What a limit is, and why f(a) is beside the point
The limit of f as x approaches a is the value the outputs close in on as the inputs close in on a — deliberately excluding a itself. The exclusion is not a technicality. The whole reason limits exist is to describe functions at points where they are undefined or misbehaved.
The clearest case is the derivative. The difference quotient [f(x+h) − f(x)]/h is undefined at h = 0, which is precisely the value you care about. The limit gives you the answer that direct substitution cannot.
The second reason is that a function can be defined at a point and still not equal its limit there. That mismatch is the definition of a discontinuity, and it is why substitution is a shortcut rather than a proof. The calculator reports f(a) and the limit as separate outputs so you can see when they disagree.
The formal definition is a promise about tolerances. For every tolerance ε you name, there is a distance δ such that every x within δ of a (other than a itself) has f(x) within ε of L. That epsilon-delta statement is what the numerical approach on this page illustrates: the step shrinks by a factor of ten each row, and if the column settles, the promise is being kept at ever tighter tolerances.
How this calculator finds the limit, and what that method can and cannot see
For a finite point, the calculator evaluates f(a − h) and f(a + h) for six values of h, starting at one tenth of the scale of the problem and dividing by ten each time down to one part in a million. It then examines the last two entries in each column. If they agree to about four significant digits, that column has converged and the value is reported as the one-sided limit. If the magnitude is large and still growing by more than a factor of two and a half per row, the column is reported as unbounded.
For a limit at infinity, it evaluates at ten, one hundred, and so on up to ten million, and applies the same test.
The two-sided verdict compares the two columns. Both finite and equal gives a limit. Both unbounded in the same direction gives an infinite limit. Anything else — different finite values, or opposite infinities — means no two-sided limit exists, and the calculator reports a blank rather than a number.
The method's limits are real and worth stating. It cannot see behaviour closer than one part in a million of the scale, so a feature that only appears at 1e−9 is invisible. It is defeated by oscillation: sin(1/x) near zero never settles, and the calculator says so rather than reporting whichever value the last sample happened to take. And it is defeated by catastrophic cancellation, which is why the step stops at 1e−6 rather than pushing to 1e−15 — subtracting two nearly equal floating-point numbers destroys precision faster than the smaller step recovers it.
A numerical approach is evidence, not proof. For a proof you factor, rationalise, or apply L'Hôpital's rule. What this page gives you is a reliable check on whichever algebra you did.
Worked example: three limits at x = 0 and x = 1
First, sin x over x at zero. Substitution gives 0/0, which is an indeterminate form and tells you nothing. The approach table settles quickly: at h = 0.1 both sides give 0.9983342, at h = 0.01 both give 0.9999833, and at h = 0.001 both give 0.9999998. The limit is 1.
You can prove it three ways. Geometrically, the squeeze theorem traps sin x / x between cos x and 1. By L'Hôpital, differentiate top and bottom to get cos x / 1, which is 1 at zero. By series, sin x = x − x³/6 + …, so sin x / x = 1 − x²/6 + …, which tends to 1.
Second, a removable discontinuity. For (x² − 1)/(x − 1) at x = 1, substitution again gives 0/0. Factor the numerator: (x − 1)(x + 1)/(x − 1) = x + 1 for every x except 1. So the limit is 1 + 1 = 2, and the graph is the line y = x + 1 with one point punched out. The calculator returns 2 and reports f(1) as blank.
Third, a limit at infinity. For (2x² + 3)/(x² − 5), divide top and bottom by the highest power: (2 + 3/x²)/(1 − 5/x²). As x grows both fractions vanish and the expression tends to 2/1 = 2. The table confirms it: at x = 10 the value is 2.1368421, at x = 100 it is 2.0013007, at x = 1000 it is 2.0000130. The general rule for a ratio of polynomials of equal degree is the ratio of the leading coefficients, which here is 2/1.
How to read the verdict
A blank limit with two finite one-sided values means a jump. abs(x)/x at zero gives −1 on the left and +1 on the right. The function is perfectly well behaved on each side; it simply has two different destinations, and no single value can be the limit.
A blank limit with opposite infinities means a pole with a sign change. 1/x at zero falls without bound on the left and grows without bound on the right. Some texts write this as an infinite discontinuity; none of them write a limit for it.
An infinite limit is a description, not a number. When the calculator reports that both sides grow without bound, as for 1/x² at zero, that is a statement about how the function fails to converge, not a value you can substitute into later algebra.
A limit that exists while f(a) does not is the ordinary case. Every interesting limit in a first calculus course has this shape, because the point of the exercise is to fill a hole in a function.
A limit that exists and differs from f(a) means a removable discontinuity that has not been removed. The calculator warns explicitly, because this is the case where substitution silently produces the wrong answer.
Standard limits worth recognising on sight
| Limit | Value | Form on substitution | Technique |
|---|---|---|---|
sin x / x as x → 0 | 1 | 0/0 | Squeeze theorem |
(1 − cos x)/x as x → 0 | 0 | 0/0 | Conjugate multiplication |
(1 − cos x)/x² as x → 0 | 1/2 | 0/0 | Series or L'Hôpital twice |
(1 + 1/x)^x as x → ∞ | e = 2.7182818 | 1^∞ | Definition of e |
(e^x − 1)/x as x → 0 | 1 | 0/0 | Series or L'Hôpital |
ln(1 + x)/x as x → 0 | 1 | 0/0 | Series or L'Hôpital |
(x² − 1)/(x − 1) as x → 1 | 2 | 0/0 | Factor and cancel |
(2x² + 3)/(x² − 5) as x → ∞ | 2 | ∞/∞ | Divide by the highest power |
x·ln x as x → 0⁺ | 0 | 0·∞ | Rewrite as a quotient, then L'Hôpital |
sin(1/x) as x → 0 | does not exist | oscillates | No technique applies |
Most of these reproduce here directly; the <code>x·ln x</code> row needs the direction set to <em>right</em>, because the function has no real value on the left. Two rows are instructive failures rather than confirmations. <code>sin(1/x)</code> is reported as no stable value because it oscillates instead of settling, which is the point of the row. <code>x·ln x</code> approaches zero so slowly that it still misses the convergence test at the smallest step used, so this page reports no stable value there too — a fair warning that sampling proves nothing on a slowly converging limit.
Traps in limit problems
- Treating 0/0 as zero or as one. It is an indeterminate form, meaning the answer depends entirely on which functions produced it.
x/xgives 1,x²/xgives 0 andx/x²is unbounded, all at zero. - Substituting and stopping. Substitution is valid only when the function is continuous at the point. If it is, you did not need a limit; if it is not, substitution can give the wrong answer.
- Assuming a two-sided limit exists because one side does. Both sides must converge, and to the same value. Check them separately, which is why they are separate outputs here.
- Reporting infinity as a limit value. An infinite limit is a statement that the limit fails to exist in a particular way. Writing it in an equation as though it were a number leads to errors later.
- Applying L'Hôpital's rule to a form that is not indeterminate. It applies only to 0/0 and ∞/∞. Using it on
(x + 1)/(x + 2)at zero converts a correct answer of 0.5 into a wrong answer of 1. - Trusting a numerical table alone. Sampling cannot distinguish a limit of 0 from a function that dips to 1e−9 between the samples. Use the table to check algebra, not to replace it.
Where limits lead
Continuity is defined by a limit. A function is continuous at a when the limit exists, f(a) exists, and the two are equal. All three conditions are needed, and each one fails in a recognisable way: a hole, a jump, or a pole.
Derivatives are limits. f′(a) is the limit of the difference quotient, and every rule in the derivative calculator is a theorem proved from that limit.
Integrals are limits. The definite integral is the limit of Riemann sums as the mesh shrinks, which is what the Riemann sum calculator illustrates numerically.
Series are limits of partial sums. Whether a series converges is a question about the limit of its partial sums, and the tests in the ratio test calculator are shortcuts for answering it without computing that limit directly.
Asymptotes are limits at infinity. A horizontal asymptote at y = L is exactly the statement that the limit at infinity is L, and a vertical asymptote at x = a is the statement that the one-sided limits there are unbounded.
For indeterminate forms specifically, L'Hôpital's rule converts 0/0 and ∞/∞ into a limit of derivatives, which is usually easier and often immediate.