What Is the Jacobian?
A map from the plane to the plane, (x, y) ↦ (u(x, y), v(x, y)), has four first partial derivatives, and the Jacobian matrix organizes them:
J = [ ∂u/∂x ∂u/∂y ; ∂v/∂x ∂v/∂y ]
J is the map's local linearization: near any point, the possibly wild nonlinear transformation acts, to first order, like multiplication by this matrix, the multivariable version of "zoom in far enough and everything looks linear." Its determinant, det J = uxvy − uyvx carries the headline geometry: it is the local area scale factor. A tiny patch of area dA near the point lands as a patch of area |det J|·dA after the map; a negative determinant additionally means the map flips orientation, like a reflection.
How to Use This Calculator
Enter both component functions. The calculator computes the four partials symbolically, assembles J, states the determinant formula, and, given a point, evaluates the matrix and det J numerically, reporting the area scale and flagging orientation reversal when the determinant is negative. The chips include the two most famous Jacobians in mathematics: the polar-coordinate map (whose determinant r is the extra factor in every polar integral) and the complex squaring map z ↦ z².
Worked Example: Why Polar Integrals Grow an r
The polar map is x-as-radius: u = x·cos(y), v = x·sin(y), with x playing r and y playing θ. Partials:
ux = cos y, uy = −x sin y, vx = sin y, vy = x cos y
det J = cos y · x cos y − (−x sin y) · sin y = x(cos²y + sin²y) = x that is, det J = r.
This single line explains the mysterious r in ∬f·r dr dθ: converting an integral to polar coordinates rescales area patches by exactly |det J| = r, so patches near the origin count for little and distant ones for more. Every change-of-variables factor in every coordinate system, spherical's ρ²sin φ included, is a Jacobian determinant earning its keep.
What the Determinant's Value Tells You
| det J | Local behavior |
|---|---|
| > 1 | Expands areas, preserves orientation |
| between 0 and 1 | Shrinks areas, preserves orientation |
| < 0 | Flips orientation (mirror-like), scales by |det J| |
| = 0 | Collapses areas, the map crushes the neighborhood onto a curve or point; not locally invertible |
The zero case is the inverse function theorem's warning light: where det J ≠ 0, the map is locally invertible with the inverse's Jacobian being J⁻¹; where det J = 0, inversion breaks down, the multivariable analogue of f'(x) = 0 obstructing an inverse function.
Common Mistakes to Avoid
- Scrambling rows and columns. Convention: each row is one output's gradient, row one is (ux uy), row two (vx vy). Transposing changes nothing for the determinant but corrupts matrix-vector uses.
- Dropping the absolute value in integrals. Change of variables uses |det J|, areas are positive even when the map reverses orientation. The sign is information, but not for the integral.
- Confusing Jacobian and Hessian. The Jacobian holds first partials of a vector map; the Hessian holds second partials of a scalar function. (They meet: the Hessian of f is the Jacobian of ∇f.)
- Evaluating symbolically forever. Most applications need det J at a specific point, the local scale there. The symbolic determinant is the map's whole scaling story; the number is today's answer.
Real-World Applications
Change of variables in probability runs on Jacobians: transforming random variables (say, from (X, Y) to (X+Y, X−Y), or to polar form in the Box–Muller normal-generator every statistics library ships) rescales density functions by exactly 1/|det J|, get the Jacobian wrong and probabilities stop summing to one. Machine learning's normalizing flows build entire generative models from chains of maps chosen so their Jacobian determinants stay computable, because the model's likelihood is a product of them.
In robotics, the Jacobian relates joint velocities to end-effector velocity, the fundamental object of manipulator control, and configurations where det J = 0 are singularities poses where the arm loses a direction of motion and controllers must tread carefully (the straight-elbow arm is the classic example). Fluid dynamics tracks deformation of material elements by Jacobians of the flow map; continuum mechanics measures strain the same way; and the complex-analysis chips illustrate a gem: maps coming from complex functions have det J = |f′(z)|² ≥ 0, they can never flip orientation, one reason conformal maps preserve angles.
Frequently Asked Questions
Why is the determinant the area scale factor?
A linear map sends the unit square to a parallelogram whose area is |det|, a fact of linear algebra. Since J is the map's local linear stand-in, tiny squares near the point map to tiny parallelograms scaled by |det J|; integrating that local fact over a region is the change-of-variables theorem.
What shape does the Jacobian take for maps between different dimensions?
An m×n matrix for a map from n inputs to m outputs, gradients of scalar functions are 1×n Jacobians, parametric curves have n×1 ones. Only square Jacobians have determinants; rectangular ones still linearize the map, and their singular values play the scaling role.
How does the chain rule look in Jacobian language?
Beautifully: the Jacobian of a composition is the matrix product of the Jacobians, Jg∘f = Jg·Jf. The one-variable chain rule is the 1×1 case; backpropagation through a neural network is this identity applied layer by layer.
What does the complex squaring chip demonstrate?
The map (x² − y², 2xy) is the complex function z ↦ z² in disguise, whose derivative is 2z, so det J = |2z|² = 4(x² + y²), which at (1, 1) gives 8. The chip confirms it numerically, and the general lesson is elegant: for any map coming from a complex function f, det J = |f′(z)|², automatically non-negative.
What is the inverse function theorem in one sentence?
Wherever det J ≠ 0, the map is locally invertible and the inverse is differentiable with Jacobian J⁻¹, nonzero local scaling guarantees nothing was crushed, so the transformation can be undone near the point.
What does the sign of the Jacobian determinant mean?
A positive determinant means the transformation preserves orientation locally, while a negative one means it flips orientation, like a reflection. The absolute value gives the local area or volume scaling factor used in the change-of-variables formula for multiple integrals.