What Are Directional Derivatives and Gradients?
A directional derivative measures how fast a multivariable function changes at a point as you move in a chosen direction. In single-variable calculus, a derivative tells you the slope of a curve as \(x\) changes. In multivariable calculus, there are many possible directions to move, so the rate of change depends on the direction.
For a function such as \(f(x,y)\), the usual partial derivatives \(f_x\) and \(f_y\) describe change in the coordinate directions. The directional derivative extends that idea: instead of asking only “what happens if \(x\) changes?” or “what happens if \(y\) changes?”, it asks “what happens if we move from the point along this vector?”
The gradient is the vector that organizes the partial derivatives into one object. For a function of two variables, the gradient is
$$
\nabla f(x,y)=\left\langle f_x(x,y), f_y(x,y) \right\rangle.
$$
For a function of three variables, it is
$$
\nabla f(x,y,z)=\left\langle f_x(x,y,z), f_y(x,y,z), f_z(x,y,z) \right\rangle.
$$
The gradient points in the direction where the function increases most rapidly, and its magnitude gives that maximum rate of increase at the point. A directional derivative is the part of the gradient that points along the direction you choose.
Why Directional Derivatives Matter
Directional derivatives are useful whenever a quantity depends on more than one input. They help answer questions such as:
-
How quickly does height increase if you walk northeast on a terrain map?
-
How does temperature change if you move in a specific direction across a plate?
-
How sensitive is a model output to a combined change in several variables?
-
Which direction gives the fastest increase or fastest decrease of a function?
For students, directional derivatives connect partial derivatives, vectors, dot products, unit vectors, and contour plots into one idea. For applied work, they are a compact way to describe local change when several inputs can vary at the same time.
Key Terms to Know
-
Multivariable function: A function with more than one input, such as \(f(x,y)\) or \(f(x,y,z)\).
-
Point: The location where the function and its local rate of change are evaluated, such as \((1,2)\) or \((1,2,-1)\).
-
Direction vector: A nonzero vector that tells which way to move from the point.
-
Unit vector: A vector with length \(1\). Directional derivatives use a unit direction so the result is a rate per one unit of distance.
-
Gradient: The vector of partial derivatives of a function.
-
Dot product: A vector operation that measures how much two vectors point in the same direction.
-
Centered finite difference: A numerical method that estimates a derivative by evaluating the function on both sides of the point.
-
Contour plot: A graph of level curves, where each curve represents points with the same function value.
How Directional Derivatives Work
The direction vector must first be converted to a unit vector. If the entered direction vector is \(v\), then the unit direction is
$$
u=\frac{v}{\|v\|}.
$$
The directional derivative of \(f\) at a point \(a\) in the unit direction \(u\) is
$$
D_u f(a)=\nabla f(a)\cdot u.
$$
This formula says: take the gradient at the point, take the unit vector in the desired direction, and compute their dot product.
For a two-variable function, if
$$
\nabla f(a,b)=\left\langle f_x(a,b),f_y(a,b)\right\rangle
$$
and
$$
u=\left\langle u_1,u_2\right\rangle,
$$
then
$$
D_u f(a,b)=f_x(a,b)u_1+f_y(a,b)u_2.
$$
For a three-variable function, if
$$
\nabla f(a,b,c)=\left\langle f_x(a,b,c),f_y(a,b,c),f_z(a,b,c)\right\rangle
$$
and
$$
u=\left\langle u_1,u_2,u_3\right\rangle,
$$
then
$$
D_u f(a,b,c)=f_x(a,b,c)u_1+f_y(a,b,c)u_2+f_z(a,b,c)u_3.
$$
The relationship with the angle \(\theta\) between the gradient and the chosen direction is
$$
D_u f(a)=\|\nabla f(a)\|\cos\theta.
$$
This explains the main interpretations:
-
If the direction matches the gradient, then \(\theta=0\) and the directional derivative is as large as possible.
-
If the direction is opposite the gradient, then \(\theta=180^\circ\) and the function decreases as fast as possible.
-
If the direction is perpendicular to the gradient, then \(\theta=90^\circ\) and the directional derivative is \(0\).
When exact symbolic partial derivatives are not being used, the gradient can be estimated numerically. A centered finite difference estimates each partial derivative by moving a small distance \(h\) forward and backward along one coordinate axis. For example, in two variables,
$$
f_x(a,b)\approx \frac{f(a+h,b)-f(a-h,b)}{2h}
$$
and
$$
f_y(a,b)\approx \frac{f(a,b+h)-f(a,b-h)}{2h}.
$$
In general coordinate form, the component in the \(i\)-th coordinate direction is approximated by
$$
\frac{\partial f}{\partial x_i}(a)\approx \frac{f(a+h e_i)-f(a-h e_i)}{2h},
$$
where \(e_i\) is the unit coordinate vector in that direction. The value of \(h\) matters: a large \(h\) can miss local behavior, while an extremely small \(h\) can run into floating-point roundoff problems.
Examples of Directional Derivatives in Practice
Example 1: A Direction of Steepest Increase
Let
$$
f(x,y)=x^2+y^2
$$
at the point \((1,2)\), with direction vector
$$
v=\left\langle 2,4\right\rangle.
$$
First find the gradient:
$$
\nabla f(x,y)=\left\langle 2x,2y\right\rangle.
$$
At \((1,2)\),
$$
\nabla f(1,2)=\left\langle 2,4\right\rangle.
$$
Now normalize the direction vector:
$$
\|v\|=\sqrt{2^2+4^2}=\sqrt{20}=2\sqrt{5}.
$$
So
$$
u=\frac{v}{\|v\|}=\left\langle \frac{1}{\sqrt{5}},\frac{2}{\sqrt{5}}\right\rangle.
$$
The directional derivative is
$$
D_u f(1,2)=\left\langle 2,4\right\rangle\cdot\left\langle \frac{1}{\sqrt{5}},\frac{2}{\sqrt{5}}\right\rangle.
$$
$$
D_u f(1,2)=\frac{2}{\sqrt{5}}+\frac{8}{\sqrt{5}}=\frac{10}{\sqrt{5}}=2\sqrt{5}\approx 4.4721.
$$
Because the direction vector points the same way as the gradient, this is also the maximum possible directional derivative at that point.
Example 2: A Three-Dimensional Directional Derivative
Let
$$
f(x,y,z)=xy+z^2
$$
at the point \((1,2,-1)\), with direction vector
$$
v=\left\langle 3,-1,2\right\rangle.
$$
The gradient is
$$
\nabla f(x,y,z)=\left\langle y,x,2z\right\rangle.
$$
At \((1,2,-1)\),
$$
\nabla f(1,2,-1)=\left\langle 2,1,-2\right\rangle.
$$
Normalize the direction vector:
$$
\|v\|=\sqrt{3^2+(-1)^2+2^2}=\sqrt{14}.
$$
So
$$
u=\left\langle \frac{3}{\sqrt{14}},\frac{-1}{\sqrt{14}},\frac{2}{\sqrt{14}}\right\rangle.
$$
Now compute the dot product:
$$
D_u f(1,2,-1)=\left\langle 2,1,-2\right\rangle\cdot\left\langle \frac{3}{\sqrt{14}},\frac{-1}{\sqrt{14}},\frac{2}{\sqrt{14}}\right\rangle.
$$
$$
D_u f(1,2,-1)=\frac{6}{\sqrt{14}}-\frac{1}{\sqrt{14}}-\frac{4}{\sqrt{14}}=\frac{1}{\sqrt{14}}\approx 0.2673.
$$
The result is positive, so the function is increasing slightly in that direction at the point.
Example 3: A Zero-Gradient Edge Case
Let
$$
f(x,y)=x^2+y^2
$$
at the point \((0,0)\). The gradient is
$$
\nabla f(x,y)=\left\langle 2x,2y\right\rangle,
$$
so
$$
\nabla f(0,0)=\left\langle 0,0\right\rangle.
$$
For any nonzero direction vector \(v\), the unit direction \(u\) exists, but
$$
D_u f(0,0)=\left\langle 0,0\right\rangle\cdot u=0.
$$
At this point, there is no nonzero gradient direction to compare angles with. The function is locally flat to first order, even though it may still curve upward nearby.
How to Interpret the Result
The directional derivative is a local rate of change. It describes what happens at the entered point as you begin moving in the normalized direction vector.
A positive directional derivative means the function is increasing in that direction at the point. A larger positive value means a faster local increase per coordinate unit.
A negative directional derivative means the function is decreasing in that direction at the point. A more negative value means a faster local decrease.
A value near zero means the function is locally level in that direction, at least within the calculator’s numerical tolerance. This often happens when the direction is close to perpendicular to the gradient, or when the gradient itself is near zero.
The function value \(f(a)\) is the value of the function at the entered point. It is not the rate of change; it is the height, temperature, model output, or function output at that location.
The gradient shows the estimated vector of partial derivatives. Its magnitude \(\|\nabla f(a)\|\) is the largest possible directional derivative at the point. The largest decrease has value \(-\|\nabla f(a)\|\) and occurs in the opposite direction.
The unit direction is the normalized version of the direction vector. This is the direction actually used in the dot product. The magnitude of the original direction vector does not affect the directional derivative after normalization.
The angle to the gradient describes alignment. An angle near \(0^\circ\) means the chosen direction is close to steepest increase. An angle near \(90^\circ\) means the function is locally level in that direction. An angle near \(180^\circ\) means the direction is close to steepest decrease.
For a three-variable function, a two-dimensional contour preview should be read as a slice, not a full three-dimensional surface. It can help visualize local behavior in the \(x\)-\(y\) plane at the selected \(z\) value, but it does not show every direction in three-dimensional space.
Common Mistakes and Misconceptions
One common mistake is using a direction vector with the wrong number of components. A two-variable function needs a two-component point and a two-component direction vector. A three-variable function needs three of each.
Another mistake is entering the zero vector as a direction. A direction vector must have nonzero length, because the unit direction \(v/\|v\|\) is undefined when \(\|v\|=0\).
A larger direction vector does not make the directional derivative larger. The vector \(\left\langle 2,4\right\rangle\) and the vector \(\left\langle 200,400\right\rangle\) point in the same direction, so they produce the same unit direction after normalization.
It is also easy to confuse the function value with the directional derivative. The value \(f(a)\) tells where the function is at the point. The directional derivative tells how the function begins changing from that point in a chosen direction.
When entering formulas, use the variables that match the selected dimension. Use \(x\) and \(y\) in 2D mode, and \(x\), \(y\), and \(z\) in 3D mode. Use explicit multiplication between variables, such as x*y, rather than xy. Use periods for decimals, such as 3.14, rather than comma decimal notation.
The entered step is an initial scale rather than a fixed difference. The calculator accepts \(0<h\leq1\), repeatedly halves coordinate-scaled steps, compares successive centered differences, and reports the accepted steps and estimated numerical error. A poor initial scale can still prevent reliable convergence; safeguards can report unresolved precision or convergence, but numerical estimates should still be checked independently.
A gradient requires the function to be differentiable at the selected point. Functions such as abs, min, max, floor, and ceil may be valid away from corners or jumps. When samples give stable evidence of a corner or jump, the calculator reports that the function is not differentiable; other unresolved estimates report a neutral convergence limitation. A numerical gradient remains an approximation based on nearby function evaluations, not an exact symbolic derivative.
When to Use Directional Derivatives
Use a directional derivative when you need to study local change in a specific direction rather than only along a coordinate axis.
This is especially useful for:
-
comparing increase or decrease along different movement directions;
-
checking whether a direction is close to steepest increase or steepest decrease;
-
understanding contour plots and level curves;
-
studying functions of two or three variables in calculus courses;
-
estimating local sensitivity when multiple inputs change together;
-
connecting gradient vectors with dot products and unit vectors.
If you only need the rate of change with respect to one variable while holding the others fixed, a partial derivative may be enough. If you need the rate of change along a general vector, use a directional derivative.
Limitations and Things to Keep in Mind
The standard gradient-dot-direction formula assumes the function is differentiable at the point. If the function has a corner, cusp, discontinuity, vertical behavior, or undefined value near the point, a directional derivative may fail to exist or a numerical estimate may be misleading.
This calculator works with real-valued functions in 2D or 3D. It does not handle complex-valued functions, dimensions other than two or three, or exact symbolic simplification.
The direction vector is used only for its direction. Its length is normalized away before the directional derivative is computed.
The gradient is estimated numerically using centered differences. That means the entered step \(h\) affects the result. For smooth functions, centered differences are often a good practical approximation, but no single \(h\) is best for every function and every point.
Displayed values may be rounded for readability. The near-zero threshold affects the status interpretation and angle note, not the ordinary numeric formatter: a finite nonzero result can still be displayed in exponential notation. Very large or very small values may use exponential notation.
The contour preview is a visual aid, not a proof of the result. It may be incomplete if nearby function samples are not finite. In 3D mode, the preview is an \(x\)-\(y\) slice at the entered \(z\) value, not a full three-dimensional graph.
No physical unit conversion is applied. If your variables represent physical quantities, the directional derivative has the function’s units per coordinate unit, assuming the coordinate axes are being compared on the same scale.
How to Use This Calculator
-
Choose whether the function uses 2D variables \((x,y)\) or 3D variables \((x,y,z)\).
-
Enter the function expression. Supported constants include
pi and e; supported functions include sin, cos, tan, sqrt, abs, exp, ln, log, floor, ceil, min, and max.
-
Enter the point with the correct number of finite coordinates. Values may be separated by spaces, commas, or semicolons.
-
Enter a nonzero direction vector with the same number of components as the point.
-
Enter a finite-difference step with \(0<h\leq1\).
-
Review the directional derivative, status message, \(f(\text{point})\), estimated gradient, unit direction, angle note, calculation steps, and contour preview.
-
Use the example presets to test sample 2D or 3D inputs, or download the visual preview as a PNG when available.
Frequently Asked Questions
Does the direction vector have to be a unit vector?
No. You may enter any nonzero direction vector with the correct number of components. The calculator normalizes it to a unit vector before computing the dot product.
Why does the calculator normalize the direction vector?
A directional derivative is a rate of change per one unit of distance in the chosen direction. Normalizing the vector makes the result depend only on direction, not on how long the entered vector happens to be.
What does a negative directional derivative mean?
A negative directional derivative means the function is decreasing at the point as you move in the chosen direction. Moving in the opposite direction would reverse the sign for a differentiable function at the same point.
What does it mean if the gradient is zero?
If the gradient is zero, every gradient-based directional derivative at that point is zero. There is no nonzero gradient direction for the angle comparison, and the point may be a local maximum, local minimum, saddle point, or another type of stationary point.
How should I choose the finite-difference step \(h\)?
Choose a value with \(0<h\leq1\) that is small enough to measure local behavior but not so small that roundoff dominates. If you are unsure, compare results from a few nearby accepted values of \(h\); stable results are usually more trustworthy than results that change dramatically.
Sources and References
Books and Textbooks
-
Gilbert Strang and Edwin “Jed” Herman. Calculus Volume 3. OpenStax, 2016. Section 4.6, “Directional Derivatives and the Gradient.” https://openstax.org/books/calculus-volume-3/pages/4-6-directional-derivatives-and-the-gradient
-
Joel Feldman, Andrew Rechnitzer, and Elyse Yeager. CLP-3 Multivariable Calculus. University of British Columbia / Mathematics LibreTexts, updated May 28, 2023. Section 2.7, “Directional Derivatives and the Gradient.” https://math.libretexts.org/Bookshelves/Calculus/CLP-3MultivariableCalculus%28FeldmanRechnitzerandYeager%29/02%3APartialDerivatives/2.07%3ADirectionalDerivativesandthe_Gradient
-
Qingkai Kong, Timmy Siauw, and Alexandre Bayen. Python Programming and Numerical Methods: A Guide for Engineers and Scientists. Academic Press, 2020. Chapter 20, “Numerical Differentiation,” especially “Finite Difference Approximating Derivatives.” https://pythonnumericalmethods.studentorg.berkeley.edu/notebooks/chapter20.02-Finite-Difference-Approximating-Derivatives.html
Online Educational Sources
-
MIT OpenCourseWare. “Part B: Chain Rule, Gradient and Directional Derivatives.” 18.02SC Multivariable Calculus, accessed June 28, 2026. https://ocw.mit.edu/courses/18-02sc-multivariable-calculus-fall-2010/pages/2.-partial-derivatives/part-b-chain-rule-gradient-and-directional-derivatives/
