because each covers the other's blind spot.
The practical split: use the derivative whenever the function is given in algebraic form and you are asked for exact answers; use graph reading whenever the function is presented visually, described piecewise, or the equation is not differentiable at the suspected extremum. For a hybrid problem — an equation plus a rough sketch — solve with the derivative and confirm with the picture.
One caution about graphs generated by software: plotting tools sample at discrete points, and a narrow peak between samples can vanish from the display. When a graph comes from a calculator or plotting app, distrust its peak count until you have also checked the underlying function if one is available.
The Same Peak Can Be Local and Global: A Worked Ranking
Ranking every extremum by value prevents the most common mix-up, where a local maximum gets called the global one just because it appears first. The sample curve from the walkthrough makes a clean case: two turning points, one above the other.
At the walkthrough curve's extremes, x = −2 holds y = 5 and x = 1 holds y = −4. Both are local extrema, and the higher one is also global because nothing on the shown domain exceeds 5. Flip the picture mentally — mirror it underground — and the old minimum becomes the global one. The classification of any single point depends on what it is compared against, so state the comparison explicitly: "local maximum at (−2, 5), which is also the global maximum over the shown domain."
Multi-peak graphs go further. On a curve with three local maxima at y = 7, y = 3, and y = 7, the global maximum is 7 and it occurs at two distinct points, both also local maxima. The local minimum between them at y = −1 is neither global nor close to it, but the global minimum is whichever valley is lower across the full domain — and if the domain is closed, check both endpoints.
Two mistakes commonly break this logic. The first is calling every local maximum "the maximum," ignoring that several may tie or compete. The second is reversing maximum and minimum because the peak looked "bigger" than the valley on screen; the labels track direction of the turn, not visual size on a squashed axis, and a graph with unequal axis scaling can make a shallow rise look taller than a deep fall. Read the axes before comparing heights.
Practicing With Real Graph Shapes: Four Patterns to Recognize
Recognition speeds up with exposure to the four shapes that account for nearly every graph-reading question. Each pattern has a signature you can learn in seconds, and spotting the signature cuts the scan from a full left-to-right sweep to a glance.
The smooth hill-and-valley pair — one peak, one dip — is the base case: check the turning points, read the values, compare. Oscillating curves such as a sine-style wave repeat peaks and valleys at regular intervals; count the cycles within the shown domain and multiply, taking care about whether the domain cuts a cycle in half. Piecewise curves have corners at the joins, and any join where the curve switches from rising to falling is a candidate extremum worth classifying. Curves with a flat shelf sit between the easy and hard cases; identify the plateau's left and right ends and report the interval.
For practice, most calculus textbooks publish a problem set of "identify local and global extrema from the graph" questions, and the Stewart Calculus chapter on maximum and minimum values is a standard source. Work five or six of those until the scan is automatic, then add a timing constraint: two minutes for a graph with six or fewer points is a reasonable pace.
On the sensory side, a graph has its own feel. A smooth peak reads as a rounded crest you could roll a ball over; a cusp feels sharp enough to catch on, and a plateau feels like a ledge. Those shapes cue the rule before the numbers do, and practicing with your finger on the curve builds the intuition faster than staring at coordinates.
For anyone working beyond coursework — reading optimization outputs, interpreting plotted survey data, examining a performance chart — the method stays identical. The vocabulary is the same whether the curve is a textbook function or a measured series, and the conventions are the only part that needs checking against local practice.
FAQ
Can a graph have a local maximum without a horizontal tangent?
Yes. A sharp corner can be a local maximum even though the derivative does not exist there, because the definition of local extremum rests on neighboring values, not on slope. Textbooks classify these with the value test rather than the derivative. Check whether your course counts corners as extrema before writing an answer either way.
Is a local maximum always the highest point on the graph?
No — local means best in the immediate neighborhood only. A curve can have three local maxima at y = 7, y = 3, and y = 7 while the global maximum is 7, attained at two of them. Ranking every extremum by value is the way to keep the terms straight.
What happens to the extrema count when you zoom a graphing app in or out?
Zooming changes what the display samples, so a narrow peak can appear or disappear between frames. Plotting tools evaluate the function at discrete points, and an extremum between two samples may not be drawn at all. Trust the algebra when an equation exists and treat the picture as a sketch.
How do endpoints count on a closed interval?
Conventions differ. Many first-year calculus courses count an endpoint as a local extremum when it beats its one-sided neighbor, treating it as a valid local endpoint maximum or minimum. More rigorous treatments require a two-sided neighborhood and exclude endpoints, so state your convention in the answer.
Why are two equal peaks both called global maxima?
The global maximum is a value, not a location. If the highest y-value on the domain is reached at two different x-values, both points are global maxima and the global maximum is still a single number. List every point that attains it.
Does a graph with a vertical asymptote ever have a maximum?
No finite maximum exists where the curve is unbounded, so a curve climbing toward a vertical asymptote has no extremum at that approach. The neighborhood test fails because no finite y-value sits above nearby points. Assess only the bounded turning points; if a suspected peak abuts an asymptote, rule it out.
Do flat horizontal shelves count as maxima or minima?
A non-zero-length horizontal segment reaches the same y-value at every point on it, and the extremum is that value. Exam keys usually accept the shelf's endpoints as the named extrema, some accept any point on the shelf, and a few accept the whole interval. State both the interval and its endpoints to cover the grading.