Implicit Differentiation

Roll a ball around the inside of a circular bowl and, at every instant, it is heading in some direction — it has a slope. Yet the circle it traces, x^2 + y^2 = 25, is not a function: above most values of x there sit two heights, one on the top arc and one on the bottom. You cannot write it as a single tidy y = f(x). So how do you differentiate a curve that refuses to be solved for y?

The answer is a beautifully sneaky move called implicit differentiation: you differentiate the equation just as it stands, treating y as a hidden function of x, and let the chain rule do the untangling. It unlocks the slope of circles, ellipses, and every knotted curve that a plain y=f(x) could never describe.

So far every curve has come solved for y: y = x^2, y = (3x+1)^5. But some curves tangle y up with x and refuse to be untangled. The circle of radius 5 is the classic example:

x^2 + y^2 = 25

We can still find the slope \frac{dy}{dx} — without ever solving for y. The trick is to differentiate both sides with respect to x, treating y as a (hidden) function of x. Then every y-term picks up an extra \frac{dy}{dx} by the chain rule:

\frac{d}{dx}\bigl[y^2\bigr] = 2y\cdot\frac{dy}{dx}

That is the chain rule with an outer function u^2 and an inner function y: bring the power down (2y), then multiply by the derivative of the inside (\frac{dy}{dx}). Differentiating the whole equation term by term:

2x + 2y\,\frac{dy}{dx} = 0

Now just solve for \frac{dy}{dx} — subtract 2x and divide by 2y:

\frac{dy}{dx} = -\frac{x}{y}

One formula gives the slope at every point of the circle. At (3,4) the slope is -\tfrac{3}{4}; at (4,3) it is -\tfrac{4}{3}. No solving for y required.

The recipe

To differentiate an equation that mixes x and y together: This is exactly how you find the slope of a curve — a circle, an ellipse, any relation — that isn't a function and so can never be written as y = f(x).

A mixed term: the product rule strikes

Not every equation splits into neat separate x- and y-terms. Consider

x^2 + xy + y^2 = 7.

The middle term xy is a product of two things that both depend on x, so it needs the product rule. Watch it differentiate term by term:

\underbrace{2x}_{x^2} \;+\; \underbrace{\Bigl(1\cdot y + x\cdot\tfrac{dy}{dx}\Bigr)}_{xy\ \text{(product rule)}} \;+\; \underbrace{2y\,\tfrac{dy}{dx}}_{y^2} \;=\; 0.

Gather the \frac{dy}{dx} terms on one side and everything else on the other:

x\,\frac{dy}{dx} + 2y\,\frac{dy}{dx} = -2x - y.

Factor out \frac{dy}{dx} and divide:

\frac{dy}{dx} = \frac{-2x - y}{x + 2y}.

Notice the pattern that appears again and again: differentiate, collect the \frac{dy}{dx} terms, factor, divide. That closing manoeuvre is the signature of every implicit-differentiation problem.

Worked example: a tangent line

Find the equation of the tangent to x^2 + y^2 = 25 at the point (-3, 4).

Step 1 — the slope. We already have \frac{dy}{dx} = -\frac{x}{y}. Substitute the point:

\left.\frac{dy}{dx}\right|_{(-3,4)} = -\frac{-3}{4} = \frac{3}{4}.

Step 2 — the line. Using point–slope form y - y_1 = m(x - x_1):

y - 4 = \tfrac{3}{4}\,(x + 3).

A quick sanity check: the radius to (-3,4) has slope \tfrac{4}{-3} = -\tfrac{4}{3}, and the tangent's slope \tfrac{3}{4} is its negative reciprocal — so the tangent is perpendicular to the radius, exactly as a circle demands. The geometry confirms the calculus.

The slope of a circle, pictured

Here is x^2 + y^2 = 25 with the tangent line at (3,4). Implicit differentiation predicts a slope of -\frac{x}{y} = -\frac{3}{4} there — and the tangent line bears it out. Notice the tangent is perpendicular to the radius, just as it should be.

Three slips account for nearly every lost mark on implicit differentiation.

Implicit differentiation is the secret engine behind related rates — problems where several changing quantities are chained together by one equation. A ladder leans against a wall; its foot slides out at a known speed. How fast does its top slide down? Pythagoras ties the two together, x^2 + y^2 = L^2, and differentiating implicitly with respect to time gives 2x\frac{dx}{dt} + 2y\frac{dy}{dt} = 0 — instantly relating the two speeds. The same trick tells you how fast a balloon's radius grows as you pump air in, or how the shadow of a walking person lengthens.

It reaches further still. Differentiating through webs of interdependent variables is exactly what backpropagation does inside a neural network — chaining derivatives through thousands of tangled quantities to work out which knob to turn. When you meet related rates, you'll recognise this very move dressed in new clothes.

See it explained