How to Find the Inverse of a Function: Swap, Solve, Restrict, Verify in Four Steps

You find the inverse of a function by swapping x and y in its equation, solving the new equation for y, and then checking that the result actually undoes the original. The whole method collapses if the function isn't one-to-one, so every honest procedure starts by testing or forcing that property. For a parabola like f(x) = x², the swap-and-solve route produces y = ±√x, which is not a function until you restrict the domain to x ≥ 0. That single constraint decides whether an inverse exists at all, and it's the step most worked examples skip.

Why Only One-to-One Functions Have Inverses

An inverse has to give back exactly one output for every input it receives. If two different inputs of the original function produce the same output, the reverse mapping can't know which one you started from, and the inverse fails the vertical line test. That's why f(x) = x² has no true inverse over all real numbers: f(2) = 4 and f(−2) = 4, so the reverse of 4 is ambiguous. The horizontal line test is the standard screen — draw any horizontal line across the graph and if it touches the curve more than once, the function isn't one-to-one and no unrestricted inverse exists.

Some functions pass automatically. Any strictly increasing or strictly decreasing function is one-to-one by definition, which covers every linear function with a nonzero slope, every exponential such as f(x) = 3^x, and the square root family. Others fail everywhere, including sine, cosine, and absolute value in their natural domains. Polynomials of even degree typically fail because they turn around, while odd-degree polynomials like f(x) = x³ + x always pass since they never reverse direction.

The distinction matters because textbooks often say "find the inverse" without mentioning the domain, and students then write y = ±√x as if it were a function. It isn't — it's a relation with two branches. Naming the restriction up front saves the confusion later, and it's the reason inverse trigonometric functions like arcsin carry restricted domains built into their definitions.

For anything you can graph easily, the test takes seconds and settles the question before algebra begins. For messier expressions, prove it algebraically: set f(a) = f(b) and show the equation forces a = b. With f(x) = 3x + 5, f(a) = f(b) gives 3a + 5 = 3b + 5, so a = b, and the inverse is safe to compute. With f(x) = x² − 4x + 7, the algebra stalls, which flags a problem the graph would have shown anyway.

The Four-Step Procedure That Works on Every Invertible Function

The method is uniform regardless of function type. Write the function as y = f(x), swap the variables, solve for the new y, and relabel. What changes between problems is the algebra in step three, and that's where practice pays off. The steps below run in the order that surfaces errors earliest.

1: Replace f(x) with y. So f(x) = (x + 3)/2 becomes y = (x + 3)/2. This is bookkeeping, but skipping it causes swap mistakes later. 2: Swap x and y. The equation becomes x = (y + 3)/2. Every x becomes y and every y becomes x, including inside roots and exponents. 3: Solve for y. Multiply both sides by 2: 2x = y + 3. Subtract 3: y = 2x − 3. 4: Rewrite as f⁻¹(x) and verify. State f⁻¹(x) = 2x − 3, then check that f(f⁻¹(x)) = x and f⁻¹(f(x)) = x. Both must hold. With x = 5: f(5) = 4, and f⁻¹(4) = 5. With x = 1: f⁻¹(1) = −1, and f(−1) = 1.

When the solve step produces a quadratic in y, you'll need the quadratic formula, and the ± sign is your signal that a domain restriction is waiting. Take f(x) = x² − 6x + 5 with x ≥ 3. Swap: x = y² − 6y + 5. Rearrange: y² − 6y + (5 − x) = 0. The quadratic formula gives y = (6 ± √(36 − 4(5 − x)))/2 = (6 ± √(16 + 4x))/2 = 3 ± √(4 + x). Since the original domain was x ≥ 3, the inverse's range must be y ≥ 3, so you keep the plus branch: f⁻¹(x) = 3 + √(4 + x). Picking the wrong sign is the most common failure here, and checking one point catches it — f(3) = −4, so f⁻¹(−4) must equal 3, and 3 + √0 = 3 confirms the choice.

Failure handling follows a fixed order. If step three stalls on a variable trapped in two places, the function may still be invertible but not expressible in elementary form — f(x) = x + e^x is the classic case, invertible everywhere but with no closed-form inverse. If the solve produces two branches, apply the domain restriction. If verification fails, you either picked the wrong branch or the original wasn't one-to-one on the stated domain.

Worked Examples Across Four Function Families

Linear functions are the gentlest case because the algebra never branches. Take f(x) = −4x + 7. Swap gives x = −4y + 7, so 4y = 7 − x and y = (7 − x)/4. Check with x = 2: f(2) = −1, and f⁻¹(−1) = (7 + 1)/4 = 2. The slope of the inverse is −1/4, the negative reciprocal of −4, which is a general rule for linear inverses and a fast sanity check on any answer you produce.

Rational functions introduce a variable that appears twice, and the fix is factoring. Let f(x) = (2x + 1)/(x − 3). Swap: x = (2y + 1)/(y − 3). Multiply out: x(y − 3) = 2y + 1, so xy − 3x = 2y + 1. Collect y terms: xy − 2y = 3x + 1, then y(x − 2) = 3x + 1, giving f⁻¹(x) = (3x + 1)/(x − 2). Notice the domain swap: the original excluded x = 3, so the inverse excludes x = 2, and the horizontal asymptote y = 2 of the original becomes the vertical asymptote x = 2 of the inverse. That reciprocal relationship between asymptotes is a reliable structural check.

Radicals need the domain carried through carefully. For f(x) = √(x − 4), the domain is x ≥ 4 and the range is y ≥ 0. Swap: x = √(y − 4). Square both sides: x² = y − 4, so y = x² + 4. The inverse's domain must match the original's range, so you write f⁻¹(x) = x² + 4 with x ≥ 0. Drop that restriction and you've invented a function that accepts inputs the original could never produce.

Exponentials and logarithms invert into each other, which is why they're taught as a pair. For f(x) = 2^x, swap to x = 2^y, then take log base 2 of both sides: y = log₂(x). The domain of the inverse is x > 0, matching the range of 2^x. For f(x) = e^(3x − 1), the inverse is f⁻¹(x) = (ln x + 1)/3, and checking x = 1 gives f⁻¹(1) = 1/3 while f(1/3) = e^0 = 1. The pattern holds across every base: exponentiation and logarithms of the same base undo each other, which is the definition of inverse at work.

A Comparison Table for the Four Main Families

The table below compresses the four families onto the same dimensions so you can see where the difficulty lives. Each row reflects the worked example above, and the arithmetic is checkable by substituting the stated test point into both functions.

The verdict splits by family. Linear and exponential inverses are safe to compute mechanically because the algebra can't produce ambiguity — the slope rule and the log rule are deterministic. Rational inverses need the factoring step and a domain check on the excluded value. Radical inverses are the risky ones, since squaring both sides is not reversible without tracking signs, and that's where extraneous solutions creep in. If you're teaching or tutoring, spend the practice time on radicals and rationals; the other two take care of themselves.

One structural point the table can't show: the domain and range always exchange roles. The original's domain becomes the inverse's range, and the original's range becomes the inverse's domain. For the rational example, the original's range excludes y = 2, and indeed the inverse's domain excludes x = 2. That exchange is the single most useful error-detector when you're unsure whether a restriction belongs.

Reading Inverses Off a Graph and the y = x Mirror

Every inverse is the original reflected across the line y = x, and that geometric fact gives you a second route to the answer when algebra gets ugly. If the point (a, b) lies on f, then (b, a) lies on f⁻¹. Plot a few points, swap their coordinates, and the shape of the inverse emerges. For f(x) = x³, the points (2, 8) and (−1, −1) map to (8, 2) and (−1, −1), and the resulting curve is the familiar cube root.

The reflection also explains the domain and range exchange visually: whatever the original covers horizontally, the inverse covers vertically. A function with a horizontal asymptote at y = 2 reflects into an inverse with a vertical asymptote at x = 2, which is exactly what the rational example produced. Spotting that correspondence on a graph tells you the restriction before any algebra confirms it.

Graphing calculators and Desmos handle this directly. Enter the function, enter y = x, and enter your candidate inverse; if the third curve is the mirror image of the first across the second, the algebra is right. A single mismatched point is usually a sign error in a square root or a forgotten domain restriction, and the visual check catches it faster than substitution.

Use the graph as the primary tool when the function is given pictorially or as a table of values. If a table lists (1, 3), (2, 5), (4, 9), the inverse table is (3, 1), (5, 2), (9, 4) — provided the original outputs are all distinct, which they are here. If two rows share an output, the table describes a function with no inverse, and the fix is to restrict the input set until the outputs separate.

Verifying With Composition Instead of Trusting the Algebra

Verification is non-negotiable because the swap-and-solve procedure can produce a plausible-looking wrong answer. The test is composition in both directions: f(f⁻¹(x)) must equal x, and f⁻¹(f(x)) must equal x, for every x in the relevant domains. One direction alone isn't enough. Take f(x) = x² with the restriction x ≥ 0 and the candidate f⁻¹(x) = √x. Then f(f⁻¹(x)) = (√x)² = x for x ≥ 0, which passes, but the reverse check f⁻¹(f(x)) = √(x²) = |x|, not x, for negative inputs — and that failure tells you the domains don't line up the way you assumed.

Work the composition with actual numbers as a second layer. For f(x) = 3x − 2 and f⁻¹(x) = (x + 2)/3, pick x = 4: f(4) = 10, and f⁻¹(10) = 4. Pick x = 0: f(0) = −2, and f⁻¹(−2) = 0. Two clean round trips plus the symbolic composition is enough evidence for a homework answer, though a proof requires the general algebra.

Extraneous solutions are the failure case worth naming. Squaring both sides during the solve step can admit values that don't satisfy the original equation, which is the same hazard that appears when solving radical equations. If your candidate inverse passes composition for positive inputs but fails for negative ones, the domain restriction is doing real work and needs to be written explicitly, not assumed.

A fast structural check supplements composition: the graphs of f and f⁻¹ must intersect on the line y = x wherever they cross at all. If your inverse's graph crosses the original somewhere off that line, one of the two is wrong. This catches sign errors that survive a single test point, and it's the check most students skip.

Domain Restrictions That Force an Inverse Into Existence

When a function fails the horizontal line test, you can still invert it by cutting the domain down to a piece where it passes. The standard move for f(x) = x² is to keep x ≥ 0, which yields f⁻¹(x) = √x, or keep x ≤ 0, which yields f⁻¹(x) = −√x. Both are legitimate; they're inverses of different restrictions of the same parabola. Which one you want depends on the context, and the answer must state it.

Trigonometry is where this becomes unavoidable. Sine, cosine, and tangent all fail globally, so their inverses are defined on chosen intervals: arcsin takes values in [−π/2, π/2], arccos in [0, π], and arctan in (−π/2, π/2). Those intervals aren't arbitrary — they're the largest stretches on which each function is one-to-one and covers its full range. Forget them and you get wrong answers: sin(5π/6) = 1/2, but arcsin(1/2) = π/6, not 5π/6, because π/6 is the value inside the defined interval.

Quadratics in general need completing the square before the restriction is visible. For f(x) = x² − 6x + 5, completing the square gives (x − 3)² − 4, so the vertex sits at x = 3. Restricting to x ≥ 3 makes the function increasing and invertible, which is why the worked example earlier kept the plus branch. Restricting to x ≤ 3 would have kept the minus branch instead. The vertex is the dividing line, always.

For periodic functions beyond trigonometry, the same principle applies with a chosen period. A model of a tidal cycle or a seasonal temperature curve is invertible only on one monotonic stretch, and any real application must name which stretch. That's not a mathematical technicality — it's the difference between an answer that means something and one that doesn't.

Mistakes That Produce Confidently Wrong Answers

The most frequent error is forgetting to swap before solving. Students solve y = 2x + 1 for x, getting x = (y − 1)/2, and then stop, leaving the answer in terms of y. The inverse must be expressed as a function of x, so the final relabel is mandatory, and a grader will mark the unrelabeled version wrong even though the algebra is sound.

The second is dropping the ± and picking a branch by habit rather than by domain. If the original domain was x ≥ 3, the inverse's range is y ≥ 3, and only the plus branch satisfies that. Choosing the minus branch produces a function that fails composition, but it looks tidy enough that the error survives casual review. Always test one point from the restricted domain.

A third mistake is confusing f⁻¹(x) with 1/f(x). The notation invites it, but the reciprocal of a function is almost never its inverse. For f(x) = 2x, the inverse is x/2 while the reciprocal is 1/(2x), and they agree only at x = 1. The superscript −1 signals function inversion, nothing else.

The fourth is misreading the domain-and-range exchange. Writing f⁻¹(x) = x² + 4 without the x ≥ 0 condition, as noted earlier, creates a function that accepts inputs the original radical could never output. Every restricted-domain original demands a matching restriction on the inverse, and omitting it is the error that graphing software will expose immediately.

A fifth, subtler mistake: assuming every invertible function has an elementary inverse. f(x) = x + e^x is invertible on all reals but its inverse can't be written with standard functions — it requires the Lambert W function, which most courses never introduce. Recognizing when to stop and say "no closed form" is part of knowing the method.

Inverse Functions in Real Models and Calculations

Inverses are how you run models backward, and that's their main practical use. If a conversion formula turns Celsius into Fahrenheit as F = (9/5)C + 32, the inverse C = (5/9)(F − 32) turns Fahrenheit back into Celsius, and the swap-and-solve procedure derives it in two lines. Any unit conversion, currency conversion at a fixed rate, or scaling relationship is a linear function whose inverse is the reverse conversion, and the negative-reciprocal slope rule tells you the inverse rate before you compute it.

Compound growth is the exponential case. An investment growing at 5 percent annually follows A = P(1.05)^t, and asking "how long until the balance doubles" means solving 2P = P(1.05)^t, which gives t = ln 2 / ln 1.05 ≈ 14.2 years. That's the inverse function at work: the logarithm undoes the exponential, and the calculation is the rule of 72 in exact form. The approximation says 72/5 = 14.4 years, close enough for mental math, and the exact figure comes from the inverse.

Geometry supplies a clean quadratic example. The area of a circle is A = πr², and the inverse r = √(A/π) recovers the radius from the area. The domain restriction is automatic here since radius is nonnegative, which is why this inverse never causes trouble — the physical context supplies the branch choice that pure algebra would leave open.

Cryptography and computing lean on inverses constantly, particularly modular inverses in RSA encryption, where the decryption exponent is the modular inverse of the encryption exponent. The arithmetic is different from real-number inversion, but the logic is identical: find the operation that undoes the original. That continuity across domains is why the four-step method is worth learning properly rather than memorizing per problem type.

FAQ

Does every function have an inverse?

No — only one-to-one functions do, and the horizontal line test settles it in seconds. A function that maps two inputs to the same output has no inverse unless you restrict its domain to a monotonic piece. Strictly increasing and strictly decreasing functions always qualify, which covers all linear functions with nonzero slope and all exponentials.

What's the difference between f⁻¹(x) and 1/f(x)?

They're unrelated. f⁻¹(x) is the function that undoes f, so f(f⁻¹(x)) = x, while 1/f(x) is the reciprocal, so f(x) · 1/f(x) = 1. For f(x) = 2x, the inverse is x/2 and the reciprocal is 1/(2x), and the two agree only at x = 1. The superscript −1 never means division when applied to a function name.

How do I find the inverse of a quadratic?

Restrict the domain to one side of the vertex first, then swap and solve using the quadratic formula. For f(x) = x² − 6x + 5 with x ≥ 3, the inverse is 3 + √(4 + x), keeping the plus branch because the original domain forces a range of y ≥ 3. Without the restriction, the swap produces ±√ and no function.

Can I find an inverse without algebra?

Yes, when the function is given as a graph or table. Reflect the graph across y = x, or swap the coordinates of each table row. A table of (1, 3), (2, 5), (4, 9) inverts to (3, 1), (5, 2), (9, 4), provided the original outputs are distinct — duplicate outputs mean no inverse exists on that input set.

Why does arcsin(1/2) equal π/6 and not 5π/6?

Because arcsin is defined to return values only in [−π/2, π/2], and π/6 falls inside that interval while 5π/6 doesn't. Sine itself equals 1/2 at both angles, which is exactly why the unrestricted sine has no inverse. The interval restriction is what makes the inverse a function rather than a two-valued relation.

What happens if the inverse can't be written with standard functions?

Some invertible functions have no elementary inverse — f(x) = x + e^x is the standard example, invertible everywhere but expressible only through the Lambert W function. The function still has an inverse; it just can't be written with the operations most courses cover. Recognizing that limit prevents hours of futile algebra.

Do the domains of f and f⁻¹ have to match?

No — they exchange. The domain of f becomes the range of f⁻¹, and the range of f becomes the domain of f⁻¹. For f(x) = √(x − 4) with domain x ≥ 4, the inverse x² + 4 carries the domain x ≥ 0, matching the original's range. Mismatched domains are the most common source of composition failures.

Is the inverse of a linear function always linear?

Yes, as long as the slope isn't zero. For f(x) = mx + b with m ≠ 0, the inverse is (x − b)/m, another linear function whose slope is 1/m — the reciprocal, not the negative reciprocal, since the perpendicular relationship applies to graphs crossing at right angles rather than to inverse pairs. A horizontal line has no inverse because it fails the horizontal line test.