How this calculation works
A partial derivative changes one input while keeping the others constant. The gradient collects these slopes into an input-space vector.
∇f = (∂f/∂x, ∂f/∂y, ∂f/∂z)A worked example
For f=x²+y²+z² at (1,1,0), the gradient is (2,2,0), with magnitude sqrt(8).
Good to know
Bounded real expressions and floating-point arithmetic. Finite path and surface samples do not prove arbitrary limits or differentiability.
Hold the other variables constant
When differentiating x²y+z with respect to x, y acts as a constant multiplier and z differentiates to zero. The result is 2xy. Differentiating the same expression with respect to y gives x², while its z derivative is one.
This distinction matters when several variables represent physical inputs. A partial derivative describes a controlled change in one input, not a change along an arbitrary path where all inputs may vary together. The table shows both symbolic expressions and their evaluated values.
Read the gradient as an input-space vector
For a differentiable scalar function, the dot product of its gradient with a unit direction gives the rate of change per unit distance in that direction. The maximum directional rate equals the gradient magnitude, attained in the gradient direction when the gradient is nonzero.
A zero gradient does not classify a stationary point. A bowl, an inverted bowl and a saddle can all have zero gradient at the center. Further local analysis is needed before calling the point a minimum or maximum.
Understand the fixed-z surface
The picture holds the entered z input constant and uses graph height for f(x,y,z). For f=x²+y²+z², increasing the fixed z input shifts the pictured bowl upward. This makes the slice consistent with the selected three-input point.
The arrow on this slice uses only the x,y gradient components. The full three-component gradient belongs to three-dimensional input space, while graphing a scalar function of three inputs would require a fourth coordinate. A slice cannot show all that information at once.
Check smoothness before applying the gradient rule
Formal partial derivatives do not by themselves prove differentiability. A function can have partial derivatives along coordinate directions and still behave badly on other approach paths. Nonsmooth roots and absolute values need particular care.
Use the multivariable limit path explorer to inspect different paths near a suspicious point. The directional derivative calculator with interactive surface handles a selected two-input direction. These related views help separate local geometry, algebraic derivatives and the assumptions required by the gradient theorem.
A few questions, answered
Can I enter three variables?
Yes. Use x, y and z with explicit multiplication.
What does the surface height mean?
It is f(x,y,z0), with z0 fixed to your entered z value.
Does zero gradient mean minimum?
No. It is a critical-point condition, not a classification.