betaFeedback

WorksPhysics and Astronomy

Hamiltonian dynamics with nonconservative forces on extended phase space: An analytical mechanics derivation

Abstract

This note derives the full Hamiltonian equations of motion on extended phase space for nonconservative and/or nonautonomous dynamical systems, with particular attention given to the momentum coordinate and generalized force associated with time. Time reparameterization is also considered, clarifying when it does or does not preserve Hamiltonian structure. The analytical mechanics framework is used throughout, with all developments formulated in terms of arbitrary symplectic coordinates; the main developments are not predicated on a Lagrangian description.

Introduction

Although nonconservative forces and extended phase space are, individually, established topics in Hamiltonian dynamics, their combination is seldom developed explicitly. Nonconservative forces are routinely incorporated as generalized forces through virtual work and the Hamilton–d’Alembert principle, as treated in classical texts such as [1, 2, 3, 4]. Independently, the (symplectic) extended phase space formulation, in which time is appended as a generalized coordinate paired with its conjugate momentum coordinate, is developed notably by Lanczos [1], Greenwood [5], Johns [6], and Struckmeier [7, 8]. Although some sources address both nonconservative forces and extended Hamiltonian dynamics (e.g., [1, 5, 6]), their combination within the extended Hamiltonian framework does not appear to be developed, leaving the two subjects largely separate. We aim to fill that gap.

Analytical mechanics precedents to this work’s Hamiltonian construction are found mainly on the Lagrangian side. Sharma et al. [9] formulate an extended Lagrange-d’Alembert principle for time-dependent systems with nonconservative forces and account for virtual work associated with variations in time. Johns [6] likewise develops extended Lagrangian equations with nonconservative forces, including the additional generalized force and virtual displacement associated with the time coordinate. Yet, such developments have not been carried over to the Hamiltonian description in arbitrary symplectic coordinates.

Why extended phase space? While the consideration of nonconservative forces requires little explanation, the role of extended phase space is less immediately apparent and warrants brief motivation. Two common situations motivate the use of extended phase space:

  1. Nonautonomous systems. Nonautonomous Hamiltonian systems arise whenever the Hamiltonian function itself depends explicitly on tt (time1), as occurs throughout dynamical systems and optimal control problems. A standard “trick” to render such a system as autonomous is to simply append tt as an additional

dependent state variable, with governing equation dtdt=1\frac{\mathrm{d}t}{\mathrm{d}t}=1, and thereby conveniently circumvent the additional complications associated with explicit tt-dependence. Yet, this typically produces a state space of odd dimension 2n+12n+1 and therefore does not permit a canonical Hamiltonian formulation. Although such a (2n+1)(2n+1)-dim formulation may be treated elegantly using cosymplectic structures or related presymplectic formulations [10, 11, 12], doing so requires familiarity with concepts and notation specialized to the differential geometric formalism (avoided in this work). An alternative presentation, accessible within the analytical mechanics framework, is to introduce an additional momentum coordinate, πt\pi_t, conjugate to tt. The resulting (2n+2)(2n+2)-dim extended phase space again admits the familiar Hamiltonian formulation [1, 6, 7].

  1. Evolution parameter transformations. Extended phase space also arises naturally whenever one wishes to replace some Hamiltonian system’s evolution parameter, tt, by some other parameter, ε\varepsilon, prescribed via dt=ldε\mathrm{d}t=l\mathrm{d}\varepsilon for some function l>0l>0 (a conformal scaling factor). At the level of first-order ODEs, such transformations can be easily implemented by directly multiplying the original ODEs by dtdε=l\frac{\mathrm{d}t}{\mathrm{d}\varepsilon}=l. Unfortunately, for a general phase-space-dependent ll, such direct conformal scaling does not preserve the Hamiltonian structure of the original system (apart from certain special cases) [13]. Fortunately, passing to extended phase space provides a systematic implementation of the same evolution parameter transformation while retaining the canonical Hamiltonian form of the original system, albeit on a higher-dimensional phase space [14, 15, 7]. Such transformations of the evolution parameter in Hamiltonian systems arise prominently in regularization and linearization methods in celestial mechanics [16, 17, 18].

The above scenarios are not mutually exclusive and the extended phase space formulation accommodates either or both as standard, not exceptional. Conversely, neither scenario is required for its use.

This work. We develop the extended phase space construction of Hamiltonian dynamics in the presence of general nonconservative forces using the framework of classic analytical mechanics. Developments are formulated in arbitrary symplectic coordinates (x,π)(\mathbf{x},\boldsymbol{\pi}), which need not be configuration and momentum coordinates in the physical sense. Three points receive particular attention. (1) The complete forced extended equations of motion are derived, including the generalized force associated with tt. (2) The often-seen relation πt=−H\pi_t=-\mathcal{H} is clarified: it characterizes an invariant level set of the extended Hamiltonian which selects the family of extended solution curves that reproduce the original physical dynamics. (3) The Hamiltonian structure (or lack thereof) of the original, directly reparameterized, fully extended, and restricted extended dynamics is compared. If the nonconservative forces are omitted, the same developments give a concise account of the extended phase space construction of classical conservative Hamiltonian systems.

Organization. We first review nonconservative Hamiltonian dynamics and its Hamilton-d’Alembert derivation on ordinary phase space (§ 2). Transforming the evolution parameter by direct conformal scaling is then considered to motivate the extended construction (§ 3). The full nonconservative dynamics are subsequently derived on extended phase space (§ 4.1) and it is shown that the extended Hamiltonian is always conserved (§ 4.2). The physical significance of the extended Hamiltonian’s zero level set is discussed and the relation πt=−H\pi_t=-\mathcal{H} is clarified (§ 4.3). Finally, we summarize how the Hamiltonian structure of these various presentations of the dynamics compare (§ 4.4). In Appx. A.1, the extended Lagrangian and its Legendre transformation reaffirm the interpretation of πt\pi_t and its relation with the Hamiltonian.

Notation and conventions. This work is written in the coordinate-dominant manner of classic analytical dynamics with all developments formulated in terms of local coordinates or functions thereof. Our mathematical arena is only ever regarded as some finite-dimensional real coordinate vector space, RN\mathbb{R}^N, equipped with the usual structures, operations, and abuses of notation.

  • Expressions such as u∈RN\mathbf{u}\in\mathbb{R}^N are often an abuse of notation indicating that u\mathbf{u} is an object that takes values in RN\mathbb{R}^N. For any such u\mathbf{u}, we make no distinction between u\mathbf{u} as an ordered NN-tuple, (u1,…,uN)(u_1,\ldots,u_N), and u\mathbf{u} as a column vector, [u1…uN]T[u_1\ldots u_N]^{\mathsf{T}}.

  • The “dot product” denotes contraction in the simplest sense. For u,v∈RN\mathbf{u},\mathbf{v}\in\mathbb{R}^N, then u⋅v=uivi\mathbf{u}\cdot\mathbf{v}=u_i v_i (summation implied over i=1…Ni=1\ldots N). Contraction with some A∈RN×NA\in\mathbb{R}^{N\times N} is denoted either with or without a dot. That is, Au=A⋅u=Aijuji^iA\mathbf{u}=A\cdot\mathbf{u}=A_{ij}\mathbf{u}_j\hat{\mathbf{i}}_i and ATu=AT⋅u=u⋅A=Aijuii^j=Ajiuji^iA^{\mathsf{T}}\mathbf{u}=A^{\mathsf{T}}\cdot\mathbf{u}=\mathbf{u}\cdot A=A_{ij}u_i\hat{\mathbf{i}}_j=A_{ji}u_j\hat{\mathbf{i}}_i (with i^i\hat{\mathbf{i}}_i the “standard basis” for RN\mathbb{R}^N).

  • We abuse notation y=y(x)y=y(x), or just y(x)y(x), to indicate something called yy is a “function of” something called xx. Similarly, given some implied function transformation x↔x~x\leftrightarrow\tilde{x}, then y(x)y(x) and y(x~)y(\tilde{x}) should be interpreted, respectively, as “yy expressed as a function of xx” and “yy expressed as a function of x~\tilde{x}”.

  • ∂xy≡∂y∂x\partial_x y\equiv\frac{\partial y}{\partial x} are equivalent notations for the partial derivative of some xx-dependent yy with respect to xx. If y∈RMy\in\mathbb{R}^M and

x∈RN\mathbf{x} \in\mathbb{R}^{N}, then ∂xy≡∂y∂x∈RM×N\partial_{\mathbf{x}}\mathbf{y} \equiv\frac{\partial\mathbf{y}}{\partial\mathbf{x}} \in\mathbb{R}^{M \times N} has components (∂xy)aj=∂ya∂xj(\partial_{\mathbf{x}}\mathbf{y})_{aj} = \frac{\partial y_a}{\partial x_j} (for a=1…Ma = 1 \ldots M and j=1…Nj = 1 \ldots N). For any z∈RM\mathbf{z} \in\mathbb{R}^{M}:

z⋅∂xy≡∂yT∂x⋅z(1)\mathbf{z} \cdot\partial_{\mathbf{x}}\mathbf{y} \equiv\frac{\partial\mathbf{y}^{\mathsf{T}}}{\partial\mathbf{x}} \cdot\mathbf{z} \tag*{(1)}

More specific notation used in this work includes:

  • r=(r1,…,rN)∈RN\mathbf{r} = (r_1, \ldots, r_N) \in\mathbb{R}^{N} denote cartesian position coordinates associated with an orthonormal inertial basis for NN-dim Euclidean vector space, and Fnc∈RN\mathbf{F}^{\mathrm{nc}} \in\mathbb{R}^{N} denotes components of any/all nonconservative forces in that same basis.

  • (x,π)∈R2n(\mathbf{x}, \boldsymbol{\pi}) \in\mathbb{R}^{2n} denote arbitrary symplectic coordinates (i.e., canonical coordinates), where n≤Nn \leq N is the degrees of freedom (DOF) of the system. By arbitrary symplectic coordinates we mean that (x,π)(\mathbf{x}, \boldsymbol{\pi}) satisfy Hamilton’s equations of motion in their canonical form, but that x\mathbf{x} and π\boldsymbol{\pi} need not be configuration and momentum coordinates, respectively, in the physical sense.4 That is, we allow that r=r(x,π,t)\mathbf{r} = \mathbf{r}(\mathbf{x}, \boldsymbol{\pi}, t) may be a “function of” all 2n2n coordinates (x,π)(\mathbf{x}, \boldsymbol{\pi}), and possibly also explicitly tt-dependent.

  • (q,p)∈R2n(\mathbf{q}, \mathbf{p}) \in\mathbb{R}^{2n} denote symplectic coordinates where q\mathbf{q} and p\mathbf{p} are configuration-level and momentum-level coordinates, respectively, in the typical physical sense such that r=r(q,t)\mathbf{r} = \mathbf{r}(\mathbf{q}, t) does not depend on momentum coordinates p\mathbf{p}.

  • {⋅,⋅}\{\cdot,\cdot\} denotes the canonical symplectic Poisson bracket, expressed in arbitrary symplectic coordinates (x,π)(\mathbf{x}, \boldsymbol{\pi}) as follows for any scalar-valued functions ff and hh:

{f,h}=∂xf⋅∂πh−∂πf⋅∂xh,{f,h}=−{h,f}(2)\{f,h\} = \partial_{\mathbf{x}}f \cdot\partial_{\boldsymbol{\pi}}h - \partial_{\boldsymbol{\pi}}f \cdot\partial_{\mathbf{x}}h,\qquad\{f,h\} = -\{h,f\} \tag*{(2)}
  • □˙:=d□dt\dot{\square} := \frac{\mathrm{d}\square}{\mathrm{d}t} and □˚:=d□dε\mathring{\square} := \frac{\mathrm{d}\square}{\mathrm{d}\varepsilon} denote, respectively, differentiation with respect to evolution parameters tt and ε\varepsilon.

  • Acronyms. DOF = degree(s) of freedom; ODE = ordinary differential equation(s); EoM = equation(s) of motion.

Review of classical Hamiltonian dynamics with nonconservative forces

Before considering extended phase space, we first summarize the canonical Hamiltonian equations of motion for systems subject to nonconservative forces. We further allow for the system to be nonautonomous.

The equations of motion. Consider some nn-DOF system described in arbitrary symplectic coordinates, (x,π)∈R2n(\mathbf{x}, \boldsymbol{\pi}) \in\mathbb{R}^{2n}, by some Hamiltonian function H(x,π,t)\mathcal{H}(\mathbf{x}, \boldsymbol{\pi}, t). If nonconservative forces also act on the system then Hamilton’s equations of motion incorporating such forces are stated as follows (derived below):

\begin{aligned} \dot{\mathbf{x}} &= \partial_{\boldsymbol{\pi}}\mathcal{H} - \boldsymbol{\alpha}_{\boldsymbol{\pi}} \\ \dot{\boldsymbol{\pi}} &= -\partial_{\mathbf{x}}\mathcal{H} + \boldsymbol{\alpha}_{\mathbf{x}}, \end{aligned} \qquad \boldsymbol{\alpha}_{\mathbf{x}} := \mathbf{F}^{\mathrm{nc}} \cdot\partial_{\mathbf{x}}\mathbf{r},\qquad \boldsymbol{\alpha}_{\boldsymbol{\pi}} := \mathbf{F}^{\mathrm{nc}} \cdot\partial_{\boldsymbol{\pi}}\mathbf{r} \tag*{(3)}\end{aligned}

where □˙:=d□dt\dot{\square} := \frac{\mathrm{d}\square}{\mathrm{d}t} and where r,Fnc∈RN\mathbf{r}, \mathbf{F}^{\mathrm{nc}} \in\mathbb{R}^{N} are cartesian coordinate vectors of position and force as previously described. It follows from Eq. (3) that the equation governing the evolution of any function f(x,π,t)f(\mathbf{x}, \boldsymbol{\pi}, t) is then

f˙={f,H}+∂tf+αx⋅∂πf−απ⋅∂xf={f,H}+∂tf+Fnc⋅{r,f}(4)\dot{f} = \{f,\mathcal{H}\} + \partial_t f + \boldsymbol{\alpha}_{\mathbf{x}} \cdot\partial_{\boldsymbol{\pi}}f - \boldsymbol{\alpha}_{\boldsymbol{\pi}} \cdot\partial_{\mathbf{x}}f = \{f,\mathcal{H}\} + \partial_t f + \mathbf{F}^{\mathrm{nc}} \cdot\{\mathbf{r},f\} \tag*{(4)}

from which we see that H\mathcal{H} itself is generally not conserved by the dynamics in Eq. (3):

H˙=∂tH+αx⋅∂πH−απ⋅∂xH=∂tH+Fnc⋅{r,H}(5)\dot{\mathcal{H}} = \partial_t\mathcal{H} + \boldsymbol{\alpha}_{\mathbf{x}} \cdot\partial_{\boldsymbol{\pi}}\mathcal{H} - \boldsymbol{\alpha}_{\boldsymbol{\pi}} \cdot\partial_{\mathbf{x}}\mathcal{H} = \partial_t\mathcal{H} + \mathbf{F}^{\mathrm{nc}} \cdot\{\mathbf{r},\mathcal{H}\} \tag*{(5)}

▪ Note. If (q,p)∈R2n(\mathbf{q}, \mathbf{p}) \in\mathbb{R}^{2n} are typical configuration-level and momentum-level coordinates in the physical sense such that r=r(q,t)\mathbf{r} = \mathbf{r}(\mathbf{q}, t), then απ=0\boldsymbol{\alpha}_{\boldsymbol{\pi}} = 0 and the ODEs in Eq. (3)–Eq. (4) simplify as follows:

q˙=∂pH,p˙=−∂qH+αq,f˙={f,H}+∂tf+αq⋅∂pf(6)\dot{\mathbf{q}} = \partial_{\mathbf{p}}\mathcal{H},\qquad \dot{\mathbf{p}} = -\partial_{\mathbf{q}}\mathcal{H} + \boldsymbol{\alpha}_{\mathbf{q}},\qquad \dot{f} = \{f,\mathcal{H}\} + \partial_t f + \boldsymbol{\alpha}_{\mathbf{q}} \cdot\partial_{\mathbf{p}}f \tag*{(6)}

with H(q,p,t)\mathcal{H}(\mathbf{q}, \mathbf{p}, t) being the relevant Hamiltonian with governing equation H˙=∂tH+αq⋅∂pH\dot{\mathcal{H}} = \partial_t\mathcal{H} + \boldsymbol{\alpha}_{\mathbf{q}} \cdot\partial_{\mathbf{p}}\mathcal{H}.

Forced Hamiltonian dynamics in the form of Eq. (6) are standard [3, 6, 19]. The more general form given in Eq. (3) is less common, though Tyeter’s derivations are found in [20, 21]. We nevertheless provide our own derivation of Eq. (3) in order to establish a reference point for subsequent developments in § 4.

*Derivation of Eq. (3). *Classically, the equations of motion may be derived using Hamilton’s principle; all admissible paths between given endpoints, the physical trajectories are those that render the action integral stationary:

δI=δ∫t0tf(π⋅x˙−H) dt=0.\delta I = \delta\int_{t_0}^{t_f}(\boldsymbol{\pi}\cdot\dot{\mathbf{x}}-\mathcal{H})\,\mathrm{d}t = 0.

(conservative) {#eq:1.5}

The above assumes purely conservative dynamics. To account for nonconservative forces (with cartesian components Fnc\mathbf{F}^{\mathrm{nc}}), we add a virtual work term, δWnc\delta W^{\mathrm{nc}}, such that the above is replaced by:

δI+∫t0tfδWncdt=0,δWnc=Fnc⋅δr(7)\delta I + \int_{t_0}^{t_f}\delta W^{\mathrm{nc}}\mathrm{d}t = 0,\qquad\delta W^{\mathrm{nc}} = \mathbf{F}^{\mathrm{nc}}\cdot\delta\mathbf{r} \tag*{(7)}

where, despite our notation, δWnc\delta W^{\mathrm{nc}} is not the variation of a smooth function called WncW^{\mathrm{nc}}. Rather, δWnc\delta W^{\mathrm{nc}} is merely an abuse of notation for the virtual work, expressed more properly in cartesian coordinates as above (see footnote5).

Let us start with the variation of action integral, δI\delta I, in Eq. (7). Carrying out the variation leads to:

δI=δ∫totf(π⋅x˙−H) dt=(π⋅x˙−H)δt∣totf+∫totfδ(π⋅x˙−H) dt=∫totf(π⋅δx˙+x˙⋅δπ−δH) dt=π⋅δx∣totf+∫totf(−π˙⋅δx+x˙⋅δπ−δH) dt=∫totf[−(π˙+∂xH)⋅δx+(x˙−∂πH)⋅δπ] dt.(8)\begin{aligned} \delta I &= \delta\int_{t_o}^{t_f}(\boldsymbol{\pi}\cdot\dot{\mathbf{x}}-\mathcal{H})\,dt \\ &=(\boldsymbol{\pi}\cdot\dot{\mathbf{x}}-\mathcal{H})\delta t\big|_{t_o}^{t_f}+\int_{t_o}^{t_f}\delta(\boldsymbol{\pi}\cdot\dot{\mathbf{x}}-\mathcal{H})\,dt \\ &=\int_{t_o}^{t_f}(\boldsymbol{\pi}\cdot\delta\dot{\mathbf{x}}+\dot{\mathbf{x}}\cdot\delta\boldsymbol{\pi}-\delta\mathcal{H})\,dt \\ &=\boldsymbol{\pi}\cdot\delta\mathbf{x}\big|_{t_o}^{t_f}+\int_{t_o}^{t_f}(-\dot{\boldsymbol{\pi}}\cdot\delta\mathbf{x}+\dot{\mathbf{x}}\cdot\delta\boldsymbol{\pi}-\delta\mathcal{H})\,dt \\ &=\int_{t_o}^{t_f}\left[-(\dot{\boldsymbol{\pi}}+\partial_{\mathbf{x}}\mathcal{H})\cdot\delta\mathbf{x}+(\dot{\mathbf{x}}-\partial_{\boldsymbol{\pi}}\mathcal{H})\cdot\delta\boldsymbol{\pi}\right]\,dt . \tag*{(8)} \end{aligned}

where the above has used the following steps. 1st line: the variation is brought inside the integral using the fact δto=δtf=0\delta t_o=\delta t_f=0. 2nd line: integration by parts is applied to the term with δx˙\delta\dot{\mathbf{x}}, along with the fact that (δx,δπ)(\delta\mathbf{x},\delta\boldsymbol{\pi}) vanish at tot_o and tft_f. 3rd line: simply use δH=∂xH⋅δx+∂πH⋅δπ\delta\mathcal{H}=\partial_{\mathbf{x}}\mathcal{H}\cdot\delta\mathbf{x}+\partial_{\boldsymbol{\pi}}\mathcal{H}\cdot\delta\boldsymbol{\pi}. These steps are common in standard textbooks.

Next, the nonconservative virtual work term in Eq. (7) is given in cartesian coordinates by δWnc=Fnc⋅δr\delta W^{\mathrm{nc}}=\mathbf{F}^{\mathrm{nc}}\cdot\delta\mathbf{r}. We need it expressed in phase space coordinates (x,π)(\mathbf{x},\boldsymbol{\pi}). For the most general case that r=r(x,π,t)\mathbf{r}=\mathbf{r}(\mathbf{x},\boldsymbol{\pi},t):

δWnc=Fnc⋅δr=Fnc⋅(∂xr⋅δx+∂πr⋅δπ)=αx⋅δx+απ⋅δπ,(9)\delta W^{\mathrm{nc}}=\mathbf{F}^{\mathrm{nc}}\cdot\delta\mathbf{r}=\mathbf{F}^{\mathrm{nc}}\cdot\left(\partial_{\mathbf{x}}\mathbf{r}\cdot\delta\mathbf{x}+\partial_{\boldsymbol{\pi}}\mathbf{r}\cdot\delta\boldsymbol{\pi}\right)=\boldsymbol{\alpha}_{\mathbf{x}}\cdot\delta\mathbf{x}+\boldsymbol{\alpha}_{\boldsymbol{\pi}}\cdot\delta\boldsymbol{\pi}, \tag*{(9)}

with αx:=Fnc⋅∂xr\boldsymbol{\alpha}_{\mathbf{x}}:=\mathbf{F}^{\mathrm{nc}}\cdot\partial_{\mathbf{x}}\mathbf{r} and απ:=Fnc⋅∂πr\boldsymbol{\alpha}_{\boldsymbol{\pi}}:=\mathbf{F}^{\mathrm{nc}}\cdot\partial_{\boldsymbol{\pi}}\mathbf{r}. Substitution of Eq. (9) and Eq. (8) into the Hamilton–d’Alembert condition of Eq. (7) finally leads to:

0=δI+∫totfδWnc dt=∫totf[−(π˙+∂xH−αx)⋅δx+(x˙−∂πH+απ)⋅δπ] dt.(10)0=\delta I+\int_{t_o}^{t_f}\delta W^{\mathrm{nc}}\,dt=\int_{t_o}^{t_f}\left[-(\dot{\boldsymbol{\pi}}+\partial_{\mathbf{x}}\mathcal{H}-\boldsymbol{\alpha}_{\mathbf{x}})\cdot\delta\mathbf{x}+(\dot{\mathbf{x}}-\partial_{\boldsymbol{\pi}}\mathcal{H}+\boldsymbol{\alpha}_{\boldsymbol{\pi}})\cdot\delta\boldsymbol{\pi}\right]\,dt . \tag*{(10)}

For the above to hold for any and all (δx,δπ)(\delta\mathbf{x},\delta\boldsymbol{\pi}), it is required that their coefficients vanish, thus verifying the governing equations of motion claimed in Eq. (3). □

Canonical transformation of generalized forces. Consider any, possibly tt-dependent, transformation of arbitrary symplectic coordinates (i.e., a canonical transformation), (x,π,t)↔(x~,π~,t)(\mathbf{x},\boldsymbol{\pi},t)\leftrightarrow(\widetilde{\mathbf{x}},\widetilde{\boldsymbol{\pi}},t). The dynamics for (x~,π~)(\widetilde{\mathbf{x}},\widetilde{\boldsymbol{\pi}}) then, by definition, have the same structure as in Eq. (3):

x~˙=∂π~H~−α~π,π~˙=−∂x~H~+α~x,α~x:=Fnc⋅∂x~r,α~π:=Fnc⋅∂π~r,(11)\dot{\widetilde{\mathbf{x}}}=\partial_{\widetilde{\boldsymbol{\pi}}}\widetilde{\mathcal{H}}-\widetilde{\boldsymbol{\alpha}}_{\boldsymbol{\pi}},\qquad \dot{\widetilde{\boldsymbol{\pi}}}=-\partial_{\widetilde{\mathbf{x}}}\widetilde{\mathcal{H}}+\widetilde{\boldsymbol{\alpha}}_{\mathbf{x}},\qquad \widetilde{\boldsymbol{\alpha}}_{\mathbf{x}}:=\mathbf{F}^{\mathrm{nc}}\cdot\partial_{\widetilde{\mathbf{x}}}\mathbf{r},\qquad \widetilde{\boldsymbol{\alpha}}_{\boldsymbol{\pi}}:=\mathbf{F}^{\mathrm{nc}}\cdot\partial_{\widetilde{\boldsymbol{\pi}}}\mathbf{r}, \tag*{(11)}

with H~(x~,π~,t)\widetilde{\mathcal{H}}(\widetilde{\mathbf{x}},\widetilde{\boldsymbol{\pi}},t) the new Hamiltonian. What then is the relation between (α~x,α~π)(\widetilde{\boldsymbol{\alpha}}_{\mathbf{x}},\widetilde{\boldsymbol{\alpha}}_{\boldsymbol{\pi}}) above and (αx,απ)(\boldsymbol{\alpha}_{\mathbf{x}},\boldsymbol{\alpha}_{\boldsymbol{\pi}}) in Eq. (3)? It follows from their definitions in Eq. (11) and Eq. (3).6

α~x=αx⋅∂x~x+απ⋅∂x~π,α~π=αx⋅∂π~x+απ⋅∂π~π.(12)\widetilde{\boldsymbol{\alpha}}_{\mathbf{x}}=\boldsymbol{\alpha}_{\mathbf{x}}\cdot\partial_{\widetilde{\mathbf{x}}}\mathbf{x}+\boldsymbol{\alpha}_{\boldsymbol{\pi}}\cdot\partial_{\widetilde{\mathbf{x}}}\boldsymbol{\pi},\qquad \widetilde{\boldsymbol{\alpha}}_{\boldsymbol{\pi}}=\boldsymbol{\alpha}_{\mathbf{x}}\cdot\partial_{\widetilde{\boldsymbol{\pi}}}\mathbf{x}+\boldsymbol{\alpha}_{\boldsymbol{\pi}}\cdot\partial_{\widetilde{\boldsymbol{\pi}}}\boldsymbol{\pi}. \tag*{(12)}

The above is unaffected by whether or not the transformation between (x,π)(\mathbf{x},\boldsymbol{\pi}) and (x~,π~)(\widetilde{\mathbf{x}},\widetilde{\boldsymbol{\pi}}) is explicitly tt-dependent. Such tt-dependence matters only for the relation between H\mathcal{H} and H~\widetilde{\mathcal{H}}.

Transforming the evolution parameter via direct conformal scaling

As in §1, we consider a system described in arbitrary symplectic coordinates (x,π)∈R2n(\mathbf{x},\boldsymbol{\pi})\in\mathbb{R}^{2n} by a Hamiltonian H(x,π,t)\mathcal{H}(\mathbf{x},\boldsymbol{\pi},t). The usual tt-parameterized Hamiltonian dynamics (plus nonconservative forces) are given in Eq. (3). Suppose that, instead of time tt, we wish to use some other evolution parameter, ϵ\epsilon, that is related to tt by some given differential relation:

dt=l(x,π,t) dϵi.e.,t˙:=dtdϵ=l(x,π,t),(13)dt=l(\mathbf{x},\boldsymbol{\pi},t)\,d\epsilon \qquad\text{i.e.,}\qquad \dot{t}:=\frac{dt}{d\epsilon}=l(\mathbf{x},\boldsymbol{\pi},t), \tag*{(13)}

where □˙:=d□dϵ\dot{\square}:=\frac{d\square}{d\epsilon} and l≠0l\ne0 may be any smooth, nowhere-vanishing, possibly tt-dependent, function on phase space. Often, one further requires that l>0l>0 be strictly positive everywhere. We will call ll the conformal factor.

If one regards the tt-parameterized equations of motion in Eq. (3) simply as first-order ODEs, without regard to their Hamiltonian/symplectic structure, then their reparameterization using ϵ\epsilon is not complicated; simply scale the original ODEs from Eq. (3) by t˙=l\dot{t}=l.7

\begin{aligned} \dot{\mathbf{x}}&=\partial_{\boldsymbol{\pi}}\mathcal{H}-\boldsymbol{\alpha}_{\boldsymbol{\pi}}\\ \dot{\boldsymbol{\pi}}&=-\partial_{\mathbf{x}}\mathcal{H}+\boldsymbol{\alpha}_{\mathbf{x}} \end{aligned} \qquad \overset{dt=l\,d\epsilon}{\Longrightarrow} \qquad \begin{aligned} \dot{\mathbf{x}}&=\dot{t}\mathbf{x}'=l(\partial_{\boldsymbol{\pi}}\mathcal{H}-\boldsymbol{\alpha}_{\boldsymbol{\pi}})\\ \dot{\boldsymbol{\pi}}&=\dot{t}\boldsymbol{\pi}'=l(-\partial_{\mathbf{x}}\mathcal{H}+\boldsymbol{\alpha}_{\mathbf{x}) \end{aligned}} \tag*{(14)}\end{aligned}

However, the canonical Hamiltonian structure of the conservative part of the dynamics generally does not survive the above conformal scaling by ll (i.e., the evolution parameter transformation dt=ldε\mathrm{d}t = l\mathrm{d}\varepsilon).

Conformally Hamiltonian dynamics. To isolate the effect of the evolution parameter transformation in Eq. (14) on the Hamiltonian structure — without the separate issue of tracking the nonconservative forces — we momentarily limit consideration to the conservative case αx=απ=0\boldsymbol{\alpha}_{x} = \boldsymbol{\alpha}_{\pi} = 0. Eq. (14) simplifies to

x˙=∂πH,π˙=−∂xH→dt=ldεx˚=l∂πH,π˚=−l∂xH.(15)\dot{\mathbf{x}} = \partial_{\pi}\mathcal{H}, \qquad\dot{\boldsymbol{\pi}} = -\partial_{x}\mathcal{H} \quad\xrightarrow{\mathrm{d}t=l\mathrm{d}\varepsilon}\quad \mathring{\mathbf{x}} = l\partial_{\pi}\mathcal{H}, \qquad\mathring{\boldsymbol{\pi}} = -l\partial_{x}\mathcal{H}. \tag*{(15)}

where the original tt-parameterized system on the left has the canonical Hamiltonian structure. However, for a general conformal factor ll, the ε\varepsilon-parameterized system on the right does not necessarily retain that structure. That is, for general choices of l(x,π,t)l(\mathbf{x},\boldsymbol{\pi},t), there does not always exist a new Hamiltonian M(x,π,t)\mathcal{M}(\mathbf{x},\boldsymbol{\pi},t) such that the conformally scaled equations in Eq. (15) are equivalently written as a canonical Hamiltonian system:

x˚=l∂πH=∂πM,π˚=−l∂xH=−∂xM.(16)\mathring{\mathbf{x}} = l\partial_{\pi}\mathcal{H} = \partial_{\pi}\mathcal{M}, \qquad\mathring{\boldsymbol{\pi}} = -l\partial_{x}\mathcal{H} = -\partial_{x}\mathcal{M}. \tag*{(16)}

The ll-scaled system (i.e., ε\varepsilon-parameterized system) is thus generally not Hamiltonian in the original phase-space coordinates. It is sometimes said to be a conformally9 Hamiltonian system (with conformal factor ll); it need not preserve the canonical symplectic form or associated phase space volume.

Special cases. The loss of Hamiltonian structure of the ε\varepsilon-parameterized dynamics in Eq. (15) is generic, but not guaranteed; for certain special cases of ll, a new Hamiltonian M\mathcal{M} satisfying Eq. (16) can indeed be found. In particular, suppose that l=l(H,t)l = l(\mathcal{H},t) depends on (x,π)(\mathbf{x},\boldsymbol{\pi}) only through the original Hamiltonian:

l=l(H(x,π,t),t).(17)l = l\bigl(\mathcal{H}(\mathbf{x},\boldsymbol{\pi},t),t\bigr). \tag*{(17)}

In this case, there exists a function M\mathcal{M} that satisfies [13]:10

∂πM=l∂πH,∂xM=l∂xH,(18)\partial_{\pi}\mathcal{M} = l\partial_{\pi}\mathcal{H}, \qquad\partial_{x}\mathcal{M} = l\partial_{x}\mathcal{H}, \tag*{(18)}

such that the conformally scaled dynamics in Eq. (15) retain Hamiltonian form with Hamiltonian M\mathcal{M}, and can be written as in Eq. (16). Three simple examples are (for k≠−1k \ne-1):11

l=l(t)⇒M=lH∣l=Hk⇒M=1k+1Hk+1=1k+1lH∣l=H−1⇒M=ln⁡∣H∣.(19)l = l(t) \quad\Rightarrow\quad\mathcal{M} = l\mathcal{H} \qquad\bigg|\qquad l = \mathcal{H}^{k} \quad\Rightarrow\quad\mathcal{M} = \frac{1}{k+1}\mathcal{H}^{k+1} = \frac{1}{k+1}l\mathcal{H} \qquad\bigg|\qquad l = \mathcal{H}^{-1} \quad\Rightarrow\quad\mathcal{M} = \ln\lvert\mathcal{H}\rvert. \tag*{(19)}

Such cases require a particular relation between ll and H\mathcal{H} in order for Hamiltonian structure of the ε\varepsilon-parameterized dynamics to be preserved in the sense of Eq. (16). For a general conformal scaling function l(x,π,t)l(\mathbf{x},\boldsymbol{\pi},t), the Hamiltonian structure cannot generally be preserved on standard (nonextended) phase space; however, it can indeed be retained systematically by instead making use of the extended phase space, as detailed in the following.

Nonconservative Hamiltonian dynamics on extended phase space

The preceding section shows that direct conformal scaling generally does not preserve Hamiltonian structure on nonextended phase space (with certain special exceptions). We now show how a general prescribed evolution-parameter transformation can instead be implemented within the extended Hamiltonian framework in order to preserve this structure. The arbitrary nonconservative forces of §1 are incorporated throughout.

The general extended equations of motion

We first state the extended phase space dynamics, then derive them. To start, define extended phase space coordinates, x^:=(x,t)\widehat{\mathbf{x}} := (\mathbf{x},t) and π^:=(π,πt)\widehat{\boldsymbol{\pi}} := (\boldsymbol{\pi},\pi_t),

(x^,π^):=(x,t,π,πt)∈R2n+2(20)(\widehat{\mathbf{x}},\widehat{\boldsymbol{\pi}}) := (\mathbf{x},t,\boldsymbol{\pi},\pi_t) \in\mathbb{R}^{2n+2} \tag*{(20)}

where phase space is extended to include tt as a generalized configuration coordinate with conjugate momentum coordinate πt\pi_t. A physical interpretation of πt\pi_t is unclear at the moment, but will soon emerge in § 4.3.

The ε\varepsilon-parameterized canonical equations of motion for (x^,π^)∈R2n+2(\hat{x},\hat{\pi}) \in\mathbb{R}^{2n+2}, including nonconservative forces, are then obtained not from the original Hamiltonian H\mathcal{H} but, rather, from its corresponding extended Hamiltonian, H^:=l(H+πt)\hat{\mathcal{H}} := l(\mathcal{H}+\pi_t) — whose definition is motivated in the subsequent derivation — as follows:

H^(x^,π^):=l(H+πt)x^˙=∂π^H^−Aπ,t˙=∂πtH^=l,π^˙=−∂x^H^+Ax,π˙t=−∂tH^+At\begin{aligned} \hat{\mathcal{H}}(\hat{x},\hat{\pi}) &:= l(\mathcal{H}+\pi_t) \\ \dot{\hat{x}} &= \partial_{\hat{\pi}}\hat{\mathcal{H}}-\mathbf{A}_{\pi}, \qquad\dot{t}=\partial_{\pi_t}\hat{\mathcal{H}}=l, \\ \dot{\hat{\pi}} &= -\partial_{\hat{x}}\hat{\mathcal{H}}+\mathbf{A}_{x}, \qquad\dot{\pi}_t=-\partial_t\hat{\mathcal{H}}+A_t \end{aligned}
Ax:=lαx=lFnc⋅∂xr,Aπ:=lαπ=lFnc⋅∂πr,At=Eq.(3.3),Aπt=0(21)\begin{aligned} \mathbf{A}_x &:= l\boldsymbol{\alpha}_x=l\mathbf{F}^{\mathrm{nc}}\cdot\partial_x\mathbf{r}, \\ \mathbf{A}_{\pi} &:= l\boldsymbol{\alpha}_{\pi}=l\mathbf{F}^{\mathrm{nc}}\cdot\partial_{\pi}\mathbf{r}, \\ A_t &= \text{Eq.}(3.3), \\ A_{\pi_t} &= 0 \tag*{(21)} \end{aligned}

with αx=Fnc⋅∂xr\boldsymbol{\alpha}_x=\mathbf{F}^{\mathrm{nc}}\cdot\partial_x\mathbf{r} and απ=Fnc⋅∂πr\boldsymbol{\alpha}_{\pi}=\mathbf{F}^{\mathrm{nc}}\cdot\partial_{\pi}\mathbf{r} the usual generalized forces on nonextended space as in Eq. (3), and where we have introduced a new generalized “force” AtA_t which we soon show is given as follows:

At=−(αx⋅∂πH^−απ⋅∂xH^)=−Fnc⋅{r,H^}=−(αx⋅x˙+απ⋅π˙)(22)A_t=-(\boldsymbol{\alpha}_x\cdot\partial_{\pi}\hat{\mathcal{H}}-\boldsymbol{\alpha}_{\pi}\cdot\partial_x\hat{\mathcal{H}})=-\mathbf{F}^{\mathrm{nc}}\cdot\{\mathbf{r},\hat{\mathcal{H}}\}=-(\boldsymbol{\alpha}_x\cdot\dot{x}+\boldsymbol{\alpha}_{\pi}\cdot\dot{\pi}) \tag*{(22)}

Note the dynamics for (x^,π^)=(x,t,π,πt)(\hat{x},\hat{\pi})=(x,t,\pi,\pi_t) in Eq. (21) are also written in the familiar form,

x^˙=∂π^H^−A^x^,π^˙=−∂x^H^+A^π^(23)\dot{\hat{x}}=\partial_{\hat{\pi}}\hat{\mathcal{H}}-\hat{\mathbf{A}}_{\hat{x}}, \qquad\dot{\hat{\pi}}=-\partial_{\hat{x}}\hat{\mathcal{H}}+\hat{\mathbf{A}}_{\hat{\pi}} \tag*{(23)}

with

(A^x^,A^π^):=(Ax,At,Aπ,Aπt)=(lαx,At,lαπ,0)∈R2n+2.(24)(\hat{\mathbf{A}}_{\hat{x}},\hat{\mathbf{A}}_{\hat{\pi}}):=(\mathbf{A}_x,A_t,\mathbf{A}_{\pi},A_{\pi_t})=(l\boldsymbol{\alpha}_x,A_t,l\boldsymbol{\alpha}_{\pi},0)\in\mathbb{R}^{2n+2}. \tag*{(24)}

Derivation of Eq. (21). In §1, nonconservative Hamiltonian dynamics in standard, nonextended phase space coordinates were derived using the Hamilton-d’Alembert principle, δI+∫t0tfδWncdt=0\delta I+\int_{t_0}^{t_f}\delta W^{\mathrm{nc}}\mathrm{d}t=0 (see Eq. (7)). For extended phase space, we begin by reparameterizing the integrals in Eq. (7) using dt=t˙dε\mathrm{d}t=\dot{t}\mathrm{d}\varepsilon:

I=∫t0tf(π⋅x˙−H)dt=∫ε0εft˙(π⋅x˙−H)dε=∫ε0εf(π^⋅x^˙−t˙(H+πt))dε,I=\int_{t_0}^{t_f}(\pi\cdot\dot{x}-\mathcal{H})\mathrm{d}t=\int_{\varepsilon_0}^{\varepsilon_f}\dot{t}(\pi\cdot\dot{x}-\mathcal{H})\mathrm{d}\varepsilon=\int_{\varepsilon_0}^{\varepsilon_f}(\hat{\pi}\cdot\dot{\hat{x}}-\dot{t}(\mathcal{H}+\pi_t))\mathrm{d}\varepsilon,
∫t0tfδWncdt=∫ε0εft˙δWncdε(25)\int_{t_0}^{t_f}\delta W^{\mathrm{nc}}\mathrm{d}t=\int_{\varepsilon_0}^{\varepsilon_f}\dot{t}\delta W^{\mathrm{nc}}\mathrm{d}\varepsilon \tag*{(25)}

where we have also expressed II in terms of the extended coordinates, (x^,π^)=(x,t,π,πt)(\hat{x},\hat{\pi})=(x,t,\pi,\pi_t), and have used the relation t˙dtdt=dtdε=t˙\dot{t}\frac{\mathrm{d}t}{\mathrm{d}t}=\frac{\mathrm{d}t}{\mathrm{d}\varepsilon}=\dot{t}.12 This expression for II motivates defining an extended Hamiltonian H^\hat{\mathcal{H}} and action integral I^\hat{I} as:

H^:=l(H+πt),I^:=∫ε0εf(π^⋅x^˙−H^)dε.(26)\hat{\mathcal{H}}:=l(\mathcal{H}+\pi_t), \qquad\hat{I}:=\int_{\varepsilon_0}^{\varepsilon_f}(\hat{\pi}\cdot\dot{\hat{x}}-\hat{\mathcal{H}})\mathrm{d}\varepsilon. \tag*{(26)}

Note I^\hat{I} is not simply another expression for the reparameterized II in Eq. (25); their integrands are related by:13

π^⋅x^˙−H^=t˙(π⋅x˙−H)+(t˙−l)(H+πt),(27)\hat{\pi}\cdot\dot{\hat{x}}-\hat{\mathcal{H}}=\dot{t}(\pi\cdot\dot{x}-\mathcal{H})+(\dot{t}-l)(\mathcal{H}+\pi_t), \tag*{(27)}

which relies only on dtdt=dtdt\frac{\mathrm{d}t}{\mathrm{d}t}=\frac{\mathrm{d}t}{\mathrm{d}t} (i.e., t˙□=□˙\dot{t}_{\square}=\dot{\square}) and the definition of H^\hat{\mathcal{H}}, but where we cannot yet assume that t˙=l\dot{t}=l.14

Motivated by Eq. (25)–Eq. (27), we formulate the Hamilton-d’Alembert principle on extended phase space, compared to the classic formulation in Eq. (7), as follows:

δI+∫t0tfδWncdt=0\delta I+\int_{t_0}^{t_f}\delta W^{\mathrm{nc}}\mathrm{d}t=0

with

I=∫t0tf(π⋅x˙−H)dt,δWnc=Fnc⋅δrI=\int_{t_0}^{t_f}(\pi\cdot\dot{x}-\mathcal{H})\mathrm{d}t,\qquad\delta W^{\mathrm{nc}}=\mathbf{F}^{\mathrm{nc}}\cdot\delta\mathbf{r}
⟺\Longleftrightarrow
δI^+∫ε0εfδW^ncdε=0\delta\hat{I}+\int_{\varepsilon_0}^{\varepsilon_f}\delta\hat{W}^{\mathrm{nc}}\mathrm{d}\varepsilon=0

with

I^:=∫ε0εf(π^⋅x^˙−H^)dε,δW^nc:=t˙δWnc=Fnc⋅(t˙δr−r˙δt)(28)\hat{I}:=\int_{\varepsilon_0}^{\varepsilon_f}(\hat{\pi}\cdot\dot{\hat{x}}-\hat{\mathcal{H}})\mathrm{d}\varepsilon,\qquad\delta\hat{W}^{\mathrm{nc}}:=\dot{t}\delta W^{\mathrm{nc}}=\mathbf{F}^{\mathrm{nc}}\cdot(\dot{t}\delta\mathbf{r}-\dot{\mathbf{r}}\delta t) \tag*{(28)}

where we have assigned the notation δW^nc:=t˙δWnc\delta\hat{W}^{\mathrm{nc}}:=\dot{t}\delta W^{\mathrm{nc}}. This virtual work term will require careful handling in the extended setting. Above, note we have claimed, without explanation, that this is given by Fnc⋅(t˙δr−r˙δt)\mathbf{F}^{\mathrm{nc}}\cdot(\dot{t}\delta\mathbf{r}-\dot{\mathbf{r}}\delta t). This will be justified shortly. For the moment, we proceed with the derivation taking this relation for granted.

Let us first address the δI^\delta\hat{I} term in Eq. (28), which is more straightforward and handled the same as the classic variation δI\delta I in the nonextended setting. Carrying out the variation yields:

δI^=δ∫ε0εf(π^⋅x^˙−H^)dε=(π^⋅δx^)∣ε0εf+∫ε0εfδ(π^⋅x^˙−H^)dε=∫ε0εf(π^⋅δx^˙+x^˙⋅δπ^−δH^)dε=π^⋅δx^∣ε0εf+∫ε0εf(−π^˙⋅δx^+δπ^⋅x^˙−δH^)dε=∫ε0εf[−(π^˙+∂x^H^)⋅δx^+(x^˙−∂π^H^)⋅δπ^]dε=∫ε0εf[−(π˙+∂xH^)⋅δx+(x˙−∂πH^)⋅δπ−(π˙t+∂tH^)δt+(t˙−∂πtH^)δπt]dε.(29)\begin{aligned} \delta\hat{I} &=\delta\int_{\varepsilon_0}^{\varepsilon_f}(\hat{\pi}\cdot\dot{\hat{x}}-\hat{\mathcal{H}})\mathrm{d}\varepsilon\\ &=(\hat{\pi}\cdot\delta\hat{x})\big|_{\varepsilon_0}^{\varepsilon_f}+\int_{\varepsilon_0}^{\varepsilon_f}\delta(\hat{\pi}\cdot\dot{\hat{x}}-\hat{\mathcal{H}})\mathrm{d}\varepsilon\\ &=\int_{\varepsilon_0}^{\varepsilon_f}(\hat{\pi}\cdot\delta\dot{\hat{x}}+\dot{\hat{x}}\cdot\delta\hat{\pi}-\delta\hat{\mathcal{H}})\mathrm{d}\varepsilon\\ &=\hat{\pi}\cdot\delta\hat{x}\big|_{\varepsilon_0}^{\varepsilon_f}+\int_{\varepsilon_0}^{\varepsilon_f}(-\dot{\hat{\pi}}\cdot\delta\hat{x}+\delta\hat{\pi}\cdot\dot{\hat{x}}-\delta\hat{\mathcal{H}})\mathrm{d}\varepsilon\\ &=\int_{\varepsilon_0}^{\varepsilon_f}\left[-(\dot{\hat{\pi}}+\partial_{\hat{x}}\hat{\mathcal{H}})\cdot\delta\hat{x}+(\dot{\hat{x}}-\partial_{\hat{\pi}}\hat{\mathcal{H}})\cdot\delta\hat{\pi}\right]\mathrm{d}\varepsilon\\ &=\int_{\varepsilon_0}^{\varepsilon_f}\left[-(\dot{\pi}+\partial_x\hat{\mathcal{H}})\cdot\delta x+(\dot{x}-\partial_{\pi}\hat{\mathcal{H}})\cdot\delta\pi-(\dot{\pi}_t+\partial_t\hat{\mathcal{H}})\delta t+(\dot{t}-\partial_{\pi_t}\hat{\mathcal{H}})\delta\pi_t\right]\mathrm{d}\varepsilon. \tag*{(29)} \end{aligned}

where the following steps were used. 1st line: the variation is taken inside the integral as usual using the fact δεo=δεf=0\delta\varepsilon_o=\delta\varepsilon_f=0. 2nd line: integration by parts is used on the term with δx˙\delta\dot{x}, along with the fact that δx^\delta\hat{x} vanish at εo\varepsilon_o and εf\varepsilon_f. 3rd and 4th lines: we simply write out δH^=∂x^H^⋅δx^+∂π^H^⋅δπ^\delta\hat{H}=\partial_{\hat{x}}\hat{H}\cdot\delta\hat{x}+\partial_{\hat{\pi}}\hat{H}\cdot\delta\hat{\pi}.

Next, momentarily taking the expression for virtual work δW^nc\delta\hat{W}^{\mathrm{nc}} in Eq. (28) for granted (it is derived further down), and considering the general case that r=r(x,π,t)\mathbf{r}=\mathbf{r}(\mathbf{x},\boldsymbol{\pi},t), we find this term is expressed in coordinates (x^,π^)(\hat{x},\hat{\pi}) as:

δW^nc:=t˙δWnc=Fnc⋅(δr−r˙δt)=Fnc⋅[t˙(∂r∂x⋅δx+∂r∂π⋅δπ)−(∂r∂x⋅x˙+∂r∂t+∂r∂π⋅π˙)δt]=Fnc⋅[t˙(∂r∂x⋅δx+∂r∂π⋅δπ)−(∂r∂x⋅x˙+∂r∂t+∂r∂π⋅π˙)δt]=t˙(αx⋅δx+απ⋅δπ)−(αx⋅x˙+απ⋅π˙)δt(30)\begin{aligned} \delta\hat{W}^{\mathrm{nc}}&:=\dot{t}\delta W^{\mathrm{nc}}=\mathbf{F}^{\mathrm{nc}}\cdot(\delta\mathbf{r}-\dot{\mathbf{r}}\delta t)\\ &=\mathbf{F}^{\mathrm{nc}}\cdot\left[\dot{t}\left(\frac{\partial\mathbf{r}}{\partial\mathbf{x}}\cdot\delta\mathbf{x}+\frac{\partial\mathbf{r}}{\partial\boldsymbol{\pi}}\cdot\delta\boldsymbol{\pi}\right)-\left(\frac{\partial\mathbf{r}}{\partial\mathbf{x}}\cdot\dot{\mathbf{x}}+\frac{\partial\mathbf{r}}{\partial t}+\frac{\partial\mathbf{r}}{\partial\boldsymbol{\pi}}\cdot\dot{\boldsymbol{\pi}}\right)\delta t\right]\\ &=\mathbf{F}^{\mathrm{nc}}\cdot\left[\dot{t}\left(\frac{\partial\mathbf{r}}{\partial\mathbf{x}}\cdot\delta\mathbf{x}+\frac{\partial\mathbf{r}}{\partial\boldsymbol{\pi}}\cdot\delta\boldsymbol{\pi}\right)-\left(\frac{\partial\mathbf{r}}{\partial\mathbf{x}}\cdot\dot{\mathbf{x}}+\frac{\partial\mathbf{r}}{\partial t}+\frac{\partial\mathbf{r}}{\partial\boldsymbol{\pi}}\cdot\dot{\boldsymbol{\pi}}\right)\delta t\right]\\ &=\dot{t}(\boldsymbol{\alpha}_{x}\cdot\delta\mathbf{x}+\boldsymbol{\alpha}_{\pi}\cdot\delta\boldsymbol{\pi})-(\boldsymbol{\alpha}_{x}\cdot\dot{\mathbf{x}}+\boldsymbol{\alpha}_{\pi}\cdot\dot{\boldsymbol{\pi}})\delta t \tag*{(30)} \end{aligned}

with the usual αx:=Fnc⋅∂xr\boldsymbol{\alpha}_{x}:=\mathbf{F}^{\mathrm{nc}}\cdot\partial_{x}\mathbf{r} and απ:=Fnc⋅∂πr\boldsymbol{\alpha}_{\pi}:=\mathbf{F}^{\mathrm{nc}}\cdot\partial_{\pi}\mathbf{r}. Substitution of Eq. (29) and Eq. (30) into Eq. (28) then gives:

0=∫εoεf[−(π˙+∂xH^−t˙αx)⋅δx+(x˙−∂πH^+t˙απ)⋅δπ+(t˙−∂πtH^)δπt−(π˙t+∂tH^+αx⋅x˙+απ⋅π˙)δt]dε.0=\int_{\varepsilon_o}^{\varepsilon_f}\left[-(\dot{\boldsymbol{\pi}}+\partial_x\hat{H}-\dot{t}\boldsymbol{\alpha}_x)\cdot\delta\mathbf{x}+(\dot{\mathbf{x}}-\partial_\pi\hat{H}+\dot{t}\boldsymbol{\alpha}_\pi)\cdot\delta\boldsymbol{\pi}+(\dot{t}-\partial_{\pi_t}\hat{H})\delta\pi_t-(\dot{\pi}_t+\partial_t\hat{H}+\boldsymbol{\alpha}_x\cdot\dot{\mathbf{x}}+\boldsymbol{\alpha}_\pi\cdot\dot{\boldsymbol{\pi}})\delta t\right]d\varepsilon.

For the above to hold for all arbitrary (δx^,δπ^)=(δx,δt,δπ,δπt)(\delta\hat{x},\delta\hat{\pi})=(\delta x,\delta t,\delta\pi,\delta\pi_t) requires that their coefficients vanish, leading to:

x˙=∂πH^−t˙απ,π˙=−∂xH^+t˙αx,t˙=∂πtH^=l,π˙t=−∂tH^−(αx⋅x˙+απ⋅π˙).(31)\dot{\mathbf{x}}=\partial_\pi\hat{H}-\dot{t}\boldsymbol{\alpha}_\pi,\qquad \dot{\boldsymbol{\pi}}=-\partial_x\hat{H}+\dot{t}\boldsymbol{\alpha}_x,\qquad \dot{t}=\partial_{\pi_t}\hat{H}=l,\qquad \dot{\pi}_t=-\partial_t\hat{H}-(\boldsymbol{\alpha}_x\cdot\dot{\mathbf{x}}+\boldsymbol{\alpha}_\pi\cdot\dot{\boldsymbol{\pi}}). \tag*{(31)}

Note that ∂πtH^=l\partial_{\pi_t}\hat{H}=l always holds (assuming ∂πtl=0\partial_{\pi_t}l=0) such that any solution curve must automatically satisfy the desired relation t˙=l\dot{t}=l. This fact allows for t˙αx\dot{t}\boldsymbol{\alpha}_x and t˙απ\dot{t}\boldsymbol{\alpha}_\pi above to be replaced by Ax=lαx\mathbf{A}_x=l\boldsymbol{\alpha}_x and Aπ=lαπ\mathbf{A}_\pi=l\boldsymbol{\alpha}_\pi, as in Eq. (21). We have, therefore, now verified the ODEs for x˙\dot{\mathbf{x}}, π˙\dot{\boldsymbol{\pi}}, and t˙\dot{t} that were claimed in Eq. (21). For the remaining ODE for π˙t\dot{\pi}_t, note that the term αx⋅x˙+απ⋅π˙\boldsymbol{\alpha}_x\cdot\dot{\mathbf{x}}+\boldsymbol{\alpha}_\pi\cdot\dot{\boldsymbol{\pi}}, after substitution of the above dynamics for (x˙,π˙)(\dot{\mathbf{x}},\dot{\boldsymbol{\pi}}), is given by:

αx⋅x˙+απ⋅π˙=αx⋅∂πH^−απ⋅∂xH^−t˙(αx⋅απ−απ⋅αx)=:−At⇒π˙t=−∂tH^+At(32)\boldsymbol{\alpha}_x\cdot\dot{\mathbf{x}}+\boldsymbol{\alpha}_\pi\cdot\dot{\boldsymbol{\pi}}=\boldsymbol{\alpha}_x\cdot\partial_\pi\hat{H}-\boldsymbol{\alpha}_\pi\cdot\partial_x\hat{H}-\dot{t}(\boldsymbol{\alpha}_x\cdot\boldsymbol{\alpha}_\pi-\boldsymbol{\alpha}_\pi\cdot\boldsymbol{\alpha}_x)=:-A_t\qquad\Rightarrow\qquad\dot{\pi}_t=-\partial_t\hat{H}+A_t \tag*{(32)}

which verifies the ODE for π˙t\dot{\pi}_t that was claimed in Eq. (21).

Additionally, upon imposing the equations of motion in Eq. (21), note that the extended virtual work in Eq. (30) is then expressed as follows (along solution curves of Eq. (21)):

δW^nc=t˙αx⋅δx+t˙απ⋅δπ−(αx⋅x˙+απ⋅π˙)δt\delta\hat{W}^{\mathrm{nc}}=\dot{t}\boldsymbol{\alpha}_x\cdot\delta\mathbf{x}+\dot{t}\boldsymbol{\alpha}_\pi\cdot\delta\boldsymbol{\pi}-(\boldsymbol{\alpha}_x\cdot\dot{\mathbf{x}}+\boldsymbol{\alpha}_\pi\cdot\dot{\boldsymbol{\pi}})\delta t
A^x⋅δx^+A^π⋅δπ^=Ax⋅δx+Aπ⋅δπ+At⋅δt+A^πt⋅δπt(33)\hat{\mathbf{A}}_x\cdot\delta\hat{\mathbf{x}}+\hat{\mathbf{A}}_\pi\cdot\delta\hat{\boldsymbol{\pi}}=\mathbf{A}_x\cdot\delta\mathbf{x}+\mathbf{A}_\pi\cdot\delta\boldsymbol{\pi}+A_t\cdot\delta t+\hat{A}_{\pi_t}\cdot\delta\pi_t \tag*{(33)}

The first line is the general kinematic expression for the extended virtual work, whereas the second line expresses the same quantity assuming the equations of motion in Eq. (21) are satisfied (with (A^x,A^π)(\hat{\mathbf{A}}_x,\hat{\mathbf{A}}_\pi) defined in Eq. (24)).

On the virtual work in Eq. (28). The above derivation is not quite complete. It remains for us to justify the expression for the nonconservative virtual work that we claimed in Eq. (28):

δW^nc:=t˙δWnc=Fnc⋅(δr−r˙δt)i.e.,δWnc=Fnc⋅(δr−r˙δt)(34)\delta\hat{W}^{\mathrm{nc}}:=\dot{t}\delta W^{\mathrm{nc}}=\mathbf{F}^{\mathrm{nc}}\cdot(\delta\mathbf{r}-\dot{\mathbf{r}}\delta t)\qquad\text{i.e.,}\qquad\delta W^{\mathrm{nc}}=\mathbf{F}^{\mathrm{nc}}\cdot(\delta\mathbf{r}-\dot{\mathbf{r}}\delta t) \tag*{(34)}

where δW^nc:=t˙δWnc\delta\hat{W}^{\mathrm{nc}}:=\dot{t}\delta W^{\mathrm{nc}} is an assignment of notation that arises from the reparameterization in Eq. (25). What needs justification is our claimed expression for δWnc\delta W^{\mathrm{nc}} in terms of Fnc\mathbf{F}^{\mathrm{nc}} and (r,t)(\mathbf{r},t) seen above.

Note that δWnc\delta W^{\mathrm{nc}} in Eq. (34) appears, at first glance, to have a different meaning than it did in the nonextended setting in Eq. (7) where we said δWnc=Fnc⋅δr\delta W^{\mathrm{nc}}=\mathbf{F}^{\mathrm{nc}}\cdot\delta\mathbf{r}. There is, in fact, no contradiction; it appears that there is because δ\delta now implies something different on extended space than it did previously in §1. In §1, the variation δ\delta had its typical interpretation with tt as the evolution parameter; let us denote it δt\delta_t (variation at fixed tt). However, in the present context of extended space, the variation δ\delta is now with ε\varepsilon as the evolution parameter and tt a coordinate like any other; let us denote it δε\delta_\varepsilon (variation at fixed ε\varepsilon). On extended space, δr\delta\mathbf{r} always implies that at fixed ε\varepsilon, δεr\delta_\varepsilon\mathbf{r}; it consists of the displacement at fixed tt, δtr\delta_t\mathbf{r}, and the displacement during the accompanying change in time, r˙δεt\dot{\mathbf{r}}\delta_\varepsilon t, such that:

δεr=δtr+r˙δεt,δtr=δεr−r˙δεt(35)\delta_\varepsilon\mathbf{r}=\delta_t\mathbf{r}+\dot{\mathbf{r}}\delta_\varepsilon t,\qquad \delta_t\mathbf{r}=\delta_\varepsilon\mathbf{r}-\dot{\mathbf{r}}\delta_\varepsilon t \tag*{(35)}

Thus, δWnc\delta W^{\mathrm{nc}} represents the same quantity, given by Fnc⋅δtr\mathbf{F}^{\mathrm{nc}}\cdot\delta_t\mathbf{r}, on both nonextended and extended space. Yet, its expression differs on extended space (where δ≡δε≡δt\delta\equiv\delta_\varepsilon\equiv\delta_t):

nonext. phase space, δ≡δt:δWnc=Fnc⋅δtrext. phase space, δ≡δε:δWnc=Fnc⋅δtr=Fnc⋅(δεr−r˙δεt)(≠Fnc⋅δεr)(36)\begin{aligned} \text{nonext. phase space},\ \delta\equiv\delta_t:\quad&\delta W^{\mathrm{nc}}=\mathbf{F}^{\mathrm{nc}}\cdot\delta_t\mathbf{r}\\ \text{ext. phase space},\ \delta\equiv\delta_\varepsilon:\quad&\delta W^{\mathrm{nc}}=\mathbf{F}^{\mathrm{nc}}\cdot\delta_t\mathbf{r}=\mathbf{F}^{\mathrm{nc}}\cdot(\delta_\varepsilon\mathbf{r}-\dot{\mathbf{r}}\delta_\varepsilon t)\quad(\ne\mathbf{F}^{\mathrm{nc}}\cdot\delta_\varepsilon\mathbf{r}) \tag*{(36)} \end{aligned}

When writing δW^nc:=t˙δWnc\delta\hat{W}^{\mathrm{nc}}:=\dot{t}\delta W^{\mathrm{nc}} in Eq. (28) or Eq. (34), it is the second expression above, with δ≡δε\delta\equiv\delta_\varepsilon, that is relevant:

δW^nc=t˙δWnc=t˙Fnc⋅δtr=Fnc⋅(t˙δεr−r˙δεt)≡Fnc⋅(t˙δr−r˙δt)(37)\delta\hat{W}^{\mathrm{nc}}=\dot{t}\delta W^{\mathrm{nc}}=\dot{t}\mathbf{F}^{\mathrm{nc}}\cdot\delta_t\mathbf{r}=\mathbf{F}^{\mathrm{nc}}\cdot(\dot{t}\delta_\varepsilon\mathbf{r}-\dot{\mathbf{r}}\delta_\varepsilon t)\equiv\mathbf{F}^{\mathrm{nc}}\cdot(\dot{t}\delta\mathbf{r}-\dot{\mathbf{r}}\delta t) \tag*{(37)}

where we have used the purely kinematic relation t˙r=r˙\dot{t}\mathbf{r}=\dot{\mathbf{r}}.

Extended Poisson bracket & conservation of the extended Hamiltonian.

Let {⋅,⋅}\{\cdot,\cdot\} be the canonical Poisson bracket on nonextended 2n2n-dim phase space as in Eq. (2). We define an analogous Poisson bracket, {{⋅,⋅}}\{\{\cdot,\cdot\}\}, on extended (2n+2)(2n+2)-dim phase space as follows for any scalar-valued ff and hh:

{{f,h}}:={f,h}+∂f∂t∂h∂πt−∂f∂πt∂h∂t.(38)\{\{f,h\}\}:=\{f,h\}+\frac{\partial f}{\partial t}\frac{\partial h}{\partial\pi_t}-\frac{\partial f}{\partial\pi_t}\frac{\partial h}{\partial t}. \tag*{(38)}

The ODE governing the evolution of any function f^(x^,π^)\hat{f}(\hat{x},\hat{\pi}) is then given in terms of the above, and the extended generalized nonconservative forces (A^x^,A^π^):=(lαx,At,lαπ,0)(\hat{A}_{\hat{x}},\hat{A}_{\hat{\pi}}):=(l\alpha_x,A_t,l\alpha_\pi,0) from Eq. (21), as:

f^˙(x^,π^)={{f,H^}}+A^x^⋅∂f∂π^−A^π^⋅∂f∂x^={{f,H^}}+Ax⋅∂f∂π−Aπ⋅∂f∂x+At∂f∂πt={{f,H^}}+l(αx⋅∂f∂π−απ⋅∂f∂x)−∂f∂πt(αx⋅∂H^∂π−απ⋅∂H^∂x).(39)\begin{aligned} \dot{\hat{f}}(\hat{x},\hat{\pi})=\{\{f,\hat{H}\}\}+\hat{A}_{\hat{x}}\cdot\frac{\partial f}{\partial\hat{\pi}}-\hat{A}_{\hat{\pi}}\cdot\frac{\partial f}{\partial\hat{x}}=\{\{f,\hat{H}\}\}+A_x\cdot\frac{\partial f}{\partial\pi}-A_\pi\cdot\frac{\partial f}{\partial x}+A_t\frac{\partial f}{\partial\pi_t} \\ &=\{\{f,\hat{H}\}\}+l\left(\alpha_x\cdot\frac{\partial f}{\partial\pi}-\alpha_\pi\cdot\frac{\partial f}{\partial x}\right)-\frac{\partial f}{\partial\pi_t}\left(\alpha_x\cdot\frac{\partial\hat{H}}{\partial\pi}-\alpha_\pi\cdot\frac{\partial\hat{H}}{\partial x}\right). \tag*{(39)} \end{aligned}

(equivalently, f^˙={{f,H^}}+lFnc⋅{r,f}−∂f∂πtFnc⋅{r,H^}\dot{\hat{f}}=\{\{f,\hat{H}\}\}+lF^{\mathrm{nc}}\cdot\{r,f\}-\frac{\partial f}{\partial\pi_t}F^{\mathrm{nc}}\cdot\{r,\hat{H}\}).

It follows immediately that H^\hat{H} itself is always a constant of motion:

Remark 3.1. The extended Hamiltonian, H^=l(H+πt)\hat{H}=l(H+\pi_t), is always — even in the presence of nonconservative forces — a constant of motion of the extended phase space dynamics in Eq. (21).15

H^˙={{H^,H^}}+A^x^⋅∂H^∂π^−A^π^⋅∂H^∂x^=lαx⋅∂H^∂π−lαπ⋅∂H^∂x+At∂H^∂πt=l(αx⋅∂H^∂π−απ⋅∂H^∂x)+lAt=−lAt+lAt=0.\dot{\hat{H}}=\{\{\hat{H},\hat{H}\}\}+\hat{A}_{\hat{x}}\cdot\frac{\partial\hat{H}}{\partial\hat{\pi}}-\hat{A}_{\hat{\pi}}\cdot\frac{\partial\hat{H}}{\partial\hat{x}}=l\alpha_x\cdot\frac{\partial\hat{H}}{\partial\pi}-l\alpha_\pi\cdot\frac{\partial\hat{H}}{\partial x}+A_t\frac{\partial\hat{H}}{\partial\pi_t}=l\left(\alpha_x\cdot\frac{\partial\hat{H}}{\partial\pi}-\alpha_\pi\cdot\frac{\partial\hat{H}}{\partial x}\right)+lA_t=-lA_t+lA_t=0.

Thus, H^\hat{H} not just a constant of motion, but an integral of motion (i.e., evolution-parameter-independent constant of motion).

integralof motion:
H^˙=Ax⋅∂H^∂π−Aπ⋅∂H^∂x+At∂H^∂πt=0\dot{\hat{H}}=A_x\cdot\frac{\partial\hat{H}}{\partial\pi}-A_\pi\cdot\frac{\partial\hat{H}}{\partial x}+A_t\frac{\partial\hat{H}}{\partial\pi_t}=0⇒H^=H^0=const.\Rightarrow\hat{H}=\hat{H}_0=\mathit{const.}

Table integral.

This holds regardless of whether or not HH is conserved under the nonextended phase space dynamics of Eq. (3).

Now, most quantities of physical interest are functions f(x,π)f(x,\pi) on nonextended phase space, or, more generally, functions f(x,π,t)f(x,\pi,t) with explicit tt-dependence. Such expressions can be viewed as functions on extended phase space with no explicit ϵ\epsilon-dependence and which are further independent of πt\pi_t. For such functions, Eq. (39) with ∂πtf=0\partial_{\pi_t}f=0 (and ∂πtH^=l\partial_{\pi_t}\hat{H}=l) leads to:

f˙(x,π,t)={{f,H^}}+Ax⋅∂πf−Aπ⋅∂xf={{f,H^}}+l(αx⋅∂πf−απ⋅∂xf)={f,H^}+l∂tf+l(αx⋅∂πf−απ⋅∂xf).(40)\begin{aligned} \dot{f}(x,\pi,t)=\{\{f,\hat{H}\}\}+A_x\cdot\partial_{\pi}f-A_\pi\cdot\partial_xf=\{\{f,\hat{H}\}\}+l(\alpha_x\cdot\partial_{\pi}f-\alpha_\pi\cdot\partial_xf) \\ &=\{f,\hat{H}\}+l\partial_tf+l(\alpha_x\cdot\partial_{\pi}f-\alpha_\pi\cdot\partial_xf). \tag*{(40)} \end{aligned}

(equivalently, f˙={f,H^}+l∂tf+lFnc⋅{r,f}\dot{f}=\{f,\hat{H}\}+l\partial_tf+lF^{\mathrm{nc}}\cdot\{r,f\}).

Expanding {f,H^}\{f,\hat{H}\} using H^=l(H+πt)\hat{H}=l(H+\pi_t), then Eq. (40) yields:16

f˙(x,π,t)=l({f,H}+∂tf+αx⋅∂πf−απ⋅∂xf)+H^l{f,l}=lf˙+H^l{f,l}.(41)\dot{f}(x,\pi,t)=l\left(\{f,H\}+\partial_tf+\alpha_x\cdot\partial_{\pi}f-\alpha_\pi\cdot\partial_xf\right)+\frac{\hat{H}}{l}\{f,l\}=l\dot{f}+\frac{\hat{H}}{l}\{f,l\}. \tag*{(41)}

with f˙={f,H}+∂tf+αx⋅∂πf−απ⋅∂xf\dot{f}=\{f,H\}+\partial_tf+\alpha_x\cdot\partial_{\pi}f-\alpha_\pi\cdot\partial_xf being the usual tt-parameterized governing ODE for any f(x,π,t)f(x,\pi,t) on nonextended space, as per Eq. (4). The implications of Eq. (41) and Remark 3.1 are considered next.

The physical extended dynamics: restriction to the zero level set.

For arbitrary initial conditions (x^0,π^0)∈R2n+2(\hat{x}_0,\hat{\pi}_0)\in\mathbb{R}^{2n+2}, there exists a unique solution curve, (x^ϵ,π^ϵ)(\hat{x}_{\epsilon},\hat{\pi}_{\epsilon}), satisfying the general extended dynamics Eq. (21). However, the projection of such solutions back to the standard, nonextended phase space — that is, the “(x,π)(x,\pi)-part” of the solution — does not necessarily agree with the true physical dynamics on nonextended phase space given in Eq. (3), even for the same (x0,π0)∈R2n(x_0,\pi_0)\in\mathbb{R}^{2n}.

The zero level set, Σ0\Sigma_0. As established in Remark 3.1, H^\hat{H} is conserved and, therefore, for any value κ∈R\kappa\in\mathbb{R}, the level set Σκ={H^=κ}\Sigma_\kappa=\{\hat{H}=\kappa\} is invariant under the extended dynamics of Eq. (21). More specifically, each Σκ\Sigma_\kappa

Σκ:={(x^,π^)∈R2n+2∣H^(x^,π^)=κ}⊂R2n+2,(42)\Sigma_\kappa:=\left\{(\hat{x},\hat{\pi})\in\mathbb{R}^{2n+2}\mid\hat{H}(\hat{x},\hat{\pi})=\kappa\right\}\subset\mathbb{R}^{2n+2}, \tag*{(42)}

is a (2n+1)(2n+1)-dim embedded hypersurface in (2n+2)(2n+2)-dim extended phase space that is invariant under the full extended dynamics — including nonconservative forces — of Eq. (21). The extended dynamics admit solution curves on any/every such surface Σκ\Sigma_\kappa. Yet, we will show that one of them, the zero level set (κ=0)(\kappa=0),

Σ0={H^=0}={πt=−H},(43)\Sigma_0=\{\hat{H}=0\}=\{\pi_t=-H\}, \tag*{(43)}

is distinguished: when restricted to Σ0\Sigma_0, the extended dynamics universally reproduce (up to reparameterization) the original nonextended dynamics in Eq. (3), and the additional momentum πt\pi_t satisfies πt=−H\pi_t=-\mathcal{H}. 17 This is verified starting around Eq. (45) below. First, let us clarify the relations in Eq. (43):

Since Σ0⊂R2n+2\Sigma_0\subset\mathbb{R}^{2n+2} is invariant under the extended dynamics, every solution curve starting in Σ0\Sigma_0 remains in Σ0\Sigma_0 and, consequently, every such solution curve (x^ε,π^ε)∈Σ0(\widehat{x}_\varepsilon,\widehat{\pi}_\varepsilon)\in\Sigma_0 satisfies,

H^(x^ε,π^ε)=0⟺(πt)ε=−H(xε,πε,tε),(44)\widehat{\mathcal{H}}(\widehat{x}_\varepsilon,\widehat{\pi}_\varepsilon)=0 \qquad\Longleftrightarrow\qquad(\pi_t)_\varepsilon=-\mathcal{H}(x_\varepsilon,\pi_\varepsilon,t_\varepsilon), \tag*{(44)}

for all ε\varepsilon. When we use the casual notation H^=0\widehat{\mathcal{H}}=0 or πt=−H\pi_t=-\mathcal{H}, it should be interpreted as above. It does not mean, for instance, that πt\pi_t should be treated as a function, −H(x,π,t)-\mathcal{H}(x,\pi,t), of the other coordinates. This distinction is important when differentiating H^\widehat{\mathcal{H}} to obtain the extended ODEs in Eq. (21); when doing so, πt\pi_t must be treated as an independent coordinate. However, after obtaining the extended ODEs this way, we may then restrict them to Σ0\Sigma_0 by substituting into their right-hand sides the relation H^=0\widehat{\mathcal{H}}=0 (equivalently, πt=−H\pi_t=-\mathcal{H}).

We note that restriction of the extended dynamics to Σ0\Sigma_0 imposes no physical constraints on the original nonextended dynamics. More specifically, it imposes no restriction on the admissible initial data of the nonextended system: for any initial condition (x0,π0)∈R2n(x_0,\pi_0)\in\mathbb{R}^{2n} at time t0t_0, there is a unique choice of the initial value of the additional momentum coordinate, (πt)0=−H(x0,π0,t0)(\pi_t)_0=-\mathcal{H}(x_0,\pi_0,t_0), that places the corresponding extended initial condition (x^0,π^0)(\widehat{x}_0,\widehat{\pi}_0) in Σ0⊂R2n+2\Sigma_0\subset\mathbb{R}^{2n+2}. Equivalently, H^(x^0,π^0)=0\widehat{\mathcal{H}}(\widehat{x}_0,\widehat{\pi}_0)=0. Since Σ0\Sigma_0 is invariant, the relation H^=0\widehat{\mathcal{H}}=0, and the equivalent relation πt=−H\pi_t=-\mathcal{H}, remains satisfied along the resulting extended solution curve. Thus, the restriction constrains only the additional coordinate πt\pi_t; it does not further constrain or exclude any initial condition of — and hence no corresponding evolution of — the original system. Rather, it selects an invariant class of extended solution curves that are guaranteed to agree, up to the prescribed reparameterization, with the original nonextended dynamics, as shown below.

Recovering the nonextended dynamics. The role of the zero level set is already visible in Eq. (41). For any observable/function f(x,π,t)f(x,\pi,t) "lifted" to a function on extended phase space:

f˙=lf˙+H^l{f,l},(45)\dot{f}=l\dot{f}+\frac{\widehat{\mathcal{H}}}{l}\{f,l\}, \tag*{(45)}

with f˙\dot{f} as in Eq. (4). Thus, restriction to Σ0\Sigma_0 guarantees the ε\varepsilon-parameterized dynamics agree with the original tt-parameterized dynamics (up to reparameterization):

f˙∣Σ0=lf˙=t˙f˙i.e.,H^=0⟹f˙≃lf˙=t˙f˙,(46)\left.\dot{f}\right|_{\Sigma_0}=l\dot{f}=\dot{t}\dot{f} \qquad\text{i.e.,}\qquad\widehat{\mathcal{H}}=0 \quad\Longrightarrow\quad\dot{f}\simeq l\dot{f}=\dot{t}\dot{f}, \tag*{(46)}

where we will use the shorthand notation "≃\simeq" to indicate relations that hold when restricting consideration to Σ0\Sigma_0. The above Eq. (45)-Eq. (46) imply that:

Remark 3.2. Σ0⊂R2n+2\Sigma_0\subset\mathbb{R}^{2n+2} is the invariant level set on which the extended dynamics universally reproduce the original, nonextended dynamics (up to the reparameterization prescribed by dt=ldε\mathrm{d}t=l\mathrm{d}\varepsilon).

Remark 3.3. If a function f(x,π,t)f(x,\pi,t) is a constant of motion of the nonextended dynamics, then it is also a constant of motion of the extended dynamics restricted to Σ0\Sigma_0. 18

f˙=0⟹f˙∣Σ0=lf˙=0.(47)\dot{f}=0 \qquad\Longrightarrow\qquad\left.\dot{f}\right|_{\Sigma_0}=l\dot{f}=0. \tag*{(47)}

The word “universally” in Remark 3.2 warrants clarification. For a particular function ff, the term 1lH^{f,l}\frac{1}{l}\widehat{\mathcal{H}}\{f,l\} in Eq. (45) may also vanish if {f,l}=0\{f,l\}=0, which can happen even off of restriction to Σ0\Sigma_0 (i.e., even if H^≠0\widehat{\mathcal{H}}\ne0). For example, if ll is independent of (x,π)(x,\pi) — including the case l=l(t)l=l(t) — all such correction terms vanish on every level set. Therefore, H^=0\widehat{\mathcal{H}}=0 is always sufficient, but not always necessary. Yet, it often is necessary in the case of a nontrivial conformal scaling function l(x,π,t)l(x,\pi,t) for which {f,l}≠0\{f,l\}\ne0 for general observables ff.

Equations of motion on Σ0\Sigma_0. To illustrate Remark 3.2, we will show that the full ε\varepsilon-parameterized extended dynamics for (x^,π^)∈R2n+2(\widehat{x},\widehat{\pi})\in\mathbb{R}^{2n+2} in Eq. (21) do not necessarily agree with the original tt-parameterized nonextended dynamics for (x,π)∈R2n(x,\pi)\in\mathbb{R}^{2n} in Eq. (3), but that they do agree when the former is restricted to Σ0\Sigma_0.

Let ziz_i denote any extended coordinate other than πt\pi_t and note that

∂H^∂zi=∂∂zi(l(H+πt))=l∂H∂zi+(H+πt)∂l∂zi=l∂H∂zi+H^l∂l∂zi⇒H^=0∂H^∂zi≃l∂H∂zi,(48)\frac{\partial\widehat{H}}{\partial z_i}=\frac{\partial}{\partial z_i}\left(l(H+\pi_t)\right)=l\frac{\partial H}{\partial z_i}+(H+\pi_t)\frac{\partial l}{\partial z_i}=l\frac{\partial H}{\partial z_i}+\frac{\widehat{H}}{l}\frac{\partial l}{\partial z_i}\xRightarrow{\widehat{H}=0}\frac{\partial\widehat{H}}{\partial z_i}\simeq l\frac{\partial H}{\partial z_i}, \tag*{(48)}

such that, for the 2n2n coordinates (x,π)(\mathbf{x},\boldsymbol{\pi}), the extended dynamics in Eq. (21) are equivalently:

x˙=∂πH^−lαπ=l(∂πH−απ)+H^l∂πl=lx˙+H^l∂πlπ˙=−(∂xH^−lαx)=−l(∂xH−αx)−H^l∂xl=lπ˙−H^l∂xl(49)\begin{aligned} \dot{\mathbf{x}}&=\partial_{\boldsymbol{\pi}}\widehat{H}-l\boldsymbol{\alpha}_{\pi} =l\left(\partial_{\boldsymbol{\pi}}H-\boldsymbol{\alpha}_{\pi}\right)+\frac{\widehat{H}}{l}\partial_{\boldsymbol{\pi}}l =l\dot{\mathbf{x}}+\frac{\widehat{H}}{l}\partial_{\boldsymbol{\pi}}l\\ \dot{\boldsymbol{\pi}}&=-\left(\partial_{\mathbf{x}}\widehat{H}-l\boldsymbol{\alpha}_{x}\right) =-l\left(\partial_{\mathbf{x}}H-\boldsymbol{\alpha}_{x}\right)-\frac{\widehat{H}}{l}\partial_{\mathbf{x}}l =l\dot{\boldsymbol{\pi}}-\frac{\widehat{H}}{l}\partial_{\mathbf{x}}l \tag*{(49)} \end{aligned}

with (x˙,π˙)(\dot{\mathbf{x}},\dot{\boldsymbol{\pi}}) being the original tt-parameterized nonextended dynamics from Eq. (3). Consequently, these same dynamics are recovered if — but not quite only if19 — the extended dynamics are restricted to Σ0\Sigma_0,

H^=0⇒x˙≃l(∂πH−απ)=lx˙=t˙x˙,π˙≃l(−∂xH+αx)=lπ˙=t˙π˙.(50)\widehat{H}=0\quad\Rightarrow\quad \begin{aligned} \dot{\mathbf{x}}&\simeq l\left(\partial_{\boldsymbol{\pi}}H-\boldsymbol{\alpha}_{\pi}\right)=l\dot{\mathbf{x}}=\dot{t}\dot{\mathbf{x}},\\ \dot{\boldsymbol{\pi}}&\simeq l\left(-\partial_{\mathbf{x}}H+\boldsymbol{\alpha}_{x}\right)=l\dot{\boldsymbol{\pi}}=\dot{t}\dot{\boldsymbol{\pi}}. \end{aligned} \tag*{(50)}

The remaining extended equation, the ODE for π˙t\dot{\pi}_t in Eq. (21), also reduces consistently. Using Eq. (48), this equation is equivalently written as

π˙t=−∂tH^+At=−l∂tH+At−H^l∂tl,(51)\dot{\pi}_t=-\partial_t\widehat{H}+A_t =-l\partial_tH+A_t-\frac{\widehat{H}}{l}\partial_tl, \tag*{(51)}

with: At:=−(αx⋅∂πH^−απ⋅∂xH^)A_t:=-(\boldsymbol{\alpha}_x\cdot\partial_{\boldsymbol{\pi}}\widehat{H}-\boldsymbol{\alpha}_{\pi}\cdot\partial_{\mathbf{x}}\widehat{H}) =−l(αx⋅∂πH−απ⋅∂xH)−H^l(αx⋅∂πl−απ⋅∂xl)=-l(\boldsymbol{\alpha}_x\cdot\partial_{\boldsymbol{\pi}}H-\boldsymbol{\alpha}_{\pi}\cdot\partial_{\mathbf{x}}H)-\frac{\widehat{H}}{l}(\boldsymbol{\alpha}_x\cdot\partial_{\boldsymbol{\pi}}l-\boldsymbol{\alpha}_{\pi}\cdot\partial_{\mathbf{x}}l)

such that restriction to Σ0\Sigma_0 yields:

H^=0⇒π˙t≃−l(∂tH−αt),with: αt:=−(αx⋅∂πH−απ⋅∂xH)(52)\widehat{H}=0\quad\Rightarrow\quad\dot{\pi}_t\simeq-l(\partial_tH-\alpha_t), \qquad\text{with: }\alpha_t:=-(\boldsymbol{\alpha}_x\cdot\partial_{\boldsymbol{\pi}}H-\boldsymbol{\alpha}_{\pi}\cdot\partial_{\mathbf{x}}H) \tag*{(52)}

(equivalently: αt=−Fnc⋅{r,H}=−Fnc⋅(r˙−∂tr)\alpha_t=-\mathbf{F}^{\mathrm{nc}}\cdot\{\mathbf{r},H\}=-\mathbf{F}^{\mathrm{nc}}\cdot(\dot{\mathbf{r}}-\partial_t\mathbf{r})).20 Note from the above that, on Σ0\Sigma_0, the ODE for π˙t\dot{\pi}_t precisely reproduces (up to reparameterization) the governing equation for −H-H on the nonextended space from Eq. (5):

π˙t≃−l(∂tH+αx⋅∂πH−απ⋅∂xH)=−lH˙.(53)\dot{\pi}_t\simeq-l(\partial_tH+\boldsymbol{\alpha}_x\cdot\partial_{\boldsymbol{\pi}}H-\boldsymbol{\alpha}_{\pi}\cdot\partial_{\mathbf{x}}H)=-l\dot{H}. \tag*{(53)}

Altogether, the complete extended dynamics from Eq. (21), and their equivalent expressions when restricted to Σ0\Sigma_0, are summarized as:

x˙=∂πH^−Aπ≃l(∂πH−απ)=lx˙π˙=−(∂xH^−Ax)≃−l(∂xH−αx)=lπ˙t˙=∂πtH^=l=(no change)=lt˙π˙t=−(∂tH^−At)≃−l(∂tH−αt)=−lH˙(54)\begin{aligned} \dot{\mathbf{x}}&=\partial_{\boldsymbol{\pi}}\widehat{H}-\mathbf{A}_{\pi}\simeq l(\partial_{\boldsymbol{\pi}}H-\boldsymbol{\alpha}_{\pi})=l\dot{\mathbf{x}}\\ \dot{\boldsymbol{\pi}}&=-\left(\partial_{\mathbf{x}}\widehat{H}-\mathbf{A}_{x}\right)\simeq-l(\partial_{\mathbf{x}}H-\boldsymbol{\alpha}_{x})=l\dot{\boldsymbol{\pi}}\\ \dot{t}&=\partial_{\pi_t}\widehat{H}=l=(\text{no change})=l\dot{t}\\ \dot{\pi}_t&=-\left(\partial_t\widehat{H}-A_t\right)\simeq-l(\partial_tH-\alpha_t)=-l\dot{H} \tag*{(54)} \end{aligned}

where “≃\simeq” indicates relations that hold when restricting consideration to Σ0\Sigma_0. That is,

(x˙,t˙,π˙,π˙t)∣Σ0=l⋅(x˙,1,π˙,−H˙)=dtdτ⋅(x˙,1,π˙,−H˙).(55)\left(\dot{\mathbf{x}},\dot{t},\dot{\boldsymbol{\pi}},\dot{\pi}_t\right)\big|_{\Sigma_0} =l\cdot\left(\dot{\mathbf{x}},1,\dot{\boldsymbol{\pi}},-\dot{H}\right) =\frac{dt}{d\tau}\cdot\left(\dot{\mathbf{x}},1,\dot{\boldsymbol{\pi}},-\dot{H}\right). \tag*{(55)}

Note the 2n2n-dim “(x,π)(\mathbf{x},\boldsymbol{\pi})-part” of Eq. (54) is precisely the directly reparameterized, conformally Hamiltonian system in Eq. (14).

Keeping track of Hamiltonian structure.

We have now encountered four closely related forms of the equations of motion for a dynamical system described by a Hamiltonian function H(x,π,t)H(\mathbf{x},\boldsymbol{\pi},t): (1) the original nonextended dynamics in Eq. (3), (2) their directly reparameterized form in Eq. (14), (3) the extended dynamics in Eq. (21), and (4) the restriction of the extended dynamics to the zero level set Σ0={H^=0}\Sigma_0=\{\widehat{H}=0\} in Eq. (54). We pause to summarize their structural differences since they are not all canonical Hamiltonian systems.

To more clearly track — without the separate complication introduced by nonconservative forces — which aspects of the Hamiltonian structure are retained or lost among these similar presentations of the equations of motion, we momentarily restrict consideration to the conservative case, Fnc=0\mathbf{F}^{\mathrm{nc}} = 0, so that all nonconservative generalized forces vanish.21

  1. The original tt-parameterized nonextended dynamics on R2n\mathbb{R}^{2n} (Eq. (3)):

x˙=∂πH,π˙=−∂xH,(56)\dot{\mathbf{x}} = \partial_{\boldsymbol{\pi}} \mathcal{H}, \qquad\dot{\boldsymbol{\pi}} = -\partial_{\mathbf{x}} \mathcal{H}, \tag*{(56)}

are Hamiltonian equations in standard canonical form on the original 2n2n-dim phase space.

  1. The ε\varepsilon-parameterized nonextended dynamics on R2n\mathbb{R}^{2n} by direct conformal scaling (Eq. (14)):

x˚=l∂πH,π˚=−l∂xH,\mathring{\mathbf{x}} = l\partial_{\boldsymbol{\pi}}\mathcal{H}, \qquad\mathring{\boldsymbol{\pi}} = -l\partial_{\mathbf{x}}\mathcal{H},

with t˚=l\mathring{t} = l {#eq:3.39}

are obtained simply by scaling the original ODEs by l=dt/dεl = \mathrm{d}t/\mathrm{d}\varepsilon. For a general scaling function l=l(x,π,t)l = l(\mathbf{x}, \boldsymbol{\pi}, t), the resulting equations above are only conformally Hamiltonian but are generally not themselves Hamiltonian with respect to the original canonical symplectic structure. Yet, certain special choices of ll may nevertheless preserve Hamiltonian form, as discussed around Eq. (19).

If tt is to be recovered in the solution, its ODE t˚=l\mathring{t} = l must also be included, giving a (2n+1)(2n+1)-dim system.

  1. The ε\varepsilon-parameterized extended dynamics on R2n+2\mathbb{R}^{2n+2} (Eq. (21)), generated by H^=l(H+πt)\widehat{\mathcal{H}} = l(\mathcal{H} + \pi_t):

x˚=∂πH^,π˚=−∂xH^,t˚=∂πtH^=l,π˚t=−∂tH^,(57)\mathring{\mathbf{x}} = \partial_{\boldsymbol{\pi}}\widehat{\mathcal{H}}, \qquad\mathring{\boldsymbol{\pi}} = -\partial_{\mathbf{x}}\widehat{\mathcal{H}}, \qquad\mathring{t} = \partial_{\pi_t}\widehat{\mathcal{H}} = l, \qquad\mathring{\pi}_t = -\partial_t\widehat{\mathcal{H}}, \tag*{(57)}

are equivalently written in the familiar canonical form,

x^˚=∂π^H^,π^˚=−∂x^H^.(58)\mathring{\widehat{\mathbf{x}}} = \partial_{\widehat{\boldsymbol{\pi}}}\widehat{\mathcal{H}}, \qquad\mathring{\widehat{\boldsymbol{\pi}}} = -\partial_{\widehat{\mathbf{x}}}\widehat{\mathcal{H}}. \tag*{(58)}

Unlike the direct conformal scaling in Eq. (3.39), the above are canonical Hamiltonian equations on the full (2n+2)(2n+2)-dim extended phase space. The additional coordinate pair (t,πt)(t,\pi_t) therefore allows the evolution parameter transformation to be implemented while retaining canonical Hamiltonian structure fully analogous to Eq. (56).

  1. The ε\varepsilon-parameterized extended dynamics restricted to Σ0⊂R2n+2\Sigma_0 \subset\mathbb{R}^{2n+2} (Eq. (54)):

x˚=∂πH^≃l∂πH,π˚=−∂xH^≃−l∂xH,t˚=∂πtH^=l,π˚t=−∂tH^≃−l∂tH,(59)\mathring{\mathbf{x}} = \partial_{\boldsymbol{\pi}}\widehat{\mathcal{H}} \simeq l\partial_{\boldsymbol{\pi}}\mathcal{H}, \qquad\mathring{\boldsymbol{\pi}} = -\partial_{\mathbf{x}}\widehat{\mathcal{H}} \simeq-l\partial_{\mathbf{x}}\mathcal{H}, \qquad\mathring{t} = \partial_{\pi_t}\widehat{\mathcal{H}} = l, \qquad\mathring{\pi}_t = -\partial_t\widehat{\mathcal{H}} \simeq-l\partial_t\mathcal{H}, \tag*{(59)}

are not actually a fourth, separate formulation; they are the equivalent form of the full extended dynamics in Eq. (58) when restricted to their invariant level set Σ0\Sigma_0 — that is, simplifying the ODEs using H^=0\widehat{\mathcal{H}} = 0 — on which the physically relevant extended solution curves reside. However, Σ0\Sigma_0 has odd dimension 2n+12n+1 and therefore cannot itself carry a symplectic structure: the canonical symplectic structure belongs to the ambient (2n+2)(2n+2)-dim extended phase space and becomes degenerate when restricted to Σ0\Sigma_0. Correspondingly, the above 2n+22n+2 ODEs evolve subject to the conserved relation πt=−H\pi_t = -\mathcal{H} (i.e., H^=0\widehat{\mathcal{H}} = 0), so only 2n+12n+1 coordinates are independent on Σ0\Sigma_0.

We reiterate that such restriction to Σ0\Sigma_0 does not imply any physical constraints; nothing need be done to “enforce” the restriction aside from limit consideration of initial conditions to those that satisfy H^(x0,π0)=0\widehat{\mathcal{H}}(\mathbf{x}_0,\boldsymbol{\pi}_0) = 0. Equivalently, for any given (x0,π0,t0)(\mathbf{x}_0,\boldsymbol{\pi}_0,t_0), one simply selects (πt)0=−H(x0,π0,t0)(\pi_t)_0 = -\mathcal{H}(\mathbf{x}_0,\boldsymbol{\pi}_0,t_0). This is not a constraint relation; it is a conservation property. It imposes no restriction on the nonextended phase space. The dynamics on Σ0\Sigma_0 are merely a description of the physical dynamics on the subspace where they naturally evolve.

Thus, direct conformal scaling provides the simplest transformation of the evolution parameter but generally sacrifices canonical Hamiltonian structure. Extension to R2n+2\mathbb{R}^{2n+2} with the additional coordinate pair (t,πt)(t,\pi_t) retains that structure on the full extended phase space. Subsequent restriction to the surface Σ0⊂R2n+2\Sigma_0 \subset\mathbb{R}^{2n+2} singles out the extended solution curves that reproduce the original physical dynamics. The restricted equations remain the restriction of the ambient extended Hamiltonian system.

Appendix

Relation to the Lagrangian & the Legendre transformation.

The preceding Hamiltonian developments need not be predicated on an accompanying Lagrangian description. Still, to complete the picture, we will briefly show how the extended Hamiltonian and momentum coordinate conjugate to tt are recovered by considering an extended Lagrangian and the corresponding Legendre transformation. We refer the reader to [6, 7, 8] for a more complete treatment.

Nonconservative forces play no part in the relations considered here and will therefore not appear.

In contrast to the arbitrary symplectic coordinates of preceding developments, we now specialize to typical configuration coordinates q∈Rn\mathbf{q} \in\mathbb{R}^{n} (in the physical sense22) with corresponding generalized velocity coordinates v∈Rn\mathbf{v} \in\mathbb{R}^{n}. I.e., vt=q˙t\mathbf{v}_{t}=\dot{\mathbf{q}}_{t} when evaluated along a curve qt≡q(t)\mathbf{q}_{t}\equiv\mathbf{q}(t). The corresponding conjugate momenta will be denoted p∈Rn\mathbf{p}\in\mathbb{R}^{n}.

Extended Lagrangian. Let L(q,v,t)L(\mathbf{q},\mathbf{v},t) be the usual nonextended Lagrangian function for the system of interest; we assume it permits such a description and also that the associated nonextended Legendre transformation is regular. As before, suppose we wish to exchange tt for some other evolution parameter ε\varepsilon. We begin by reparameterizing the Lagrangian action integral:

IL=∫t0tfL dt=∫ε0εft˙L dε.(60)I_{L}=\int_{t_{0}}^{t_{f}}L\,\mathrm{d}t=\int_{\varepsilon_{0}}^{\varepsilon_{f}}\dot{t}L\,\mathrm{d}\varepsilon. \tag*{(60)}

where we have used only the kinematic relation dt=t˙dε\mathrm{d}t=\dot{t}\mathrm{d}\varepsilon. The above motivates defining an extended Lagrangian function L^\widehat{L} of extended configuration and velocity coordinates (q^,u^)∈R2n+2(\widehat{\mathbf{q}},\widehat{\mathbf{u}})\in\mathbb{R}^{2n+2}:

L^(q^,u^):=utL,q^:=(q,t),u^:=(u,ut),(61)\widehat{L}(\widehat{\mathbf{q}},\widehat{\mathbf{u}}):=u_{t}L,\qquad\widehat{\mathbf{q}}:=(\mathbf{q},t),\qquad\widehat{\mathbf{u}}:=(\mathbf{u},u_{t}), \tag*{(61)}

with u^=(u,ut)\widehat{\mathbf{u}}=(\mathbf{u},u_{t}) related to the usual nonextended velocity coordinates v\mathbf{v} as follows (for ut≠0u_{t}\ne0):

v=1utu⟺u=utv.(62)\mathbf{v}=\frac{1}{u_{t}}\mathbf{u}\quad\Longleftrightarrow\quad\mathbf{u}=u_{t}\mathbf{v}. \tag*{(62)}

The extended velocity coordinates u^=(u,ut)\widehat{\mathbf{u}}=(\mathbf{u},u_{t}) are understood as follows: when evaluated along any extended configuration coordinate curve, q^ε=(qε,tε)\widehat{\mathbf{q}}_{\varepsilon}=(\mathbf{q}_{\varepsilon},t_{\varepsilon}), they satisfy:

along a curve:u^ε=q^˙εi.e.,uε=q˙ε=t˙εq˙t,(ut)ε=t˙ε.(63)\text{along a curve:}\qquad\widehat{\mathbf{u}}_{\varepsilon}=\dot{\widehat{\mathbf{q}}}_{\varepsilon}\qquad\text{i.e.,}\qquad\mathbf{u}_{\varepsilon}=\dot{\mathbf{q}}_{\varepsilon}=\dot{t}_{\varepsilon}\dot{\mathbf{q}}_{t},\qquad(u_{t})_{\varepsilon}=\dot{t}_{\varepsilon}. \tag*{(63)}

Thus, along such curves, L^\widehat{L} in Eq. (A.2) reproduces the integrand of the action integral in Eq. (A.1).

Note. For L^:=utL\widehat{L}:=u_{t}L given in Eq. (A.2), the original L(q,v,t)L(\mathbf{q},\mathbf{v},t) is regarded as a function L(q,v(u^),t)L(\mathbf{q},\mathbf{v}(\widehat{\mathbf{u}}),t) of (q^,u^)(\widehat{\mathbf{q}},\widehat{\mathbf{u}})—that is, the original expression for L(q,v,t)L(\mathbf{q},\mathbf{v},t) rewritten using direct substitution of v=u/ut\mathbf{v}=\mathbf{u}/u_{t}.

Extended Legendre transformation. The Legendre transformation associated with the extended Lagrangian L^\widehat{L} proceeds much as it does in the nonextended setting. We introduce extended momentum coordinates p^=(p,pt)∈Rn+1\widehat{\mathbf{p}}=(\mathbf{p},p_{t})\in\mathbb{R}^{n+1} using a familiar definition:

p^:=∂u^L^.(64)\widehat{\mathbf{p}}:=\partial_{\widehat{\mathbf{u}}}\widehat{L}. \tag*{(64)}

The first nn components, p:=∂uL^∈Rn\mathbf{p}:=\partial_{\mathbf{u}}\widehat{L}\in\mathbb{R}^{n}, coincide with the usual nonextended relation p=∂vL\mathbf{p}=\partial_{\mathbf{v}}L.23

p:=∂uL^=∂vL.(65)\mathbf{p}:=\partial_{\mathbf{u}}\widehat{L}=\partial_{\mathbf{v}}L. \tag*{(65)}

Eq. (A.5) also defines the momentum coordinate conjugate to tt as pt:=∂utL^p_{t}:=\partial_{u_{t}}\widehat{L}, which turns out to be −H-H.24

pt:=∂utL^=L−∂vL⋅v=L−p⋅v=−H,(66)p_{t}:=\partial_{u_{t}}\widehat{L}=L-\partial_{\mathbf{v}}L\cdot\mathbf{v}=L-\mathbf{p}\cdot\mathbf{v}=-H, \tag*{(66)}

with H:=p⋅v−LH:=\mathbf{p}\cdot\mathbf{v}-L the usual nonextended Hamiltonian, with v\mathbf{v} expressed in terms of (q,p,t)(\mathbf{q},\mathbf{p},t) by inversion of Eq. (A.6) (assumed possible).

Next, in analogy to the usual relation H=p⋅v−LH=\mathbf{p}\cdot\mathbf{v}-L, consider the analogous extended expression p^⋅u^−L^\widehat{\mathbf{p}}\cdot\widehat{\mathbf{u}}-\widehat{L}:

p^⋅u^−L^=p⋅u+ptut−utL=ut(p⋅v−L+pt)=ut(H+pt).(67)\widehat{\mathbf{p}}\cdot\widehat{\mathbf{u}}-\widehat{L}=\mathbf{p}\cdot\mathbf{u}+p_{t}u_{t}-u_{t}L=u_{t}(\mathbf{p}\cdot\mathbf{v}-L+p_{t})=u_{t}(H+p_{t}). \tag*{(67)}

This would recover the definition of the extended Hamiltonian, H^:=ι(H+pt)\widehat{H}:=\iota(H+p_{t}) from § 4.1, if an identification can be made between the velocity coordinate utu_{t} and the conformal scaling function ι(q,p,t)\iota(\mathbf{q},\mathbf{p},t). Such an identification can indeed be made, though it does not arise directly from the extended Legendre transformation, which is nonregular (even if the original nonextended Legendre transformation is regular).25 The velocity coordinate utu_{t} cannot be solved for in terms of the momenta. However, along any extended configuration curve it holds by definition that ut=t˙=dtdεu_{t}=\dot{t}=\frac{\mathrm{d}t}{\mathrm{d}\varepsilon}, as per Eq. (A.4). An evolution parameter transformation prescribed by dt=ιdε\mathrm{d}t=\iota\mathrm{d}\varepsilon (as considered previously) therefore gives the clear choice

pt:=∂utL^=∂ut(utL)=L+ut∂utL=L+ut∂vL⋅∂utv=L+ut∂vL⋅(−ut−1v)=L−∂vL⋅v=L−p⋅v.p_{t}:=\partial_{u_{t}}\widehat{L}=\partial_{u_{t}}(u_{t}L)=L+u_{t}\partial_{u_{t}}L=L+u_{t}\partial_{\mathbf{v}}L\cdot\partial_{u_{t}}\mathbf{v}=L+u_{t}\partial_{\mathbf{v}}L\cdot(-u_{t}^{-1}\mathbf{v})=L-\partial_{\mathbf{v}}L\cdot\mathbf{v}=L-\mathbf{p}\cdot\mathbf{v}.

where the final equality has made use of Eq. (A.6). of this otherwise undetermined relation to be ut=lu_t = l. Substitution of this choice into Eq. (A.8) defines the corresponding extended Hamiltonian as:

H^(q^,p^):=l(H+pt).(68)\widehat{H}(\widehat{q},\widehat{p}) := l(H + p_t). \tag*{(68)}

Thus, the extended Lagrangian and its Legendre transformation independently recover the relation pt=−Hp_t = -H associated with the zero level set and, upon imposing the prescribed evolution-parameter transformation, also yield the extended Hamiltonian H^=l(H+pt)\widehat{H} = l(H + p_t).

References

References

  1. [1]C. Lanczos, The Variational Principles of Mechanics. New York: Dover Publications, 4th ed., 1986.
  2. [2]E. T. Whittaker, A Treatise on the Analytical Dynamics of Particles and Rigid Bodies: With an Introduction to the Problem of Three Bodies. Cambridge: Cambridge University Press, 4 ed., 1937.
  3. [3]D. T. Greenwood, Advanced Dynamics. Cambridge University Press, 2003.
  4. [4]J. G. Papastavridis, Analytical Mechanics: A Comprehensive Treatise on the Dynamics of Constrained Systems. Singapore: World Scientific Publishing Co., reprint ed., 2014. Reprint of the 2002 edition published by Oxford University Press.
  5. [5]D. T. Greenwood, Classical Dynamics. Mineola, NY: Dover Publications, 1997. Unabridged and corrected republication of the work first published by Prentice-Hall, Inc., Englewood Cliffs, N.J., in 1977.
  6. [6]O. D. Johns, Analytical Mechanics for Relativity and Quantum Mechanics. Oxford Graduate Texts, Oxford; New York: Oxford University Press, 2nd ed., 2011.
  7. [7]J. Struckmeier, “Hamiltonian dynamics on the symplectic extended phase space for autonomous and non-autonomous systems,” Journal of Physics A: Mathematical and General, vol. 38, no. 6, p. 1257, 2005.DOI
  8. [8]J. Struckmeier, “Extended hamilton–lagrange formalism and its application to feynman’s path integral for relativistic quantum physics,” International Journal of Modern Physics E, vol. 18, no. 01, pp. 79–108, 2009.DOI
  9. [9]H. Sharma, M. Patil, and C. Woolsey, “Energy-preserving variational integrators for forced lagrangian systems,” Communications in Nonlinear Science and Numerical Simulation, vol. 64, pp. 159–177, 2018.DOI
  10. [10]A. Echeverría-Enríquez, M. C. Muñoz-Lecanda, and N. Román-Roy, “Geometrical setting of time-dependent regular systems: Alternative models,” Reviews in Mathematical Physics, vol. 3, no. 3, pp. 301–330, 1991.
  11. [11]M. de León and C. Sardón, “Cosymplectic and contact structures to resolve time-dependent and dissipative hamiltonian systems,” Journal of Physics A: Mathematical and Theoretical, vol. 50, no. 25, p. 255205, 2017.
  12. [12]M. C. Muñoz-Lecanda and N. Román-Roy, Geometry of Mechanics. Advanced Textbooks in Mathematics, World Scientific Publishing Company, 2025.
  13. [13]J. F. Cariñena, E. Martínez, and M. C. Muñoz-Lecanda, “Infinitesimal time reparametrisation and its applications,” Journal of Nonlinear Mathematical Physics, vol. 29, no. 3, pp. 523–555, 2022.DOI
  14. [14]J. F. Cariñena, L. A. Ibort, and E. A. Lacomba, “Time scaling as an infinitesimal canonical transformation,” Celestial Mechanics, vol. 42, pp. 201–213, 1988.
  15. [15]J. F. Cariñena and M. F. Rañada, “Hamiltonian formalism and time transformations in the extended phase space T*(Q × R),” Reports on Mathematical Physics, vol. 30, no. 1, pp. 89–106, 1991.DOI
  16. [16]E. L. Stiefel and G. Scheifele, Linear and Regular Celestial Mechanics: Perturbed Two-Body Motion, Numerical Methods, Canonical Theory. Berlin and New York: Springer-Verlag, 1971.
  17. [17]V. R. Bond and G. Janin, “Canonical orbital elements in terms of an arbitrary independent variable,” Celestial mechanics, vol. 23, no. 2, pp. 159–172, 1981.DOI
  18. [18]J. T. Peterson, M. Majji, and J. L. Junkins, “Projective transformations for linearized and regularized central-force dynamics: Hamiltonian formulation,” Celestial Mechanics and Dynamical Astronomy, vol. 138, no. 5, p. 54, 2026.DOI
  19. [19]H. Schaub and J. L. Junkins, Analytical Mechanics of Space Systems. Reston, VA: AIAA Education Series, 5th ed., 2026.
  20. [20]F. T. Tveter, “Hamilton’s equations of motion for non-conservative systems,” Celestial Mechanics and Dynamical Astronomy, vol. 60, no. 4, pp. 409–419, 1994.DOI
  21. [21]F. T. Tveter, “Deriving the hamilton equations of motion for a nonconservative system using a variational principle,” Journal of Mathematical Physics, vol. 39, pp. 1495–1500, 03 1998.DOI

Paper details

vlnc:W_83d8jswz
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 10

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents