Multivariable Calculus

How local change, approximation, integration, and constraints work when functions depend on several variables at once.
Modified

April 26, 2026

Keywords

multivariable calculus, partial derivatives, gradient, Jacobian, Hessian, chain rule

1 Why This Module Matters

Single-variable calculus teaches local change when there is one input.

Multivariable calculus asks what happens when there are many coordinates, many directions of motion, and many ways to vary the system.

That is the setting for most real problems in CS, AI, and engineering:

  • a loss depends on many parameters
  • a model depends on many features
  • a physical state depends on space coordinates and time
  • a density depends on several variables at once
  • constraints and objectives interact in high-dimensional spaces

This module is the bridge that makes gradients, Jacobians, Hessians, and constrained local models feel natural instead of abrupt.

Prerequisites Single-variable calculus first. Linear algebra first. Both matter here.

Unlocks Optimization, probability densities, dynamical systems, modern ML math

Research Use Reading gradient-based arguments, local models, curvature claims, and multivariable approximations in papers

2 First Pass Through This Module

The intended first-pass order is:

  1. Partial Derivatives and Gradients
  2. Chain Rule and Linearization
  3. Jacobians and Hessians
  4. Multiple Integrals
  5. Constrained Optimization
  6. Vector Fields and Divergence / Curl

4 Core Concepts

  • Partial Derivatives and Gradients: introduces coordinate-wise rates of change, directional sensitivity, and the gradient as the vector of steepest local increase.
  • Chain Rule and Linearization: explains how local effects compose through multistage maps and how differentiable functions become local linear maps.
  • Jacobians and Hessians: turns multivariable derivatives into matrix language for transformation and curvature.
  • Multiple Integrals: turns accumulation into several variables, including density, mass, probability, and volume viewpoints.
  • Constrained Optimization: gives the calculus-side bridge into Lagrange multipliers before the full optimization module.
  • Vector Fields and Divergence / Curl: gives the field-view used in geometry, engineering, PDE intuition, and later vector-calculus theorems.

5 Proof Patterns In This Module

  • One coordinate at a time: partial derivatives isolate one direction while holding the others fixed.
  • Best local linear model: multivariable differentiability means a linear map approximates the function nearby.
  • Geometry of level sets: gradients point normal to level surfaces and encode local direction of fastest increase.

6 Applications

6.1 Optimization And Machine Learning

The language of gradient descent, backpropagation, Jacobians, Hessians, and constrained updates lives here. This module is the real mathematical bridge between first calculus and the optimization/ML layer already on the site.

6.2 Probability, Geometry, And Engineering Models

Multivariable densities, coordinate transforms, surfaces, fields, and local sensitivity analysis all depend on the same multivariable tools. This is where calculus starts to look like the language of high-dimensional systems rather than only of curves.

7 Go Deeper By Topic

Start with Partial Derivatives and Gradients.

If that page feels slippery, revisit:

8 Optional Deep Dives After First Pass

The strongest official deeper references for this module are:

9 Study Order

The intended first pass is the six-step sequence above.

You are ready to move beyond the first-pass multivariable module when you can:

  • explain what a partial derivative means
  • compute simple partial derivatives
  • interpret the gradient as a local direction of steepest increase
  • connect level sets with gradient geometry
  • explain why chain rule is composition of local effects
  • write a first multivariable linearization at a point
  • explain why the Jacobian is a local linear map and why the Hessian captures curvature
  • explain why constrained extrema obey tangency logic rather than unconstrained critical-point logic
  • explain why a double or triple integral is accumulation over a region rather than just a repeated symbol pattern
  • explain how divergence and curl diagnose different local behaviors of a vector field

10 Sources and Further Reading

Back to top