Nabla: One Operator, Three Questions
How does the nabla operator turn local change into gradient, divergence and curl?
How this lesson unfolds.
Begin with Why can one triangle describe three kinds of change? The sequence then connects A field assigns something everywhere and builds toward Parse the expression before calculating.
Why can one triangle describe three kinds of change?
The operator nabla combines directional derivatives; attached to a scalar or vector field with no product, a dot product or a cross product, it produces gradient, divergence or curl.
- Helpful foundationsfunctions · vectors · partial derivatives
- Next connectionA field assigns something everywhere
More context
A large nabla symbol branches to gradient, divergence and curl, showing the input, output and question for each operation.
What does the word field add to a number or vector?
A scalar field assigns one number to every point, while a vector field assigns a directed quantity to every point in a region.
- Helpful foundationsMeet del, also called nabla
- Next connectionA vector is a local instruction
More context
Paired cards contrast a scalar with a vector, then a scalar field with a vector field, emphasizing that a field assigns a quantity throughout a region.
How should we read a vector attached to a point?
At a point P=(1,2), the vector (2,4) describes direction and magnitude of a possible displacement or flow; it is not the destination point.
At a point P=(1,2), the vector (2,4) describes direction and magnitude of a possible displacement or flow; it is not the destination point.
- Helpful foundationsA field assigns something everywhere
- Next connectionA scalar field becomes a landscape
More context
A point P at coordinates one, two carries the vector two, four as a displacement instruction; adding them produces the destination three, six.
How can one number at every point become a surface?
For f(x,y)=x squared plus y squared, each coordinate selects a location and the function value supplies its height, producing a bowl-shaped surface.
For f(x,y)=x squared plus y squared, each coordinate selects a location and the function value supplies its height, producing a bowl-shaped surface.
- Helpful foundationsA vector is a local instruction
- Next connectionThe gradient collects directional slopes
More context
Concentric level curves for f of x y equals x squared plus y squared show higher values farther from the bowl's center.
How does nabla turn two partial derivatives into one vector?
For f(x,y)=x squared plus y squared, the partial derivatives are 2x and 2y, so the gradient at (1,2) is the vector (2,4).
For f(x,y)=x squared plus y squared, the partial derivatives are 2x and 2y, so the gradient at (1,2) is the vector (2,4).
- Helpful foundationsA scalar field becomes a landscape
- Next connectionGradient gives direction; step size gives distance
More context
The x and y partial derivatives are evaluated at one, two and combined into the gradient vector two, four.
Why is the gradient vector not another point?
The gradient (2,4) points in the direction of steepest increase. A chosen step size eta converts it into a displacement, such as (1,2)+0.1(2,4)=(1.2,2.4).
A chosen step size eta converts it into a displacement, such as (1,2)+0.1(2,4)=(1.2,2.4).
- Helpful foundationsThe gradient collects directional slopes
- Next connectionA gradient is local
More context
A gradient arrow at P is scaled by eta point one to move from one, two to one point two, two point four.
Why must we calculate the gradient again after moving?
For a curved field, the gradient depends on position; moving to a new point changes the local arrow, so a trajectory repeats evaluate, move and reevaluate.
- Helpful foundationsGradient gives direction; step size gives distance
- Next connectionThe same geometry trains a model
More context
Four positions along a curved path carry different local gradient arrows, followed by an evaluate-and-move loop.
How does a height landscape become an AI loss landscape?
Replace coordinates by model parameters and height by loss. Gradient descent updates parameters opposite the loss gradient: w new equals w old minus eta times the gradient of L.
Gradient descent updates parameters opposite the loss gradient: w new equals w old minus eta times the gradient of L.
- Helpful foundationsA gradient is local
- Next connectionAttachment determines the question
More context
A loss landscape shows the negative gradient pointing downhill beside the gradient descent update for model parameters.
What changes when nabla meets a scalar field or a vector field?
Nabla acting on a scalar gives a gradient vector; dotting nabla with a vector field gives scalar divergence; crossing nabla with a three-dimensional vector field gives vector curl.
- Helpful foundationsThe same geometry trains a model
- Next connectionDivergence matches change to flow
More context
Three rows map nabla f, nabla dot F and nabla cross F to their input and output types and intuitive questions.
How does the dot product compress local flow change into one number?
In two dimensions, divergence adds the x-change of the x-component and the y-change of the y-component: partial Fx by x plus partial Fy by y.
- Helpful foundationsAttachment determines the question
- Next connectionRead the sign of divergence
More context
Matched cards connect x-change with x-flow and y-change with y-flow before adding them in the divergence formula.
Is the local flow expanding, compressing or balanced?
F=(x,y) has divergence 2, F=(-x,-y) has divergence -2, and constant F=(2,0) has divergence 0; the sign measures net local outflow, not simple arrow direction.
F=(x,y) has divergence 2, F=(-x,-y) has divergence -2, and constant F=(2,0) has divergence 0; the sign measures net local outflow, not simple arrow direction.
- Helpful foundationsDivergence matches change to flow
- Next connectionCharge density creates electric-field divergence
More context
Three vector fields show outward flow with divergence two, inward flow with divergence minus two and constant flow with divergence zero.
How does Gauss's law connect a source to a field?
In differential form, divergence of E equals rho divided by epsilon zero. Positive charge density produces positive divergence and negative charge density produces negative divergence.
In differential form, divergence of E equals rho divided by epsilon zero.
- Helpful foundationsRead the sign of divergence
- Next connectionCurl asks whether a tiny wheel turns
More context
Positive and negative charge-density cards sit above outward and inward electric fields beside differential Gauss law.
Why can moving flow have zero curl?
A constant translation field moves a tiny wheel without spinning it, while F=(-y,x) produces local counterclockwise rotation and nonzero curl.
A constant translation field moves a tiny wheel without spinning it, while F=(-y,x) produces local counterclockwise rotation and nonzero curl.
- Helpful foundationsCharge density creates electric-field divergence
- Next connectionCurl has strength and an axis
More context
A wheel in constant flow translates without spinning, while a wheel in the rotating field F equals negative y comma x turns.
Why is curl a vector in three dimensions?
The cross-product structure gives both rotational strength and an oriented axis; rotation in the xy-plane points along the z-axis by the right-hand rule.
- Helpful foundationsCurl asks whether a tiny wheel turns
- Next connectionCompute curl for a rotating field
More context
Rotation drawn in the xy-plane has an arrow along positive z, connecting curl magnitude with its oriented axis.
How do circular arrows become the vector 2k?
For F=(-y,x,0), the z-component of curl is partial x by x minus partial negative y by y, equal to 1-(-1)=2, so curl F=2k.
For F=(-y,x,0), the z-component of curl is partial x by x minus partial negative y by y, equal to 1-(-1)=2, so curl F=2k.
- Helpful foundationsCurl has strength and an axis
- Next connectionParse the expression before calculating
More context
The z-component of curl for F equals negative y comma x comma zero is calculated as one minus negative one, producing two k.
How can the notation itself reveal the output and the question?
Identify the attached object and connector: nabla f asks fastest scalar increase, nabla dot F asks local expansion, and nabla cross F asks local rotation.
- Helpful foundationsCompute curl for a rotating field
- Next connectionline integrals and the fundamental theorems of vector calculus
More context
A final three-row guide parses gradient, divergence and curl by expression, input object, output type and physical question.