An equation fails to represent a linear function when its graph is not a straight line. The four failures are: any exponent other than 1 (y = x²), a variable in a denominator (y = 1/x), a variable under a radical or inside absolute-value bars, and any product of variables (xy = 6). Equations in slope-intercept form, y = mx + b, stay linear because every x term has power 1 and every y appears alone to power 1. Non-linearity is a property of structure, not appearance. Read every equation through that four-part filter and the answer resolves in seconds.

The structural test for linearity

The definition does most of the work. A linear function in two variables is any equation that can be rewritten as y = mx + b, where m and b are constants. The "can be rewritten" clause matters, because a professional caveat hides inside it. Rearrangement is legal. Multiplying both sides by a constant, or moving terms across the equals sign, never changes a graph's shape. So an equation that looks messy can still be linear, and an equation that looks tidy can still fail.

Work through the standard candidate. Start with x + y = 4. Subtract x from both sides and y = −x + 4 emerges. Slope −1, intercept 4, straight line. Now test the classic wrong answer: y = x² + 2. Try to rearrange it into y = mx + b. The squared term survives every legal operation. No constant m can pair with a variable exponent, so the equation is not linear, cannot be made linear, and its graph is a parabola. The same logic dispatches y = 1/x: clearing the denominator gives xy = 1, a product of variables that no rearrangement removes.

Here is the counterintuitive detail most short explainers miss: coefficients do not affect linearity. y = 0.5x − 7, y = 1000x, and y = −3x + 0.0001 are all linear because the coefficient is constant in every case. A craggy graph for y = 1000x makes students reach for "non-linear," but steepness and linearity are different properties. Slope measures steepness; a constant slope measures linearity. Only when the slope itself changes, as it does at every point of a curve, does the function leave the linear class.

A second trap is that the word linear appears on two floors of the math building. At the algebra level, f(x) = mx + b counts as a linear function even though its graph often misses the origin. At the linear-algebra level, linear transformations must map zero to zero, which excludes b ≠ 0. This explainer uses the algebra-class definition because that is the one exam questions and graphing calculators use. One plain sentence: for school purposes, straight line equals linear, endpoint to endpoint.

Quantified context: what non-linearity does to outputs

Numbers make the distinction vivid. Compare y = 3x + 1 with y = x² + 1 at the same inputs. Table both.

x y = 3x + 1 Change per step y = x² + 1 Change per step
0 1 — 1 —
1 4 +3 2 +1
2 7 +3 5 +3
3 10 +3 10 +5
4 13 +3 17 +7

The linear function's output grows by exactly 3 for every unit of x, forever. The quadratic function's output grows by 1, then 3, then 5, then 7. The differences themselves climb by 2 each time, which is the fingerprint of a quadratic. That constant-difference property is the arithmetic shadow of linearity, and it gives you a calculator-free test: subtract consecutive outputs, and if the differences refuse to settle, the function is not linear.

Rates of change matter outside the classroom. Over the interval [3, 4], the quadratic's average rate is 7, but over [0, 1] it is 1. That ratio of 7 to 1 shows why curve-fitting a line to quadratic data produces errors that balloon with distance from the fitted points. A linear model's error stays bounded; a mismatched quadratic model's error grows with the square of the gap.

Absolute value earns its own paragraph because it is the most commonly misclassified entry. The graph of y = |x| makes a V: two line segments meeting at the origin. Neither piece fails the slope test, yet the function as a whole has no single slope. At x = 0 the slope jumps from −1 to +1. The definition requires one m for the entire domain, so the V is not linear, no matter how straight each arm looks.

Where the slope-intercept view breaks

The deeper mechanism is that linearity is about differences, not drawing styles. For y = mx + b, pick any two inputs x₁ and x₂. The output change is m(x₂ − x₁), which depends only on the gap between the inputs, never on where they sit. Professional mathematicians call this translation invariance, and it is the engine of every linear system in engineering and statistics.

Exponential equations are the quiet failure. y = 2ˣ looks tamable on a small table: at x = 0, 1, 2, 3 it gives 1, 2, 4, 8. Students see tidy integers and assume straightness. Plot twenty points and the curve takes off; by x = 10 the output is 1024, while the straight line y = 2x would sit at 20. Doubling every step is the signature of exponential growth, and growth by a constant multiplier is the one pattern no constant slope can imitate.

Also watch compound equations. y = x(x + 2) is nonlinear despite containing no explicit exponent, because expanding it gives y = x² + 2x. Products of variables, factored or not, are quadratic or worse. The test is not "does a squared symbol appear" but "does any legal rearrangement leave y with power 1 and alone on one side." Converses trap students too: 3x + 4y − 8 = 0 is linear, since solving for y gives a valid mx + b form. Being written in general form means nothing by itself.

The non-linear family: comparison table

Once an equation fails the test, identify which family it joined. The families are named, comparable, and each fails a different clause of the definition.

Equation Graph shape Why it fails Linear?
y = x² + 2 Parabola x exponent is 2 No
y = 1/x Hyperbola, two branches Variable in denominator No
y = x V shape
y = 2ˣ Exponential curve Growth by constant multiplier No
xy = 6 Hyperbola Product of variables No
y = 0.5x − 7 Line None; slope is constant 0.5 Yes
3x + 4y = 12 Line Rewrites as y = −0.75x + 3 Yes

Every "No" row violates exactly one clause, which is why the four-part filter works. Vandermonde and other professional evaluators use a formal version of this table: they compute second differences, and any nonzero entry disqualifies the candidate.

Misconceptions and red flags

Three bad habits produce most wrong answers, and each has an antidote.

"Ugly means non-linear." Coefficients rarely matter; only the variable's power and position do. Verify by rearranging first. "Some negative answers are possible." In an exam asking for the equation that does NOT represent a linear function, exactly one option violates the definition. Most choices are distractors that look complex but simplify to lines. "Any equation can be linearized into y = mx + b." Rearranging is not transformation. Taking the square root of both sides of y² = x² + 1 substitutes a radical for the square and changes the very object being graphed. Red flag: treating y = |x| as two lines glued together and therefore linear. A discontinuous slope is a fatal disqualifier. Red flag: accepting y = x + x² because a linear term appears alongside the squared one. One offending term poisons the whole equation. Red flag: judging from the first three integer test points. Cubics and exponentials masquerade as lines briefly; sample at least five points spanning negative and positive inputs.

Practical test procedure and thresholds

For any equation, run a two-minute checklist. First, isolate y. Second, confirm every x sits at power 1 and every constant term is plain. Third, scan for divisions, radical signs, absolute-value bars, and variable products. Fourth, plug in three points spread wide, say x = −2, 0, and 2, and confirm the slope between pairs is identical. Fifth, table consecutive differences; if they are not constant, stop. These five steps settle every exam, worksheet, and calculator entry.

When choosing a model for real data, thresholds decide. A set of points fits a linear function only if the average rate of change across the data range stays within the measurement's noise band from beginning to end. For clean data with negligible noise, treat any systematic change in average rate as evidence of curvature. Practicing adults can benchmark their accuracy skills against a 16-question linear-equation diagnostic on the SUMS (Skills Upgrading for the Manufacturing Sector) evidence base, part of the U.S. Department of Education's assessment program. Accuracy below 13 of 16 signals a checkpoint worth revisiting before moving to graphing calculators, since the same model also flags graduation odds: adults without major literacy skill gains were about 2.4 times as likely to drop out of high school.

One implementation detail for teachers: the slope consistency check needs no graph paper. Compute m = (y₂ − y₁)/(x₂ − x₁) for two disjoint pairs of points. If the values match, the equation is linear. If they differ, the answer is settled, and the differing slope values themselves reveal the family. A slope that climbs from 1 to 3 to 5 across equal steps marks a quadratic; a ratio between slopes rather than a difference marks the exponential.

Limits of the test

The structural filter classifies two-variable equations perfectly, but it has boundaries. Piecewise functions and other members of the broader nonlinear class are not the target of exam questions, and calling them "not linear" is technically true but functionally unhelpful. Domain matters: it is true that every linear function has slope m, but that linear function's slope is only constant over its domain. And it is true that non-linear variables stay non-linear, but that non-linear structure has its own vocabulary, separate from the linear one.

Some formats add cost. Equations using absolute-value notation, set notation, or functions mixed with constants require a re-check before the filter applies. In particular, an absolute value equation is not a linear function in the school sense, but its component lines do satisfy y = mx + b, which is why Venn diagrams are a more honest graphic than a list. The practical takeaway is unchanged: when the question asks which equation does not represent a linear function, find the clause violation and stop.