Step 1: Pick the variable to differentiate
To pick the variable, read it off the notation: ∂f/∂y and fy both mean "with respect to y". y is the only variable that moves.

This partial derivative calculator finds first, second, mixed and higher-order partial derivatives of f(x, y, z) with steps. It holds the other variables constant, checks Clairaut's theorem, builds the Hessian and evaluates everything at a point.
Results
Your derivative, as a slope and step by step
The orange curve is the slice y = b and the teal curve is the slice x = a. Their dashed tangent lines have slopes fx and fy at the point.
Interactive diagram
Surface and slice slopes
Partial derivative
All first partials
Second partials (Hessian)
Step-by-step solution
—Step by step
To use the partial derivative calculator, follow these 5 steps. The preview, the result and the 3D view update as you type.
Type the function
Use x, y and z, for example x^2*y^3 + 3*x*y or exp(-(x^2 + y^2)).
Build the order
Press + ∂x, + ∂y or + ∂z in the order you differentiate. ∂x then ∂y gives fxy.
Add a point (optional)
Enter values such as 1, 2 to evaluate the derivative and get the tangent plane.
Read the result and the tables
Check the gradient, the Hessian and the Clairaut check that fxy = fyx.
Follow the steps and the 3D view
Each step differentiates term by term while the other variables stay constant.
Definition
A partial derivative is the rate of change of a function of several variables with respect to one variable, while every other variable is held constant. For f(x, y), the partial derivative with respect to x measures how fast f changes when x moves and y stays fixed.
The definition is the ordinary derivative applied to one slice of the function. Fix y = b, and f(x, b) becomes a function of x alone; its derivative at x = a is fx(a, b). For f = x²y³ + 3xy at (1, 2), the slice is 8x² + 6x and the slope at x = 1 is 22.
Partial derivatives are the basic tool of multivariable calculus. They give the gradient vector, tangent planes, optimization of functions of several variables, and the partial differential equations of heat, waves and fluids. Economists use them for marginal cost and marginal utility; machine learning uses them for every gradient step.
Red dashed: the secant through x = 1 and x = 1 + h. Teal: the tangent with slope fx = 22. As h shrinks, the 2 lines merge.
Symbols
There are 3 common notations for partial derivatives: Leibniz notation ∂f/∂x, subscript notation fₓ, and operator notation Dₓf. All 3 mean the same thing; the curly ∂ (read "partial" or "del") marks a partial derivative and separates it from the ordinary d.
| Meaning | Leibniz | Subscript | Operator |
|---|---|---|---|
| First partial in x | |||
| First partial in y | |||
| Second partial in x | |||
| Mixed: x first, then y | |||
| Value at a point |
Method
To find a partial derivative, pick the variable, freeze every other variable as a constant, and differentiate with the ordinary rules. The 3 steps below find ∂f/∂y for f = x²y³ + 3xy.
To pick the variable, read it off the notation: ∂f/∂y and fy both mean "with respect to y". y is the only variable that moves.
To treat x as a constant, read it like a number: x² is a coefficient and 3x is a coefficient. f becomes (x²)·y³ + (3x)·y, a function of y alone.
To finish, apply the power rule term by term in y:
At (1, 2): fy = 12 + 3 = 15.
Rules
The rules of differentiation carry over to partial derivatives unchanged: the power rule, the constant multiple rule, the sum rule, the product rule, the quotient rule and the chain rule. The only new idea is which factors count as constants.
| Rule | Formula (in x) | Example |
|---|---|---|
| Power rule | ||
| Constant multiple | ||
| Sum rule | ||
| Constant term |
The product and quotient rules apply when both factors contain the variable being differentiated. When only one factor does, the constant multiple rule is enough.
The chain rule multiplies the outer derivative by the partial derivative of the inside: ∂/∂x f(g(x, y)) = f′(g) · ∂g/∂x. For sin(x²y), the inside x²y has ∂/∂x = 2xy:
The multivariable chain rule handles paths: for z = f(x(t), y(t)), dz/dt = fx x′(t) + fy y′(t). Implicit differentiation follows from it: on a curve F(x, y) = 0, dy/dx = −Fx/Fy, so x² + y² = 25 has slope −3/4 at (3, 4).
Higher order
A function of 2 variables has 4 second-order partial derivatives: 2 pure ones, fxx and fyy, and 2 mixed partials, fxy and fyx. Each one differentiates a first-order partial again.
Clairaut's theorem says the mixed partials are equal, fxy = fyx, wherever both are continuous. The order of differentiation does not matter for smooth functions. Alexis Clairaut stated it in 1740, and Hermann Schwarz gave a rigorous proof in 1873, so it goes by Schwarz's theorem too.
For f = x³y² + exy, both orders give 6x²y + (1 + xy)exy, which is 6 + 2e ≈ 11.4366 at (1, 1). The calculator checks the equality for every function, first symbolically and then at sample points.
The Hessian matrix collects all second-order partial derivatives in a square table, H = [[fxx, fxy], [fyx, fyy]] for 2 variables. Clairaut's theorem makes it symmetric. Its determinant drives the second-derivative test: at a critical point, D = fxx fyy − fxy² > 0 with fxx > 0 means a minimum, D > 0 with fxx < 0 a maximum, and D < 0 a saddle.
— Top row: first partials. Lower 2 rows: the Hessian; the 2 shaded mixed partials always match for these smooth functions.
Geometry
Geometrically, fx(a, b) is the slope of the curve cut from the surface z = f(x, y) by the plane y = b, and fy(a, b) is the slope of the curve cut by the plane x = a. Each partial derivative measures steepness in one compass direction: east-west for x, north-south for y.
The 2 slopes together describe the surface near the point. A positive fx means the surface climbs as x increases; fx = fy = 0 means a flat spot, a candidate for a maximum, minimum or saddle point. Slopes in other directions come from the gradient vector ∇f = (fx, fy): the directional derivative in unit direction u is ∇f · u.
For a slope in any direction, the steepest direction and a contour map, use the Gradient Calculator.
Orange: the slice y = b and its tangent. Teal: the slice x = a and its tangent. Drag to turn the surface.
Linear approximation
The tangent plane to z = f(x, y) at (a, b) is z = f(a, b) + fx(a, b)(x − a) + fy(a, b)(y − b). It contains both tangent lines of the slices, and it gives the best linear approximation of f near the point.
For f = x² + y² at (1, 2): f = 5, fx = 2 and fy = 4, so the tangent plane is z = 5 + 2(x − 1) + 4(y − 2). The approximation error shrinks with the square of the distance: 0.02 at a step of 0.1, and 0.5 at a step of 0.5. The calculator prints the tangent plane whenever a point is entered.
| Point | f(x, y) | Tangent plane | Error |
|---|---|---|---|
| (1.01, 1.98) | 4.9405 | 4.94 | 0.0005 |
| (1.1, 2.1) | 5.62 | 5.6 | 0.02 |
| (1.5, 2.5) | 8.5 | 8 | 0.5 |
More variables
A function of 3 variables, f(x, y, z), has 3 first-order partial derivatives, fx, fy and fz, each found by holding the other 2 variables constant. The method does not change; there is one more variable to freeze.
For f = xyz + x²z³, the first partials are fx = yz + 2xz³, fy = xz and fz = xy + 3x²z². At (1, 1, 1) they form the gradient vector (3, 1, 4). The 3 × 3 Hessian has 9 entries; Clairaut's theorem leaves 6 different ones.
The count grows fast. A function of n variables has nk partial derivatives of order k, but only C(n + k − 1, k) different ones when the mixed partials are equal. AnalyzeMath's partial derivative calculator handles 2 variables; this calculator handles 3 and goes up to fourth order.
| Order k | 2 variables: all / different | 3 variables: all / different |
|---|---|---|
| 1 | 2 / 2 | 3 / 3 |
| 2 | 4 / 3 | 9 / 6 |
| 3 | 8 / 4 | 27 / 10 |
| 4 | 16 / 5 | 81 / 15 |
Worked examples
Here are 4 worked examples of partial derivatives: a first partial, a mixed partial at a point, a third-order partial in 3 variables and a quotient. Each graph shows the derivative as the slope of a curve.
Answer:
Answer: . The other order, fyx = ∂/∂x (3x²y² + 3x) = 6xy² + 3, gives the same 27.
f(x, y, z) = xyz + x²z³
Answer:
f(x, y) = x/(x + y), with respect to y.
Answer: . f falls as y grows, since y sits only in the denominator.
Comparison
An ordinary derivative applies to a function of 1 variable, a partial derivative changes 1 variable of several while freezing the rest, and a total derivative lets every variable change at once along a path.
| Ordinary | Partial | Total | |
|---|---|---|---|
| Function | f(x) | f(x, y, …) | f(x(t), y(t)) |
| Symbol | |||
| Other variables | none | held constant | change with t |
| Example | x = cos t, y = sin t: |
Handheld calculators cover the ordinary case only: the TI-84 Plus nDeriv( gives a numeric derivative of one variable. Symbolic partial derivatives need a computer algebra system, such as Wolfram Alpha, or this partial derivative calculator.
Questions
Holding a variable constant means treating it as a fixed number while you differentiate with respect to another variable. For ∂/∂x of x²y³, y³ acts like a coefficient such as 8, so the result is 2x·y³. Geometrically, you stay on one slice of the surface, the plane y = constant.
No, mixed partial derivatives are equal when they are continuous (Clairaut's theorem), but not always. Every polynomial, exponential and trig combination satisfies f_xy = f_yx. The classic exception is f = xy(x² − y²)/(x² + y²) with f(0, 0) = 0: at the origin, f_xy = −1 and f_yx = 1, since the second partials are not continuous there.
The difference is that ∂ marks a partial derivative, taken with the other variables held constant, while d marks an ordinary or total derivative, where the function depends on one variable or every variable moves together. ∂f/∂x for f(x, y) freezes y; df/dt for f(x(t), y(t)) lets x and y both change.
Find a third-order partial derivative by differentiating 3 times in a row, one variable at a time, in the order the subscript gives. f_xzz of xyz + x²z³ takes x (yz + 2xz³), then z (y + 6xz²), then z again: 12xz, which is 12 at (1, 1, 1). For smooth functions, the order of the 3 steps does not change the result.
The Hessian is used to classify critical points, to build quadratic (second-order) approximations and to run Newton's method in optimization. At a critical point of f(x, y), det H > 0 with f_xx > 0 means a local minimum, det H > 0 with f_xx < 0 a local maximum, and det H < 0 a saddle point. Machine learning uses it to study curvature of loss surfaces.
This partial derivative calculator finds first, second, mixed and higher-order partial derivatives with steps, checks that mixed partials agree, builds the Hessian and gives the tangent plane at a point.