324 Legendre Transform
A mapping that is used to replace a variable in a function by the derivative of that function with respect to the variable. The Legendre transform is a non-integral transformation, unlike, for example, the Laplace and Fourier transforms.
324.1 Applications
- Builds the Hamiltonian from the Lagrangian in classical mechanics.
- Changes independent variables in variational problems to canonical form.
- Defines thermodynamic potentials such as enthalpy and free energy.
- Converts convex optimization problems between dual variable sets.
- Relates action formulations that use velocity to formulations that use momentum.
definition [d] (Legendre Transform) From Courant and Hilbert: the Legendre transformation connects
- \(p = \dfrac{\partial F}{\partial\!\left(\dfrac{du}{dx}\right)}\)
with
- \(\Phi = p\dfrac{du}{dx} - F\) ,
and the inverse is
- \(\dfrac{du}{dx} = \dfrac{\partial\Phi}{\partial p}\) , \(\qquad F = p\dfrac{du}{dx} - \Phi\) .
where
- \(F\) is the original function.
- \(u\) is the dependent variable.
- \(x\) is the independent variable.
- \(\dfrac{du}{dx}\) is the derivative of \(u\) with respect to \(x\).
- \(p\) is the new independent variable.
- \(\Phi\) is the Legendre transform of \(F\).
definition [d] (Legendre Transform) From Cahill: for a function \(A(x,y)\), set
- \(v = \dfrac{\partial A}{\partial y}\)
and
- \(B = A(x,y) - v y\) ,
so the independent variables change from \((x,y)\) to \((x,v)\).
Note: Cahill’s form \(B=A-vy\) differs in sign from \(\Phi = p\dfrac{du}{dx}-F\). Both change the independent variable from the original one to its conjugate slope.
where
- \(A\) is the original function.
- \(x\) and \(y\) are the original independent variables.
- \(v\) is the new variable.
- \(B\) is the Legendre transform of \(A\) in the \(y\) direction.
- \(\dfrac{\partial A}{\partial y}\) is the partial derivative of \(A\) with respect to \(y\).
324.2 Examples
324.2.1 Simple
For \(F(x) = \dfrac{1}{2}x^{2}\),
\[ p = \dfrac{dF}{dx} = x,\qquad G(p) = p\, x - F(x) = \dfrac{1}{2}p^{2} \]
where
- \(F\) is the original function.
- \(x\) is the original variable.
- \(p\) is the new variable.
- \(G\) is the Legendre transform.
324.2.2 General
For \(F(x) = \dfrac{1}{2}m x^{2}\) with constant \(m > 0\),
\[ p = m x,\qquad x = \dfrac{p}{m},\qquad G(p) = \dfrac{p^{2}}{2m} \]
where
- \(F\) is the original function.
- \(m\) is a positive constant.
- \(x\) is the original variable.
- \(p\) is the new variable.
- \(G\) is the Legendre transform.
324.3 References
- Courant, R., and Hilbert, D. Methods of Mathematical Physics, Vol. 1. Wiley-VCH, 1991. — \(\Phi = p\dfrac{du}{dx}-F\) with \(p=\partial F/\partial(du/dx)\).
- Cahill, K. Physical Mathematics. Cambridge University Press, 2019. — \(v=\partial A/\partial y\) and \(B=A-vy\).
- MIT OpenCourseWare. 8.223 Classical Mechanics II, Lecture 15 (IAP 2017). PDF — Hamiltonian as Legendre transform of the Lagrangian.
- Basic Derivation of Jacobian
- Chain Rule
- Cross Product
- Curl
- Curve
- Definite Integral
- Derivation of Curl
- Derivation of Divergence
- Derivation of Grad
- Derivation of Jacobian
- Derivation of Taylor Expansion
- Divergence
- Dot Product
- Envelope of Tangent Lines
- General Functions
- Gradient
- Improper Integral
- Indefinite Integral
- Jacobian
- Legendre Transform
- Legendre Transform Derivation
- Limit
- Line Integral
- Notes
- Parameterization
- Partial Derivative
- Real Parameter
- Scalar Projection
- Smooth Curve
- Vector Calculus
- Vector Differential Operator
- Vector Field
- Vector Projection