The idea: differentiate the equation, not a formula for y

The circle x² + y² = 25 has a tangent at (3, 4), but it is not the graph of a function: the vertical line x = 3 meets it at y = 4 and y = −4. You could split it into two halves, y = √(25 − x²) and y = −√(25 − x²), and differentiate one of them. Implicit differentiation skips that: treat y as some unknown function y(x), differentiate both sides of the equation with respect to x, and solve for y′ = dy/dx. For the circle the answer is y′ = −x / y, so the slope at (3, 4) is −3/4. (Demo: circle at (3, 4))

The circle x squared plus y squared equals 25. A vertical line at x = 3 meets it at (3, 4) and (3, −4). The upper half is drawn in purple, and a green tangent touches the circle at (3, 4) with slope −3/4.
x = 3 meets the circle twice, so y is not a function of x; near (3, 4) the upper half is, and its tangent has slope −x / y = −3/4.

Differentiate both sides

If x² + y(x)² = 25 holds for every x on the curve, the two sides are equal as functions of x, so their derivatives are equal too. Each term is differentiated with the usual rules. A term with only x is differentiated as usual. A term with y needs the chain rule, because y depends on x: the outside is differentiated and multiplied by the inside's derivative y′. A product of x and y needs the product rule.

termd/dxrule
x²2xpower rule
y²2y · y′chain rule
y³3y² · y′chain rule
xy1 · y + x · y′product rule
sin ycos y · y′chain rule
250constant

On the canvas each term is a box; orange boxes contain y. Differentiate drops each term's derivative into the row below, with the rule it used.

Collect, factor, solve

After differentiating, y′ appears only to the first power, so the equation is linear in y′ and can always be solved: move every term with y′ to the left, factor y′ out, divide.

  • Circle: 2x + 2y·y′ = 0, so 2y·y′ = −2x and y′ = −x / y.
  • Tilted ellipse x² + xy + y² = 7: 2x + (y + x·y′) + 2y·y′ = 0, so y′(x + 2y) = −(2x + y) and y′ = −(2x + y) / (x + 2y); at (1, 2) that is −4/5. (Demo: x² + xy + y² = 7)
  • Folium of Descartes x³ + y³ = 6xy: 3x² + 3y²·y′ = 6y + 6x·y′, so y′ = (2y − x²) / (y² − 2x); at (3, 3) that is −1. (Demo: folium at (3, 3))

The answer is a formula in x and y. That is not a defect: one x can have several points, with different slopes, so the slope must depend on which point you mean. Walk the Curve moves P round the circle and the tangent follows −x / y. (Demo: walk the circle)

The shortcut y′ = −Fx / Fy

Write the curve as F(x, y) = c. Differentiating gives Fx + Fy · y′ = 0, where Fx and Fy are the partial derivatives (differentiate in x with y held fixed, and the other way round). So y′ = −Fx / Fy. For the circle Fx = 2x and Fy = 2y, the same −x / y. The tangent is perpendicular to the gradient (Fx, Fy), which is why level curves cross the gradient at right angles.

The formula breaks where Fy = 0:

  • Fy = 0 and Fx ≠ 0: a vertical tangent. The circle at (5, 0) gives y′ = −5 / 0. (Demo: vertical tangent at (5, 0))
  • Fx = Fy = 0: y′ = 0/0 says nothing. The folium crosses itself at the origin, and its two branches have different tangents, y = 0 and x = 0. (Demo: folium crossing at (0, 0))
The folium of Descartes x cubed plus y cubed equals 6xy, a loop in the first quadrant with two tails. A green tangent with slope −1 touches it at (3, 3). An orange vertical tangent touches it near (3.17, 2.52) where Fy = 0. At the origin the curve crosses itself and has two tangents, the x axis and the y axis.
On the folium, y′ = −Fx / Fy gives −1 at (3, 3), a vertical tangent where Fy = 0, and 0/0 at the crossing.

Why it works: locally a function

The implicit function theorem says: if Fy ≠ 0 at a point of F(x, y) = c, then near that point the curve is the graph of one differentiable function y(x). The vertical line test colours that piece purple: the arc through P on which x keeps moving one way, from one vertical tangent to the next. Implicit differentiation computes the slope of that function without ever writing it down, which matters because often you can't: the tilted ellipse needs the quadratic formula, the folium a cubic, and an equation like y⁵ + y = x has no formula for y at all.

Where it is used

  • Related rates: differentiate an equation like x² + y² = L² with respect to time t, and dx/dt and dy/dt appear the same way y′ does here.
  • Derivatives of inverse functions: from y = ln x write eʸ = x, then eʸ·y′ = 1, so y′ = 1/x. The same trick gives arcsin′ x = 1/√(1 − x²).
  • Tangent lines to conics and to level curves of a function of two variables, and the Lagrange multiplier condition, which compares gradients.

The chain rule behind every y′ factor is animated on The Chain Rule, and the power and product rules on Derivative Rules as Pictures.

What the page leaves out

Only three curves, each with three fixed points. The second derivative y″ (differentiate y′ = −x / y again), implicit equations with three variables (surfaces and partial derivatives ∂z/∂x), and the proof of the implicit function theorem are not shown. The slopes come from the derived formulas, not from symbolic algebra.

On the mind map

Open the Calculus mind mapCalculusLimits and derivatives: slope at a point, as a limitLimits and derivativesReading derivatives: what f′ and f″ say about fReading derivativesIntegrals: add up thin slicesIntegralsSeries: a function as an infinite sumSeriesNumerical methods: approximate when exact is hardNumerical methodsDifferential equations: know the slope, find the curveDifferential equationsChaos: simple rules, unpredictable pathsChaosVector calculus: fields that spread and spinVector calculusLimits and continuityLimits and continuityfor every ε band there is a δ windowDerivative as a limitDerivative as a limitsecant slope as h → 0 gives f′(x)Local linearityLocal linearityzoom in: a smooth curve becomes a lineDerivative rulesDerivative rulesx² grows by 2x·dx; the dx² corner vanishesChain ruleChain rulerates multiply: dy/dx = dy/du · du/dxWeierstrass functionWeierstrass functioncontinuous everywhere, smooth nowhereImplicit differentiationtreat y as y(x), then solve for y′L'Hôpital's ruleL'Hôpital's rule0/0? compare the slopes insteadOpen the Calculus mind mapCalculusslope at a point, as a limitLimits and derivativesLimits and continuityLimits and continuityfor every ε band there is a δ windowDerivative as a limitDerivative as a limitsecant slope as h → 0 gives f′(x)Local linearityLocal linearityzoom in: a smooth curve becomes a lineDerivative rulesDerivative rulesx² grows by 2x·dx; the dx² corner vanishesChain ruleChain rulerates multiply: dy/dx = dy/du · du/dxWeierstrass functionWeierstrass functioncontinuous everywhere, smooth nowhereImplicit differentiationtreat y as y(x), then solve for y′L'Hôpital's ruleL'Hôpital's rule0/0? compare the slopes instead
This page on the Calculus mind map · 42 pages in 8 branches · open it to see the whole map and test yourself in recall mode.