U.U.D.M. Project Report 2021:36
The Hamiltonian formulation of geodesics
Victor Hildebrandsson
Examensarbete i matematik, 15 hp Handledare: Georgios Dimitroglou Rizell Examinator: Martin Herschend Juli 2021
Department of Mathematics Uppsala University
Abstract We explore the Hamiltonian formulation of the geodesic equation. We start with the definition of a differentiable manifold. Then we continue with studying tensors and differential forms. We define a Riemannian manifold (M, g) and geodesics, the shortest path between two points, on such manifolds. Lastly, we define the symplectic manifold (T ∗M, d(pdq)), and study the connection between geodesics and the flow of Hamilton’s equations.
2 Contents
1 Introduction 4
2 Background 5 2.1 Calculus of variations ...... 5 2.2 Differentiable manifolds ...... 8 2.3 Tensors and vector bundles ...... 10
3 Differential forms 13 3.1 Exterior and differential forms ...... 13 3.2 Integral and exterior derivative of differential forms ...... 14
4 Riemannian manifolds 17
5 Symplectic manifolds 20
References 25
3 1 Introduction
In this thesis, we will study Riemannian manifolds and symplectic manifolds. More specifically, we will study geodesics, and their connection to the Hamilto- nian flow on the symplectic cotangent bundle. A manifold is a topological space which locally looks like Euclidean space. The study of Riemannian manifolds is called Riemannian geometry, the study of symplectic manifolds is called symplectic geometry. Riemannian geometry was first studied by Bernhard Riemann in his habil- itation address ”Uber¨ die Hypothesen, welche der Geometrie zugrunde liegen” (”On the Hypotheses on which Geometry is Based”). In particular, it was the first time a mathematician discussed the concept of a differentiable manifold [1]. Symplectic geometry has its roots in Hamiltonian formulation of classical mechanics, where phase spaces take the form of symplectic manifolds. We will begin the thesis with discussing the calculus of variations, where we will get an understanding of concepts used in classical mechanics. Then we continue to define differentiable manifolds. We also define the tangent space and tangent bundle of a differentiable manifold. Moving on, we will discuss tensors, vector bundles, and differential forms. Tensors and differential forms are important tools when studying differential geometry, and the vector bundle can be considered as a family of vector spaces parametrized by a manifold. We then define a Riemannian manifold and derive the geodesic equation on such manifolds. Lastly, we define a symplectic manifold and derive a natural symplectic structure for the cotangent bundle. On the cotangent bundle we can define a vector field, the Hamiltonian vector field, and also define the cogeodesic flow: Theorem. Let (M, g) be a Riemannian manifold. Let (q, p) be local coordi- ∗ 1 ij nates of T M and H(q, p) = 2 g (q)pipj be our Hamiltonian. Then the geodesic equation is equivalent to the cogeodesic flow on T ∗M, given by:
i ∂H ij q˙ = = g pj, ∂pi ∂H 1 p˙ = − = − gjkp p . i ∂qi 2 ,i j k
Example. On the flat torus Tn = (S1)n we have the cotangent bundle
∗ n 1 n n T T = (S ) × R .
∗ n The local coordinates on T T are (θ1, . . . , θn, p1, . . . , pn), where (θ1, . . . , θn) are 1 n standard angular coordinates of (S ) , and (p1, . . . , pn) the standard coordinates of Rn. The cogeodesic flow is then given by ˙i θ = pi,
p˙i = 0.
4 2 Background
In this chapter, we will explain some concepts that will be crucial for our un- derstanding of the content in the later chapters. Each concept will have a separate subsection. However, before we can do that, we need to refresh (and in some cases, introduce) definitions, notations, and terminology that will be used throughout this thesis.
With h·, ·i we will, if not otherwise stated, mean the Euclidean inner product. If we say that a function f is smooth, we mean that the function is continuous and all its partial derivatives exists and are continuous. Definition 2.1 (Einstein summation convention). If a free index appears twice in a single term, once elevated, and once lowered, we sum over the values of that index. The easiest way to understand Einstein summation is with an example:
Example 2.1. Let a = (a1, . . . , an) and b = (b1, . . . , bn). Let n X y = aibi. i=1 With Einstein summation, we instead notate the vector b = (b1, . . . , bn), and then we write i y = aib . Remark. In some instances below, we will use Einstein summation over partial ∂ derivatives ∂xi . Then, an elevated index in the ”demoninator” is understood as a lowered index.
Definition 2.2 (Disjoint union). Let {Ai : i ∈ I} be a family of sets indexed by I. The disjoint union of this family is the set G [ Ai = {(i, x): x ∈ Ai}. i∈I i∈I F If Ai = A for all i ∈ I, then i∈I Ai = I × A.
2.1 Calculus of variations In the calculus of variations we consider the extremals of functionals, contin- uous functions from the space C1(U) to the real numbers. Here U ⊂ Rn is compact, and C1 has norm
0 |f|C1 = sup |f(x)| + sup |f (x)|. x∈U x∈U
2 In euclidean plane R , let γ = {(t, x): x = x(t), t0 ≤ t ≤ t1} be a curve. 1 Consider the approximation γh = {(t, x): x = x(t) + h(t), h ∈ C (U)} of γ. Let Φ be a functional, consider the increment Φ(γh) − Φ(γ).
5 Definition 2.3 (Differentiable). A functional Φ is differentiable if
Φ(γh) − Φ(γ) = F (h) + R(γ, h),
1 where F : C (U) −→ R is linear and continuous, and R(γ, h) = O(|h|C1 ). F (h) is then called the differential.
Theorem 2.1. If the function L : R × R × R −→ R is differentiable, then the functional Φ(γ) = R t1 L(x(t), x˙(t), t)dt is differentiable, with differential t0
Z t1 t1 ∂L d ∂L ∂L F (h) = − hdt + h ∂x dt ∂x˙ ∂x˙ t0 t0 at any x ∈ C2(U) ⊆ C1(U). Proof. See p. 56-57 in [2]. Definition 2.4 (Extremal). An extremal of a differentiable functional Φ(γ) is a curve γ such that F (h) = 0 for all h that vanish along the boundary. Remark. Compare the definition of extremals to the definition of extreme points of a function f(x).
2 Theorem 2.2. The curve γ = {(t, x): x = x(t) ∈ C (R), t0 ≤ t ≤ t1} is an extremal of the functional Φ(γ) = R t1 L(x(t), x˙(t), t)dt if and only if t0 ∂L d ∂L − = 0 (1) ∂x dt ∂x˙ along the curve x. Proof. See p. 57-58 in [2]. Equation (1) is called the Euler-Lagrange equation, or Lagrange’s equa- tion. It can be generalized to higher dimensions. In euclidean space Rn, let ∂ ∂ ∂ x = (x1, . . . , xn) and = ,..., . ∂x ∂x1 ∂xn
Theorem 2.3. The curve γ = {(t, x): x = x(t), t0 ≤ t ≤ t1} is an extremal of the functional Φ(γ) = R t1 L(x(t), x˙(t), t)dt if and only if t0 ∂L d ∂L − = 0, i = 1, . . . , n ∂xi dt ∂x˙ i along the curve x. The condition of γ being an extremal is independent of coordinate system n on R . Therefore we will use generalized coordinates q = (q1, . . . , qn). From the Euler-Lagrange equations, we get n second-order differential equa- tions. We would like to reduce them to first-order equations instead. To do this, we first need to define the Legendre transformation.
6 Definition 2.5 (Legendre transformation). Let f : Rn −→ R be a convex T function, i.e. x Hf x is positive definite, where Hf is the hessian matrix of f. The Legendre transformation is the function
g(p) = hp, x(p)i − f(x(p)) = maxx[hp, xi − f(x)] ∂f where p = ∂x . Remark. The Legendre transformation is involutive, i.e. if the Legendre trans- form takes f to g, then the Legendre transform takes g to f.
n 2 2 Example 2.2. Let f : R −→ R,(x1, . . . , xn) 7−→ x1 + . . . xn be our function. T Then Hf = 2I, so x Hf x is positive definite, and f is convex. Then ∂f p p p(x) = = (2x ,..., 2x ) ⇐⇒ x(p) = 1 ,..., n ∂x 1 n 2 2 and g(p) = hp, x(p)i − f(x(p)) p2 p2 p2 p2 = 1 + ··· + n − 1 + ··· + n 2 2 4 4 p2 p2 = 1 + ··· + n 4 4 is the Legendre transformation of f.
Given a Lagrangian L : Rn × Rn × R −→ R, assumed to be convex with ∂L ∂L respect toq ˙, consider the equationsp ˙ = ∂q , where p = ∂q˙ . Theorem 2.4. The system of n second-order Euler-Lagrange equations is equiv- alent to the system of 2n first-order equations ∂H p˙ = − , i ∂q i (2) ∂H q˙i = ∂pi where H(p, q, t) = hp, q˙(p)i − L(q, q˙(p), t) is the Legendre transformation of the Lagrangian viewed as a function of q˙. Proof. By definition, the Legendre transform of L(q, q,˙ t), with respect toq ˙, is ∂L the function H(p, q, t) = hp, q˙i − L(q, q,˙ t), whereq ˙ =q ˙(p) and p = ∂q˙ . The total differential of H = H(p, q, t) ∂H ∂H ∂H dH = dp + dq + ∂p ∂q ∂t is equal to the total differential of H = hp, q˙i − L(q, q,˙ t) ∂L ∂L dH =qdp ˙ − dq − . ∂q ∂t
7 The expressions of dH are equal if and only if ∂H =q, ˙ ∂p ∂H ∂L = − , ∂q ∂q ∂H ∂L = − ∂t ∂t
∂L Plugging inp ˙ = ∂q , we get the equations (2). The equations (2) are called Hamilton’s equations. Later in the thesis, we will come back to these equations as they are strongly linked to the study of symplectic manifolds.
2.2 Differentiable manifolds A manifold is a topological space which locally looks like Euclidean space. To study these manifolds, we often need to use calculus. Then we need our mani- folds to be smooth, or differentiable [3]. Definition 2.6 (Topological manifold). Let M be a second-countable Hausdorff space such that for every point p ∈ M there is a neighbourhood U of p that is homeomorphic to an open set V of Rn. Then M is a n-dimensional mani- fold. Such homeomorphism, φ : U −→ V , is called a (coordinate) chart.
Definition 2.7 (Differentiable manifold). An atlas is a family {Uα, φα} of S charts such that Uα is a covering of M. A chart is compatible with an atlas if their union is again an atlas. An atlas is called maximal if any chart compatible with it is already contained in it. An atlas is called differentiable if all chart transitions −1 φβ ◦ φα : φα(Uα ∩ Uβ) −→ φβ(Uα ∩ Uβ) are smooth. A maximal differentiable atlas is called a differentiable struc- ture, and a differentiable manifold of dimension n is a n-dimensional manifold with a differantiable structure. At a point x in a manifold M we have a tangent space, which is a real vector space. Intuitively, the tangent space is the span of the directional derivatives at the point x. We can ”glue together” all tangent spaces of a manifold to get its tangent bundle. The tangent bundle is itself a manifold of twice the dimension of the original manifold.
We start by defining the tangent space and tangent bundle of an open set of Rn. Let x1, . . . , xn be euclidean coordinates of Rn,U ⊆ Rn open, p ∈ U. The ∂ ∂ tangent space of U at p is TpU := {p} × E, where E = span ,..., . ∂x1 ∂xn Here, ∂ is the i-th partial derivative at p. ∂xi
8 Let U ⊆ Rn,U 0 ⊆ Rm be open, f : U −→ U 0 be differentiable, and p ∈ U. Define the derivative dfp as
0 dfp : TpU −→ Tf(p)U , j i ∂ i ∂f ∂ v i 7−→ v i (p) . ∂x ∂x ∂fj
S ∼ n n n Let TU := p∈U TpU = U × E = U × R . Hence, TU ⊆ R × R is open, and in particular a differentiable manifold. Let π : TU −→ U be the projection onto the first factor. The triple (T U, π, U) is the tangent bundle of U. Similar to the tangent space, we define
df : TU −→ TU 0 j i ∂ i ∂f ∂ x, v i 7−→ f(x), v i (x) ∂x ∂x ∂fj Now we can define the tangent space and the tangent bundle of a general differentiable manifold M. Definition 2.8 (Tangent space). Let p ∈ M. On
{(φ, v): φ : U −→ V chart with p ∈ U, v ∈ Tφ(p)V } define the equivalence relation (φ, v) ∼ (ψ, w) ⇐⇒ w = d(ψ ◦φ−1)v. The space of equivalence classes is called the tangent space to M at p, denoted TpM.
Given a map f : M −→ N between manifolds M, N , let (UM, φ), (UN , ψ) be charts for p ∈ M and f(p) ∈ N , respectively. We then get the map ˜ f : VM −→ VN f˜ = ψ ◦ f ◦ φ−1
m n between open sets VM = φ(UM) ⊆ R , VN = ψ(UN ) ⊆ R . Represented in local coordinates by the map df˜, df is the unique map such that the following diagram commutes:
df T M T N dφ dψ df˜ TVM TVN
Definition 2.9 (Tangent bundle). T M is the disjoint union of the tangent spaces TpM, p ∈ M. Define the projection π : T M −→ M by π(p, w) = p for w ∈ TpM. (T M, π, M) is called the tangent bundle of M, T M is called the total space of the tangent bundle. Usually, we use the total space T M to denote the tangent bundle.
9 If (Uα, φ) is a chart for an n-dimensional differentiable manifold M we let TU = F T M, and define p∈Uα p
dφ : TUα −→ TV,
F n n where TV = p∈V TpR = V × R , as the charts for T M. Then [ [ G G TUα = TpM = TpM = T M
α α p∈Uα p∈M
−1 −1 and dφα ◦ dφβ = d(φα ◦ φβ ) are smooth. So T M is an 2n-dimensional differ- entiable manifold, in particular, it can be checked that T M is Haussdorf and second countable. Now we can find extremals of functionals on a differentiable manifold M. Consider a smooth function L : T M −→ R, and a map γ : R −→ M.
Theorem 2.5. In local coordinates (q1, . . . , qn) of a point γ(t), γ is an extremal of the functional Φ(γ) = R t1 L(q, q˙)dt, if it satisfies the Euler-Lagrange equation t0 with Lagrangian L(q, q˙), where (q, q˙) are local coordinates of T M.
Proof. It suffices to consider variations supported in some local chart, then use the results from Section 2.1.
2.3 Tensors and vector bundles A generalization of the tangent bundle is that of a vector bundle. Given an abstract vector bundle, there are induced vector bundles of tensors, so called tensor bundles. Sections of such bundles are called tensor fields. Both differen- tial forms and Riemannian metrics are examples of tensor fields. [4].
Definition 2.10 (Vector bundle). A smooth k-dimensional vector bundle is a triple (E, π, M) of smooth manifolds E (the total space) and M (the base space), and a surjective map π : E −→ M (the projection), satisfying
−1 (a) Each set Ep := π (p) is a vector space. (b) For each p ∈ M, there is a neighbourhood U of p and a diffeomorphism φ : π−1(U) −→ U × Rk such that
π = π1 ◦ φ
where π1 is the projection onto the first factor.
k (c) The restriction of φ to φ : Ep −→ {p} × R is a linear isomorphism.
Ep is called the fiber of E over p. The diffeomorphism φ is called a local trivialization of E.
10 Proposition 2.6. Let M be a smooth manifold, E a set, and π : E −→ M a surjective map. Suppose we are given an open covering {Uα} of M together with bijective maps −1 k φα : π (Uα) −→ Uα×R satisfying π1◦φ = π, such that whenever Uα∩Uβ 6= ∅, the composite map
−1 k k φα ◦ φβ : Uα ∩ Uβ × R −→ Uα ∩ Uβ × R is on the form −1 φα ◦ φβ (p, V ) = (p, τ(p)V ) for some smooth map τ : Uα ∩Uβ −→ GL(k, R). Then E has a unique structure as a smooth k-dimensional vector bundle over M for which the maps φα are local trivializations.
The smooth, GL(k, R) valued maps τ in the proposition are called transi- tion functions for E. In the following example we find the transition functions for T M. Example 2.3. Given an open set U of M with local coordinates x = (x1, . . . , xn), any tangent vector V ∈ TpM at a point p ∈ U can be written in terms of the i ∂ coordinate basis as V = v ∂xi . Define a bijection
−1 n φ : π (U) −→ U × R V 7−→ (p, v).
Where two sets of coordinates, x andx ˜, overlap, the coordinate bases are related by ∂ ∂x˜j ∂ = , ∂xi ∂xi ∂x˜j and then the same vector V is represented by
∂ ∂ ∂x˜j ∂ V =v ˜j = vi = vi . ∂x˜j ∂xi ∂xi ∂x˜j
j i ∂x˜j Hencev ˜ = v ∂xi and thus
φ˜ ◦ φ−1(p, v) = φ˜(V ) = (p, v˜) = (p, τ(p)v)