Lagrangian Systems
The Euler-Lagrange equations, that is the dynamical equations of a Lagrangian system, are given in terms of the Lagrangian $L(x,v)$ by
\[\frac{d}{dt} \frac{\partial L}{\partial v} - \frac{\partial L}{\partial x} = 0 .\]
For regular (i.e. non-degenerate) Lagrangians, this is a set of second-order ordinary differential equations. In many numerical applications, it is advantageous to solve the implicit form of these equations, given by
\[\begin{align*} \frac{d \vartheta}{dt} &= f , & \vartheta &= \frac{\partial L}{\partial v} , & f = \frac{\partial L}{\partial x} . \end{align*}\]
In the following, we show how these equations can be obtained for the example of a particle in a square potential.
Particle in a potential
Before any use, we need to load EulerLagrange:
using EulerLagrangeNext, we generate symbolic variables for a two-dimensional system:
t, x, v = lagrangian_variables(2)(t, (x(t))[1:2], (v(t))[1:2])With those variables, we can construct a Lagrangian
using LinearAlgebra
L = v ⋅ v / 2 - x ⋅ x / 2((v(t))[1]^2 + (v(t))[2]^2) / 2 + (-((x(t))[1]^2) - ((x(t))[2]^2)) / 2This Lagrangian together with the symbolic variables is then used to construct a LagrangianSystem:
lag_sys = LagrangianSystem(L, t, x, v)
Lagrangian system with
L = ((v(t))[1]^2 + (v(t))[2]^2 - ((x(t))[1]^2) - ((x(t))[2]^2)) / 2The constructor computes the Euler-Lagrange equations and generates the corresponding Julia code. In the last step, we can now construct a LODEProblem from the LagrangianSystem and some appropriate initial conditions, a time span to integrate over and a time step:
tspan = (0.0, 10.0)
tstep = 0.01
q₀ = [1.0, 1.0]
p₀ = [0.5, 2.0]
lprob = LODEProblem(lag_sys, tspan, tstep, q₀, p₀)Geometric Equation Problem for Lagrangian Ordinary Differential Equation (LODE)
with vector fields
ϑ = RuntimeGeneratedFunctions.RuntimeGeneratedFunction{(:ˍ₋out, :t, :X, :V, :params), EulerLagrange.var"#_RGF_ModTag", EulerLagrange.var"#_RGF_ModTag", (0xd20c7564, 0xd9f14488, 0x7c9632d0, 0x567848cf, 0x0ce88cdb), Expr}(quote
#= none:1 =#
#= none:2 =#
#= none:2 =# @inbounds begin
#= none:4 =#
begin
#= none:8 =#
#= none:8 =# @inbounds begin
#= none:10 =#
ˍ₋out[1] = getindex(V, 1)
#= none:11 =#
ˍ₋out[2] = getindex(V, 2)
#= none:13 =#
ˍ₋out
end
end
end
end)
f = RuntimeGeneratedFunctions.RuntimeGeneratedFunction{(:ˍ₋out, :t, :X, :V, :params), EulerLagrange.var"#_RGF_ModTag", EulerLagrange.var"#_RGF_ModTag", (0x02e60692, 0x993dfded, 0xb71e62d0, 0x11e23e6b, 0xb7b4f783), Expr}(quote
#= none:1 =#
#= none:2 =#
#= none:2 =# @inbounds begin
#= none:4 =#
begin
#= none:8 =#
#= none:8 =# @inbounds begin
#= none:10 =#
ˍ₋out[1] = -1 // 1 * getindex(X, 1)
#= none:11 =#
ˍ₋out[2] = -1 // 1 * getindex(X, 2)
#= none:13 =#
ˍ₋out
end
end
end
end)
g = RuntimeGeneratedFunctions.RuntimeGeneratedFunction{(:ˍ₋out, :t, :X, :Λ, :V, :params), EulerLagrange.var"#_RGF_ModTag", EulerLagrange.var"#_RGF_ModTag", (0x4df42040, 0x8fc17548, 0x11ffb7f0, 0xb9555f81, 0xc483cd8e), Expr}(quote
#= none:1 =#
#= none:2 =#
#= none:2 =# @inbounds begin
#= none:4 =#
begin
#= none:8 =#
#= none:8 =# @inbounds begin
#= none:10 =#
ˍ₋out[1] = (Differential(t, 1))(getindex(v(t), 1))
#= none:11 =#
ˍ₋out[2] = (Differential(t, 1))(getindex(v(t), 2))
#= none:13 =#
ˍ₋out
end
end
end
end)
Lagrangian: L = RuntimeGeneratedFunctions.RuntimeGeneratedFunction{(:t, :X, :V, :params), EulerLagrange.var"#_RGF_ModTag", EulerLagrange.var"#_RGF_ModTag", (0x20e03889, 0x548c5d55, 0x30ddf9f7, 0x6aedf8dc, 0x35473043), Expr}(quote
#= none:1 =#
#= none:2 =#
#= none:2 =# @inbounds begin
#= none:4 =#
begin
#= none:8 =#
(((getindex(V, 1) ^ 2 + getindex(V, 2) ^ 2) + -1 * getindex(X, 1) ^ 2) + -1 * getindex(X, 2) ^ 2) / 2
end
end
end)
Invariants:
GeometricBase.NullInvariants()
Timespan: (0.0, 10.0)
Timestep: 0.01
Initial conditions:
(t = fill(0.0), q = [1.0, 1.0], p = [0.5, 2.0], v = [0.0, 0.0])
Parameters:
GeometricBase.NullParameters()We can integrate this system using GeometricIntegrators:
using GeometricIntegrators
sol = integrate(lprob, Gauss(1))
using CairoMakie
fig = lines(parent(sol.q[:,1]), parent(sol.q[:,2]);
axis = (; xlabel = "x₁", ylabel = "x₂", title = "Particle moving in a square potential"),
figure = (; size = (800,600), fontsize = 22))┌ Warning: Hermite Extrapolation: q's history[1] and history[2] are identical!
└ @ GeometricIntegratorsBase ~/.julia/packages/GeometricIntegratorsBase/zEYTS/src/extrapolation/hermite.jl:209
Parameters
We can also include parametric dependencies in the Lagrangian. Consider, for example, a parameter α that determines the strength of the potential.
The easiest way, to account for parameters, is to create a named tuple with typical values for each parameter, e.g.,
params = (α = 5.0,)(α = 5.0,)In the next step, we use the function symbolize to generate a symbolic version of the parameters:
sparams = symbolize(params)(α = αₚ,)Now we modify the Lagrangian to account for the parameter:
L = v ⋅ v / 2 - sparams.α * (x ⋅ x) / 2((v(t))[1]^2 + (v(t))[2]^2) / 2 - (1//2)*((x(t))[1]^2 + (x(t))[2]^2)*αₚFrom here on, everything follows along the same lines as before, the only difference being that we also need to pass the symbolic parameters sparams to the LagrangianSystem constructor:
lag_sys = LagrangianSystem(L, t, x, v, sparams)
Lagrangian system with
L = ((v(t))[1]^2 + (v(t))[2]^2) / 2 - (1//2)*((x(t))[1]^2 + (x(t))[2]^2)*αₚAnalogously, we need to pass actual parameter values params to the LODEProblem constructor via the parameters keyword argument:
lprob = LODEProblem(lag_sys, tspan, tstep, q₀, p₀; parameters = params)Geometric Equation Problem for Lagrangian Ordinary Differential Equation (LODE)
with vector fields
ϑ = RuntimeGeneratedFunctions.RuntimeGeneratedFunction{(:ˍ₋out, :t, :X, :V, :params), EulerLagrange.var"#_RGF_ModTag", EulerLagrange.var"#_RGF_ModTag", (0xd20c7564, 0xd9f14488, 0x7c9632d0, 0x567848cf, 0x0ce88cdb), Expr}(quote
#= none:1 =#
#= none:2 =#
#= none:2 =# @inbounds begin
#= none:4 =#
begin
#= none:8 =#
#= none:8 =# @inbounds begin
#= none:10 =#
ˍ₋out[1] = getindex(V, 1)
#= none:11 =#
ˍ₋out[2] = getindex(V, 2)
#= none:13 =#
ˍ₋out
end
end
end
end)
f = RuntimeGeneratedFunctions.RuntimeGeneratedFunction{(:ˍ₋out, :t, :X, :V, :params), EulerLagrange.var"#_RGF_ModTag", EulerLagrange.var"#_RGF_ModTag", (0x75e7046d, 0xc0d46f16, 0xcf25a212, 0xac0cfd8a, 0xc8c2123c), Expr}(quote
#= none:1 =#
#= none:2 =#
#= none:2 =# @inbounds begin
#= none:4 =#
begin
#= none:8 =#
#= none:8 =# @inbounds begin
#= none:10 =#
ˍ₋out[1] = (-1 // 1 * getindex(X, 1)) * params.α
#= none:11 =#
ˍ₋out[2] = (-1 // 1 * getindex(X, 2)) * params.α
#= none:13 =#
ˍ₋out
end
end
end
end)
g = RuntimeGeneratedFunctions.RuntimeGeneratedFunction{(:ˍ₋out, :t, :X, :Λ, :V, :params), EulerLagrange.var"#_RGF_ModTag", EulerLagrange.var"#_RGF_ModTag", (0x4df42040, 0x8fc17548, 0x11ffb7f0, 0xb9555f81, 0xc483cd8e), Expr}(quote
#= none:1 =#
#= none:2 =#
#= none:2 =# @inbounds begin
#= none:4 =#
begin
#= none:8 =#
#= none:8 =# @inbounds begin
#= none:10 =#
ˍ₋out[1] = (Differential(t, 1))(getindex(v(t), 1))
#= none:11 =#
ˍ₋out[2] = (Differential(t, 1))(getindex(v(t), 2))
#= none:13 =#
ˍ₋out
end
end
end
end)
Lagrangian: L = RuntimeGeneratedFunctions.RuntimeGeneratedFunction{(:t, :X, :V, :params), EulerLagrange.var"#_RGF_ModTag", EulerLagrange.var"#_RGF_ModTag", (0xc8850283, 0x788ffa0a, 0x572fdb74, 0x9530fb48, 0xc012cda0), Expr}(quote
#= none:1 =#
#= none:2 =#
#= none:2 =# @inbounds begin
#= none:4 =#
begin
#= none:8 =#
(getindex(V, 1) ^ 2 + getindex(V, 2) ^ 2) / 2 + (-1 // 2 * (getindex(X, 1) ^ 2 + getindex(X, 2) ^ 2)) * params.α
end
end
end)
Invariants:
GeometricBase.NullInvariants()
Timespan: (0.0, 10.0)
Timestep: 0.01
Initial conditions:
(t = fill(0.0), q = [1.0, 1.0], p = [0.5, 2.0], v = [0.0, 0.0])
Parameters:
(α = 5.0,)This problem can again be integrated using GeometricIntegrators:
sol = integrate(lprob, Gauss(1))
fig = lines(parent(sol.q[:,1]), parent(sol.q[:,2]);
axis = (; xlabel = "x₁", ylabel = "x₂", title = "Particle moving in a square potential"),
figure = (; size = (800,600), fontsize = 22))┌ Warning: Hermite Extrapolation: q's history[1] and history[2] are identical!
└ @ GeometricIntegratorsBase ~/.julia/packages/GeometricIntegratorsBase/zEYTS/src/extrapolation/hermite.jl:209