What do the derivatives of $\Lambda$ with respect to the multipliers $\lambda_j$ tell you?

Each multiplier appears only in the term $-\lambda\cdot G(x)$. Differentiating gives $\partial \Lambda/\partial \lambda_j=-g_j(x)$. Setting these derivatives to zero forces $g_j(x)=0$ for every $j$, which is exactly the vector constraint $G(x)=0$. The entire $\lambda$-gradient collapses to $-G(x)$.

Multivariable Calculus 31

Q: How does the $x$-gradient condition relate to the original objective and the constraint gradients?

Differentiating $\Lambda$ with respect to $x$ yields $\nabla_x\Lambda(x,\lambda)=\nabla f(x)-\sum_{j=1}^m \lambda_j\nabla g_j(x)$. This is the stationarity condition under constraints: the gradient of the objective is balanced by a linear combination of the gradients of the constraint functions, with coefficients given by the multipliers.

Q: What role does the Jacobian regularity assumption for $G$ play?

The method is presented as a necessary-condition tool: if $x^*$ is a constrained local extremum of $f$ subject to $G(x)=0$, then—assuming the Jacobian of $G$ has the required regularity on the constraint set—there must exist multipliers $\lambda$ such that $\nabla_{x,\lambda}\Lambda(x^*,\lambda)=0$. Without this kind of regularity, exceptional points can break the guarantee.

Q: After solving $\nabla \Lambda=0$, how do you decide whether a candidate point is a maximum or minimum?

The transcript notes that for $C^2$ functions, a sufficient criterion can be formulated using the Hessian of the Lagrangian. In simpler applications, one may not need the full second-order test because existing knowledge about the shape of $f$ can determine whether the candidate corresponds to a maximum or minimum.

TL;DR

The Lagrangian $Λ (x, λ) = f (x) - λ \cdot G (x)$ packages both the objective and the constraints into one real-valued function.

Briefing Cornell Notes

Briefing

Constrained optimization in multivariable calculus gets a cleaner “recipe” once the Lagrange multipliers method is rewritten using a single object: the Lagrangian. Instead of solving separate equations for the constraint and for stationarity of the original function, the method bundles everything into one real-valued function $Λ (x, λ)$ whose gradient vanishes exactly when both the constraint $G (x) = 0$ and the multiplier-weighted gradient condition hold. This matters because it turns a constrained extremum problem into an unconstrained-looking system—making both computation and reasoning more systematic.

The setup starts with a target function $f \in C^{1} (R^{n}, R)$ and constraint functions $g_{1}, \dots, g_{m} \in C^{1} (R^{n}, R)$ , typically with $m \leq n$ . Writing the constraints as a vector map $G (x) = (g_{1} (x), \dots, g_{m} (x)) \in R^{m}$ , the constrained extrema candidates must satisfy $m$ constraint equations $G (x) = 0$ plus $n$ stationarity equations involving gradients. Those stationarity equations introduce $m$ unknown multipliers $λ_{1}, \dots, λ_{m}$ , giving enough freedom to match the $n$ gradient components.

The key simplification is the Lagrangian. Define a real-valued function $Λ : R^{n} \times R^{m} \to R$ by $Λ (x, λ) = f (x) - λ \cdot G (x),$ where $λ \cdot G (x)$ is the standard inner product in $R^{m}$ , i.e. $λ_{1} g_{1} (x) + \dots + λ_{m} g_{m} (x)$ . With this definition, the method becomes: compute the gradient of $Λ$ with respect to both sets of variables— $x \in R^{n}$ and $λ \in R^{m}$ —and set it to zero.

Concretely, the gradient components split into two groups. Differentiating with respect to the coordinates of $x$ yields a condition built from gradients: the $x$ -part becomes $\nabla f (x) - \sum_{j = 1}^{m} λ_{j} \nabla g_{j} (x)$ . Differentiating with respect to each multiplier gives $\partial Λ/ \partial λ_{j} = - g_{j} (x)$ , so the entire $λ$ -gradient collapses to $- G (x)$ . Therefore, $\nabla_{x, λ} Λ (x, λ) = 0$ holds exactly when both $\nabla f (x) - \sum_{j = 1}^{m} λ_{j} \nabla g_{j} (x) = 0$ and $G (x) = 0$ are satisfied. The constrained extremum candidates are the $x$ -coordinates of solutions to this combined system.

Regularity assumptions matter for the method’s reliability: if the Jacobian of $G$ has the right rank (described as a “subjective map” condition in the transcript), then at a constrained local extremum $x^{*}$ there exist multipliers $λ$ such that the Lagrangian gradient condition holds. The practical workflow is also emphasized: (1) correctly formulate the constraint function $G$ , (2) check the Jacobian regularity on the constraint set ${x : G (x) = 0}$ to avoid exceptional points, (3) build $Λ$ and solve $\nablaΛ = 0$ to get a finite list of candidate points, and (4) decide whether each candidate is a maximum or minimum, typically using second-order information via the Hessian of $Λ$ for $C^{2}$ functions—though in many applications prior knowledge about $f$ suffices. The series closes by framing this Lagrangian-based method as the necessary-condition engine for extrema under constraints.

Cornell Notes

The Lagrange multipliers method can be rewritten using a single function, the Lagrangian $Λ (x, λ)$ , which turns constrained optimization into a system where a gradient must vanish. For constraints $G (x) = 0$ with $G : R^{n} \to R^{m}$ and objective $f : R^{n} \to R$ , the Lagrangian is defined as $Λ (x, λ) = f (x) - λ \cdot G (x)$ . Setting $\nabla_{x, λ} Λ (x, λ) = 0$ produces two conditions at once: $\partial Λ/ \partial λ_{j} = - g_{j} (x)$ , which forces $G (x) = 0$ , and $\nabla_{x} Λ = \nabla f (x) - \sum_{j = 1}^{m} λ_{j} \nabla g_{j} (x) = 0$ , which gives the multiplier-weighted stationarity condition. Under a Jacobian regularity assumption, constrained local extrema must satisfy these equations for some $λ$ .

Why does defining the Lagrangian $Λ (x, λ) = f (x) - λ \cdot G (x)$ simplify constrained extrema problems?

It consolidates the constraint equations and the stationarity equations into one gradient condition. When

\nabla_{x, λ} Λ (x, λ) = 0

, the

λ

-derivatives automatically reproduce the constraints

G (x) = 0

, while the

x

-derivatives reproduce the weighted gradient balance

\nabla f (x) - \sum_{j = 1}^{m} λ_{j} \nabla g_{j} (x) = 0

. Instead of solving two separate sets of equations, both appear as parts of a single system.

What do the derivatives of $Λ$ with respect to the multipliers $λ_{j}$ tell you?

Each multiplier appears only in the term

- λ \cdot G (x)

. Differentiating gives

\partial Λ/ \partial λ_{j} = - g_{j} (x)

. Setting these derivatives to zero forces

g_{j} (x) = 0

for every

j

, which is exactly the vector constraint

G (x) = 0

. The entire

λ

-gradient collapses to

- G (x)

How does the $x$ -gradient condition relate to the original objective and the constraint gradients?

Differentiating

Λ

with respect to

x

yields

\nabla_{x} Λ (x, λ) = \nabla f (x) - \sum_{j = 1}^{m} λ_{j} \nabla g_{j} (x)

. This is the stationarity condition under constraints: the gradient of the objective is balanced by a linear combination of the gradients of the constraint functions, with coefficients given by the multipliers.

What role does the Jacobian regularity assumption for $G$ play?

The method is presented as a necessary-condition tool: if

x^{*}

is a constrained local extremum of

f

subject to

G (x) = 0

, then—assuming the Jacobian of

G

has the required regularity on the constraint set—there must exist multipliers

λ

such that

\nabla_{x, λ} Λ (x^{*}, λ) = 0

. Without this kind of regularity, exceptional points can break the guarantee.

After solving $\nablaΛ = 0$ , how do you decide whether a candidate point is a maximum or minimum?

The transcript notes that for

C^{2}

functions, a sufficient criterion can be formulated using the Hessian of the Lagrangian. In simpler applications, one may not need the full second-order test because existing knowledge about the shape of

f

can determine whether the candidate corresponds to a maximum or minimum.

Review Questions

Given $f$ and constraints $G (x) = 0$ , write the Lagrangian $Λ (x, λ)$ and state what equations result from setting $\nabla_{x, λ} Λ = 0$ .
Explain why $\partial Λ/ \partial λ_{j} = - g_{j} (x)$ forces the constraint $G (x) = 0$ .
What regularity condition on $G$ is needed so that constrained local extrema must satisfy the Lagrangian gradient condition for some multipliers?

Key Points

1
The Lagrangian $Λ (x, λ) = f (x) - λ \cdot G (x)$ packages both the objective and the constraints into one real-valued function.
2
Solving $\nabla_{x, λ} Λ (x, λ) = 0$ produces both the constraint equations $G (x) = 0$ and the stationarity condition $\nabla f (x) - \sum_{j = 1}^{m} λ_{j} \nabla g_{j} (x) = 0$ .
3
Each multiplier derivative satisfies $\partial Λ/ \partial λ_{j} = - g_{j} (x)$ , so the $λ$ -gradient vanishing is equivalent to satisfying all constraints.
4
Under Jacobian regularity assumptions for $G$ , any constrained local extremum must admit multipliers $λ$ that make $\nabla_{x, λ} Λ = 0$ true.
5
A practical workflow is: formulate $G$ , check Jacobian regularity on ${x : G (x) = 0}$ , build $Λ$ , solve $\nablaΛ = 0$ , then apply a maximum/minimum test.
6
For $C^{2}$ problems, the Hessian of the Lagrangian can provide a sufficient criterion to classify candidates, though simpler reasoning may suffice in practice.

Highlights

The Lagrangian gradient condition ∇x,λ​Λ=0 simultaneously enforces the constraints G(x)=0 and the multiplier-weighted stationarity of f.

Because ∂Λ/∂λj​=−gj​(x), the multiplier equations are nothing more than the original constraint functions set to zero.

The stationarity equation takes the form ∇f(x)=∑j=1m​λj​∇gj​(x), meaning the objective gradient lies in the span of constraint gradients.

Regularity of the Jacobian of G is what turns the Lagrange-multiplier equations into a reliable necessary condition for constrained local extrema.

Topics

Lagrangian
Lagrange Multipliers
Constrained Extrema
Gradient Conditions
Jacobian Regularity

Multivariable Calculus 31 | Lagrangian

Briefing

Cornell Notes

Why does defining the Lagrangian $Λ (x, λ) = f (x) - λ \cdot G (x)$ simplify constrained extrema problems?

What do the derivatives of $Λ$ with respect to the multipliers $λ_{j}$ tell you?

How does the $x$ -gradient condition relate to the original objective and the constraint gradients?

What role does the Jacobian regularity assumption for $G$ play?

After solving $\nablaΛ = 0$ , how do you decide whether a candidate point is a maximum or minimum?

Review Questions

Key Points

Highlights

Topics

Get summaries like this for any content

Briefing

Cornell Notes

Why does defining the Lagrangian Λ(x,λ)=f(x)−λ⋅G(x) simplify constrained extrema problems?

What do the derivatives of Λ with respect to the multipliers λj​ tell you?

How does the x-gradient condition relate to the original objective and the constraint gradients?

What role does the Jacobian regularity assumption for G play?

After solving ∇Λ=0, how do you decide whether a candidate point is a maximum or minimum?

Review Questions

Key Points

Highlights

Topics

Get summaries like this for any content

Why does defining the Lagrangian $Λ (x, λ) = f (x) - λ \cdot G (x)$ simplify constrained extrema problems?

What do the derivatives of $Λ$ with respect to the multipliers $λ_{j}$ tell you?

How does the $x$ -gradient condition relate to the original objective and the constraint gradients?

What role does the Jacobian regularity assumption for $G$ play?

After solving $\nablaΛ = 0$ , how do you decide whether a candidate point is a maximum or minimum?