Put simply, a function is continuous at a point when three things line up: the function is defined there, the two-sided limit exists there, and the limit equals the function's value. On a graph, you check those conditions visually in one order — find the x-value, look at the height of the curve, compare the approach from the left with the approach from the right, then see whether the hole, jump, or spike sits exactly on the curve. A graph that looks unbroken at normal zoom can still fail any one of those checks, so the visual pass is a filter, not a proof.
The three-condition definition comes from the standard limit-based treatment of continuity taught in calculus sequences and codified in textbooks such as Stewart's Calculus and Thomas' Calculus. Those sources agree on the order of operations and on the classification of the ways continuity fails. What they rarely emphasize is how often a graphing window hides a break. That's where most student errors start, so the method below front-loads the rescaling habit before it gets to the interesting cases.
The Three Conditions You Are Checking, in the Order That Saves Time
Start with the value, not the limit. If f(c) doesn't exist, you're done — the point fails continuity no matter how well-behaved the curve looks nearby. This first check eliminates the simplest failure: a vertical asymptote or an isolated missing point where the function simply has no output.
Next, evaluate the one-sided limits. Trace the curve from the left of c and read the y-value it approaches; trace from the right and do the same. If those two numbers differ, the two-sided limit doesn't exist, and the graph shows a jump. If they match, the shared number is the candidate limit.
Third, compare the limit to the value. A curve that approaches the same height from both sides but has a filled dot somewhere else at x = c fails here — the limit exists, the value exists, and they disagree. That's the removable discontinuity in its most common form.
The order matters for a practical reason: it prevents you from doing limit work on a point that was never a candidate. A quick check of the graph's y-axis at x = c takes seconds, and it settles whether the remaining two checks are even worth running.
The Hole, the Jump, and the Spike: Three Named Failures to Recognize on Sight
A removable discontinuity, also called a hole, looks like a break in the curve with an open circle at one point. The limit exists there; the value either doesn't or sits somewhere else. Stewart's treatment calls this removable because redefining f(c) to equal the limit repairs the function without changing its behavior anywhere else — a useful diagnostic test. If you can patch the graph with a single dot, the discontinuity is removable.
A jump discontinuity shows the curve terminating at one height and resuming at another, with a visible gap between the two branches. The one-sided limits exist separately and disagree, so no amount of redefining f(c) can fix it. A common real example is a step function like the greatest-integer function, where the value changes by exactly 1 at every integer. Any graph where the left branch ends at an open circle and the right branch begins at a filled circle at a different height is a jump.
An infinite discontinuity, sometimes called an essential or asymptotic discontinuity, appears where the curve shoots toward positive or negative infinity as x approaches c. The limit doesn't exist as a finite number, so continuity fails outright. The reciprocal function 1/x at x = 0 is the standard case, and it's the failure students most often misread as "just a steep part."
A fourth type — the oscillating discontinuity — shows up in functions like sin(1/x) near x = 0, where the graph wiggles infinitely often in any window around the point. No graphing window can render this honestly at normal scale, so treat any unusually dense cluster of ink near a vertical line as a candidate for numerical investigation rather than visual judgment.
Why a Smooth-Looking Curve Can Still Be Discontinuous at Your Zoom Level
A graphing calculator's default window typically spans something like x from −10 to 10 and y from −10 to 10 on a TI-84 Plus CE, a current model whose defaults have been stable across several generations. That window compresses a jump of 0.001 into a fraction of a pixel — invisible. Zoom in by a factor of 100 and the jump appears as a vertical gap.
Piecewise functions built from a calculator's own syntax are the usual culprit. Enter a function with a conditional branch that shifts by a small amount at x = 2, graph it, and the line looks continuous. Change the window to x from 1.9 to 2.1, and the break materializes. The lesson isn't that calculators lie; it's that resolution determines what you can see, and the visual test is only as good as the window you chose.
The same trap appears in the other direction. A function with a steep but finite slope near a point can look like a vertical asymptote in a wide window. Zooming in reveals that the curve actually crosses through, and the point is continuous. Before classifying anything as infinite, check whether the apparent asymptote survives a zoom that would resolve a large finite slope.
Numerical tables bypass the resolution problem. On a TI-84 Plus CE, the table setup screen (second function key, then the table key) lets you start a table at a chosen x-value and step by a chosen increment. Setting the start to 1.999 and the step to 0.001 shows the function's actual values approaching 2 from below — often the fastest way to distinguish a jump from a steep climb. The same logic works in Desmos by typing an expression and using the zoom controls, though Desmos's table feature requires typing explicit values rather than stepping automatically.
Reading One-Sided Limits From a Graph Without Getting Fooled by Open and Filled Circles
The open-versus-filled circle convention answers the value question directly. A filled dot at x = c means f(c) equals the height of that dot. An open circle means f(c) is undefined at that height — but the curve may still be defined elsewhere at x = c, which is exactly the removable case. If both an open circle and a filled dot appear at the same x-value at different heights, the filled dot wins: that's f(c), and the gap between it and the limit is the discontinuity.
Trace slowly. Starting from x = c − 0.5 and moving toward c, read the y-values off the curve; you're following the left-hand branch. Repeat from the right. Most graph-reading errors come from tracing the wrong branch at a corner or from assuming the curve continues through a gap because a dashed segment suggests it. There is no dashed segment in a correct graph — if you see one, the source is illustrating an asymptote or a limit, not the function itself.
For a graph drawn by hand or by a teacher, the convention holds: the function's own curve is solid, and markers indicate endpoints of removable pieces. For a computer-generated plot, the software draws the curve wherever it is defined, so a visible break is real and a filled dot appears only if the function actually attains that value. This distinction matters when you're checking homework: a printed graph may have stylistic gaps that the software version wouldn't show.
The practical takeaway: open and filled circles are information about the value, and the curve's approach is information about the limit. Read them separately, then compare.
A Worked Example Using a Named, Real Function
Take f(x) = (x² − 1)/(x − 1). This is a standard textbook example, appearing in Stewart's Calculus and in the AP Calculus Course and Exam Description published by the College Board, which lists it among the functions students should be able to analyze. The graph is the line y = x + 1 with a hole at (1, 2).
Check the conditions in order. f(1) is undefined because the denominator is zero — the first condition fails, so continuity fails immediately. The limit as x approaches 1, factored as (x − 1)(x + 1)/(x − 1) = x + 1, equals 2 from both sides. The limit exists, the value doesn't. This is the textbook removable discontinuity: a single open circle at (1, 2), repairable by defining f(1) = 2.
Now compare with g(x) = (x² − 1)/(x − 1) for x ≠ 1, with g(1) = 0. The limit is still 2, but the value is 0. The graph shows the line with an open circle at (1, 2) and a filled dot at (1, 0). The removable discontinuity is still removable — you'd redefine g(1) to 2 — but a student checking only "does the graph look continuous near 1" sees a filled dot and may call it continuous. The dot is at the wrong height, and the gap between 0 and 2 is the whole point.
A third version: h(x) = 1/(x − 1). At x = 1, no value exists, the left limit is negative infinity, and the right limit is positive infinity. This is the infinite discontinuity, and no redefinition repairs it.
Continuous Versus Merely Undefined: What the Closed Interval Adds
A function continuous at every point of an open interval is continuous on that interval. The extra work comes at endpoints: continuity on a closed interval [a, b] requires continuity on (a, b), plus the right-hand limit at a equaling f(a), plus the left-hand limit at b equaling f(b). The College Board's AP Calculus framework lists this endpoint condition explicitly, and it's the reason a function can be continuous on (0, 1) but not on [0, 1].
On a graph, endpoint continuity shows up as a filled dot at the end of the curve with the curve approaching that height from inside the interval. If the endpoint is an open circle instead, the function is not defined at that endpoint in the relevant way, and interval continuity fails at that end.
This distinction matters for the Intermediate Value Theorem, which requires a function continuous on [a, b] to guarantee a root between any two values of opposite sign. Students who skip the endpoint check sometimes invoke IVT on a function that fails at an endpoint and get a contradiction with the graph — a useful diagnostic when something seems off.
For a piecewise function defined over [0, 4] with branches meeting at x = 2, interval continuity requires checking each branch's interior, the junction point, and both endpoints. That's four checks, and a graph that looks clean in the middle can still fail at the endpoint.
Side-by-Side Comparison of the Five Failure Modes
The table below lines up the visual signature, the limit behavior, the value behavior, and whether the discontinuity is repairable for each type. Use it as a labeling tool once you've identified that a discontinuity exists; the identification method in the next section tells you which row you're in.
| Failure type | What the graph shows | Left limit vs right limit | f(c) defined? | Repairable by redefining f(c)? |
|---|---|---|---|---|
| Removable (hole) | Open circle on an otherwise continuous curve | Equal | No, or wrong value | Yes |
| Jump | Two branches ending at different heights | Unequal | Yes | No |
| Infinite | Curve running off toward ±∞ at a vertical line | ±∞ on at least one side | No | No |
| Oscillatory | Dense, non-resolving wiggle near a vertical line | Doesn't exist | Usually no | No |
| Endpoint | Filled or open dot at the end of a curve on a closed interval | One-sided only | Depends on the marker | Only if the endpoint value can be set to the interior limit |
The last column is the practical payoff. If the answer is "yes," you're looking at a removable discontinuity and the graph's only defect is a missing or misplaced dot. If the answer is "no," the function is genuinely discontinuous at that point no matter how you define it there, and you should classify it and move on.
Notice the oscillatory row: the right limit doesn't exist as a finite number and neither does the left, so it's sometimes lumped with infinite discontinuities. It isn't the same thing. The function stays bounded — sin(1/x) never leaves [−1, 1] — but it doesn't settle. Textbooks separate the two because the repair question and the limit behavior differ.
Procedure: Deciding Continuity From Any Plot in Four Moves
The procedure below works for hand-drawn graphs, textbook figures, calculator plots, and Desmos screenshots. It's written as a sequence because order matters, and it assumes you're checking continuity at a specific point — usually one flagged by the problem or by a visible anomaly.
1: Identify the x-value in question and locate the vertical line x = c on the plot. If the problem names a point, use it; if you're scanning for discontinuities, look for breaks, circles, or vertical lines the curve seems to approach. 2: Check whether the curve has a point at x = c. Look for a filled dot. If you find an open circle or nothing, f(c) is undefined and continuity fails — record that and stop. 3: Trace the curve toward c from the left and from the right. Read the y-value each branch approaches. If the two numbers disagree, you have a jump. If either runs off the top or bottom of the window, suspect an infinite discontinuity and confirm numerically. 4: Compare the common limit value, if it exists, to f(c). If they're equal, the function is continuous at c. If they differ, it's a removable discontinuity — a misplaced value rather than a missing limit.
Handle failures as they arise. If step 3 produces a limit that seems not to exist but the curve looks bounded, generate a table of values near c (on a TI-84 Plus CE, use the table setup to step by 0.001 from c − 0.01 to c + 0.01) and check whether the values oscillate or converge. If step 4's comparison fails and the problem asks whether the function could be made continuous, the answer is yes — define f(c) as the limit.
The single most common failure of this procedure: skipping step 2 and reading the limit as the value. On a graph with a filled dot away from the hole, the eye goes to the line, not the dot. Slow down at step 4.
The Asymptote Trap: When "Runs Off the Graph" Doesn't Mean Infinite
An asymptote is a line the curve approaches, and the curve's approach doesn't establish what happens at the x-value itself. A function can have a vertical asymptote at x = c, fail to be continuous there, and still be redefined to make it continuous if the limit happens to exist. The function sin(x)/x at x = 0 has a removable discontinuity, not an asymptote — the limit is 1, and the standard convention in calculus texts (including Stewart) defines sin(x)/x = 1 at 0, which makes it continuous.
Conversely, a steep curve can look asymptotic in a wide window and turn out to be perfectly continuous. The exponential function e^x at x = 10 looks like it hits the top of a default calculator window, but zoom out and it passes through (10, 22026.47) — a finite value. Treating "off-screen" as "infinite" is a window artifact, not a mathematical conclusion.
The distinguishing test is numerical. Build a table of values approaching c from both sides at decreasing increments. If the values grow without bound (or fall without bound) as you approach, the discontinuity is infinite. If the values converge to a finite number, the limit exists and the apparent asymptote is either removable or an artifact. If the values oscillate within a fixed range, it's oscillatory.
Graphing software helps less than you'd think here, because the software's window is also finite and its resolution is also finite. The table is the reliable tool. Use it whenever the visual evidence is ambiguous, and trust the numbers over the pixels.
What Teachers and Graders Actually Mark Down
Course descriptions and textbook solutions manuals converge on the same expectations: name the type, state the limit, state the value, and write the conclusion explicitly. The AP Calculus Course and Exam Description, published by the College Board, requires students to justify continuity using the limit definition rather than the graph's appearance — a graph is evidence, not proof. Solutions in Stewart's manual follow the same structure, which suggests the convention is stable across exam systems.
A frequent partial-credit loss is calling a jump a removable discontinuity because both branches end at the same x-value with a visible gap. The gap is the problem, not a detail — a jump cannot be repaired by redefining one point. Another is asserting "the graph is continuous" without addressing a specific point, when the question named one. Graders want the three conditions stated or implied: value, limit, equality.
A third is treating a filled dot as sufficient evidence of continuity when the curve isn't approaching it. If the dot sits at (2, 5) and the curve approaches (2, 3) from both sides, the function is discontinuous at 2 despite the defined value — the limit and the value disagree. The dot's presence answers only condition one.
The habit that avoids all three: write out f(c), the left limit, the right limit, and the comparison. Four lines, in that order, every time. A graph can be read; a function's continuity at a point has to be stated in the language of limits, and the four-line format forces that statement.
FAQ
Does a hole in the graph always mean the function is discontinuous there?
Yes, at the hole's x-value. A removable discontinuity means the function is either undefined or wrongly defined at that point — either way, continuity fails there. The limit may exist and the function may be repairable, but "repairable" and "continuous" are different claims. A graph with a hole is discontinuous precisely at the hole.
Can a function be continuous at a point even if the graph looks broken at it?
Yes, if the break is a rendering artifact. Steep slopes, dense oscillations, and resolution limits can all produce visual breaks that don't correspond to discontinuities. A function is continuous at c if the limit exists there and equals f(c), and that can be true at a point where a low-resolution plot shows a vertical gap. Zooming in or building a table of values resolves the question.
What's the difference between a removable discontinuity and an undefined point?
A removable discontinuity is a specific kind of undefined (or misplaced) point where the two-sided limit exists. An undefined point in a function with no limit — like 1/x at x = 0 — is a different failure. The test is whether redefining f(c) as the limit would repair it; if yes, it's removable. If the limit doesn't exist, redefinition can't help.
How does continuity on a closed interval differ from continuity at a point?
Continuity on a closed interval [a, b] requires continuity at every interior point plus one-sided continuity at each endpoint. Endpoint continuity means the limit approaching from inside the interval equals the endpoint value. A function continuous on (0, 1) can fail on [0, 1] if the endpoints aren't defined or aren't approached correctly.
Do I need calculus to determine continuity from a graph?
For a visual read, no — the three tests (value, limit, equality) can be applied by eye. For a defensible answer, yes: continuity is defined in terms of limits, and stating a conclusion without limit language is incomplete. The AP Calculus framework requires limit-based justification, and textbooks follow the same convention.
Is sin(1/x) continuous anywhere near x = 0?
The function is continuous at every nonzero x — it's a composition of continuous functions — but it's discontinuous at 0, where the oscillation prevents any limit from existing. The graph near 0 looks like a dense vertical smear in any normal window. The discontinuity is oscillatory, not infinite, since the values stay within [−1, 1].
Can a graph show a discontinuity that isn't real?
Yes, in two ways: resolution can hide a discontinuity, or a rendering choice can invent one. A plot that draws a vertical connector between two branches of an asymptote — some software does this — shows a break that doesn't exist as a curve segment; the true graph has no connector. Read the function's values near the suspected point before deciding the graph is wrong.
What's the fastest check when a graph looks continuous and you only have a minute?
Test one interior point of the visible segment by table, stepping through it at increment 0.001. If the values change smoothly across the point, treat the graph as continuous there and move on. The step-by-step procedure covers the cases where this quick pass turns up a surprise; the quick pass itself won't find the surprises.