betaFeedback

WorksPhysics and Astronomy

Hamilton Revised: The Action Principle for Initial Value Problems

Abstract

We present the variational action principle for initial value problems in classical, conservative-force point particle mechanics. We rigorously derive this formulation by taking the classical limit of the Schwinger-Keldysh expression for the time dependence of the expectation value for operators in quantum mechanics. We clarify the connection between the variation of the position and the variation of the velocity of a particle when implementing Hamilton's Principle in deriving the Euler-Lagrange Equations. We show that both the plus and minus Keldysh paths (of the average and difference of the forward/backward paths) have classical paths and fluctuations – unlike the common perception that the minus path provides the fluctuations around the single classical solution given by the plus path – and that the fluctuations of both paths are crucial for the correct normalization of the classical limit. The classical limit yields "initial conditions" and equations of motion for the minus paths such that the unique classical solution for the minus paths is that they are identically zero, and, fascinatingly, that the minus paths' solution propagates backwards in time; thus one does not need to set the minus paths to zero by hand when taking the classical limit of the Schwinger-Keldysh formalism. We note implications for the classical and quantum mechanics of non-holonomic constraints and quantum field theories with gauges dependent on the derivatives of the fields.

Introduction

The field of classical point particle mechanics concerns itself with the deterministic prediction of the future position of a particle given a particular physical setup. The setup will in general set the initial position and velocity of the particle and provide some notion of the forces that will act on the particle. Classical mechanics is, in principle, fully understood: Newton’s Second Law [1]

∑iF⃗i(t)=md2xdt2∣t(1)\sum_i \vec{F}_i(t) = m \left.\frac{d^2 x}{dt^2}\right|_t \tag*{(1)}

gives a second order differential equation in time for the position x(t)x(t) of a particle of mass mm subject to external forces F⃗i(t)\vec{F}_i(t). Given the initial position x(ti)=x0x(t_i) = x_0 and initial velocity dxdt∣ti=v0\left.\frac{dx}{dt}\right|_{t_i} = v_0, the Picard-Lindelöf Theorem [2] guarantees that a solution of the second order differential equation exists and is unique.

Despite its well-deserved fame and utility, Newton’s Second Law has a number of drawbacks. Centuries of work led to the understanding that, in many ways, a far better formulation of classical point particle mechanics can be found in terms of Hamilton’s Principle of Extremized Action [3, 4], that the path of the classical point particle xcl(t)x_{cl}(t) satisfies

δS∣xcl(t)=0,(2)\left.\delta S\right|_{x_{cl}(t)} = 0, \tag*{(2)}

where the action

S≡∫dt L(3)S \equiv\int dt\,L \tag*{(3)}

is defined in terms of the Lagrangian

L≡T−V,(4)L \equiv T - V, \tag*{(4)}

where TT is the kinetic energy and VV is the potential energy, and δS\delta S is the first variation of the action. Hamilton’s formulation has enormous advantages over Newton’s [5, 6]. A proper subset of these advantages includes

  • making the role of symmetries transparent via Noether’s Theorem [7]

  • formulating the problem in terms of scalar, rather than vector, quantities

  • allowing for the easy use of generalized coordinates

  • readily incorporating constraints (whose explicit forces do not need to be known a priori)

  • easily generalizing to continuous distributions of masses and to fields

  • leading to a phase space manifold with symplectic structure, providing a “royal road” to quantum mechanics

  • allowing for global rather than local solutions of classical paths [8]

Despite these enormous advantages, as well as ∼200\sim200 years of study, there are still several conceptual and practical open questions related to Hamilton’s formulation. To understand the conceptual issues, let us briefly review the usual derivation of the Euler-Lagrange equations from Hamilton’s Principle [5, 9]. We start with a Lagrangian L[q⃗(t),q⃗˙(t)]L[\vec{q}(t),\dot{\vec{q}}(t)] that depends on the path of the particle in generalized coordinates q⃗(t)\vec{q}(t) and the time derivative of the generalized coordinates q⃗˙(t)≡dq⃗dt∣t\dot{\vec{q}}(t) \equiv\left.\frac{d\vec{q}}{dt}\right|_{t}. Then the variation of the action yields

δS=∫titfdt[∂L∂qiδqi(t)+∂L∂q˙iδq˙i(t)].(5)\delta S = \int_{t_i}^{t_f} dt \left[\frac{\partial L}{\partial q^i}\delta q^i(t) + \frac{\partial L}{\partial\dot{q}^i}\delta\dot{q}^i(t)\right]. \tag*{(5)}

For a particle undergoing unconstrained motion, one has the transposition rule [10]

δq˙i(t)=ddt[δqi(t)].(6)\delta\dot{q}^i(t) = \frac{d}{dt}\left[\delta q^i(t)\right]. \tag*{(6)}

One then integrates by parts, which yields

δS=δqi∂L∂q˙i∣titf+∫titfdt δqi[∂L∂qi−ddt∂L∂q˙i].(7)\delta S = \left.\delta q^i\frac{\partial L}{\partial\dot{q}^i}\right|_{t_i}^{t_f} + \int_{t_i}^{t_f} dt\,\delta q^i\left[\frac{\partial L}{\partial q^i} - \frac{d}{dt}\frac{\partial L}{\partial\dot{q}^i}\right]. \tag*{(7)}

The endpoints of the path are taken to be fixed, in which case δqi(ti)=δqi(tf)=0\delta q^i(t_i) = \delta q^i(t_f) = 0, and the boundary terms are identically 0. Then, because the variations in the path δqi(t)\delta q^i(t) are independent and arbitrary, in order for the variation to be identically zero, we must have by the Fundamental Theorem of the Calculus of Variations [11] the Euler-Lagrange Equations,

∂L∂qi−ddt∂L∂q˙i=0.(8)\frac{\partial L}{\partial q^i} - \frac{d}{dt}\frac{\partial L}{\partial\dot{q}^i} = 0. \tag*{(8)}

One then solves the Euler-Lagrange Equations as an initial value problem, in which qi(ti)=q0iq^i(t_i) = q^i_0 and q˙i(ti)=q˙0i\dot{q}^i(t_i) = \dot{q}^i_0.

There are two main conceptual issues with the above derivation that we wish to highlight. The first conceptual issue is the imposition of boundary value problem conditions on the variation of the path at the initial and final times, δqi(ti)=δqi(tf)=0\delta q^i(t_i)=\delta q^i(t_f)=0 in the derivation of the Euler-Lagrange Equations, which are then posed and solved for the particle’s path qi(t)q^i(t) in terms of an initial value problem. Let us expand on this point. Hamilton’s Principle in the above derivation asks: if we know the initial position at the initial time of our point particle and a priori know the final position of the particle at the final time of interest, what is the path the particle takes? In contrast, in a realistic experiment, generally speaking it is only the initial conditions of the initial position and initial velocity at the initial time that are known. To know the final location of the particle at some future time tft_f is to possess acausal information, a fundamental violation of Einstein’s Special Relativity [12]. Further, unlike initial value problems (which are well-posed, in the sense that the Picard-Lindelöf Theorem [2] guarantees the existence and uniqueness of the solution), boundary value problems can have one, zero, or many solutions [2], contrary to our expectation and experience that experimental measurements can be made and are reproducible [13].

The second conceptual issue is as follows. We know that the velocity of the particle is related to the position of the particle through time differentiation,

q˙i(t)=ddt[qi(t)].(9)\dot{q}^i(t)=\frac{d}{dt}\left[q^i(t)\right]. \tag*{(9)}

Why, then, do we treat the variations in the velocity of the particle δq˙i(t)\delta\dot{q}^i(t) as independent of the variations of the path of the particle δqi(t)\delta q^i(t) in the derivation of the Euler-Lagrange Equations? In a similar way, in the derivation of Hamilton’s Equations from Hamilton’s Principle, one treats the variations in the momentum δpi(t)\delta p_i(t) as independent of the variations in the paths δqi(t)\delta q^i(t), even though the definition of momentum

pi≡∂L∂q˙i(10)p_i\equiv\frac{\partial L}{\partial\dot{q}^i} \tag*{(10)}

implies that the momentum is a function of the path and the velocity, pi(qi(t),q˙i(t))p_i(q^i(t),\dot{q}^i(t)).

Related to this second conceptual issue is the confusion and open questions related to the treatment of constraints in Hamilton’s formulation. Holonomic constraints fk(qi(t),t)=0f_k(q^i(t),t)=0, such as constraining the particle to move on a circle x2(t)+y2(t)−R2=0x^2(t)+y^2(t)-R^2=0, depend only on the coordinates qi(t)q^i(t). Hamilton’s Principle in this case is modified such that the variations in the path must satisfy the constraint equations, fk(qi(t)+δqi(t),t)=0f_k(q^i(t)+\delta q^i(t),t)=0. The variations are not all independent and arbitrary, but one may still find solutions to the problem. In particular, the transposition rule Eq. (6) remains unchanged. However, for non-holonomic constraints gk(q⃗ i(t),q⃗˙ i(t),t)=0g_k(\vec{q}^{\,i}(t),\dot{\vec{q}}^{\,i}(t),t)=0, such as for the problem of rolling without slipping [10, 14], a general formulation of Hamilton’s Principle eluded physicists until only recently [8]. Further, for non-holonomic constraints, the transposition rule Eq. (6) appears to be modified [10], which therefore modifies the symplectic structure of phase space [15, 16]. Modifying this symplectic structure of phase space has unknown consequences for the quantum mechanics of systems with non-holonomic constraints, such as quantum nano-carts [17] and gauge theories under the ’t Hooft-Veltman gauge [18].

In addition to the above conceptual issues related to the derivation of the Euler-Lagrange Equations from Hamilton’s Principle is the limitation that Hamilton’s formulation of classical mechanics requires that forces be derived from potentials, which is to say that the system must be conservative; i.e. dissipative systems are outside the scope of the treatment. In a recent work [19], Galley proposed a new formulation of classical mechanics. In Galley’s Principle, one also considers the variation of an action, but with two crucial changes: first, the number of degrees of freedom for the system are doubled, and, second, a new term is added to the action, which can incorporate the effects of dissipation. In particular, Galley proposed that one should consider an action dependent on two paths, q⃗1(t)\vec{q}_1(t) and q⃗2(t)\vec{q}_2(t), and on a term K(q⃗a(t),q⃗˙a(t),t)K(\vec{q}_a(t),\dot{\vec{q}}_a(t),t) that depends on both paths {q⃗a}a=12\{\vec{q}_a\}_{a=1}^{2},

SG≡∫dt[L(q⃗1(t),q⃗˙1(t))−L(q⃗2(t),q⃗˙2(t))+K(q⃗a(t),q⃗˙a(t),t)].(11)S_G \equiv\int dt\left[L(\vec{q}_1(t),\dot{\vec{q}}_1(t))-L(\vec{q}_2(t),\dot{\vec{q}}_2(t))+K(\vec{q}_a(t),\dot{\vec{q}}_a(t),t)\right]. \tag*{(11)}

Utilizing the change of coordinates

q⃗+(t)≡12(q⃗1(t)+q⃗2(t))\vec{q}_{+}(t) \equiv\frac{1}{2}\left(\vec{q}_1(t)+\vec{q}_2(t)\right)

References

  1. [2]E. A. Coddington and N. Levinson, Theory of Ordinary Differential Equations (McGraw-Hill, 1955) 9th reprint (1987).
  2. [3]W. R. Hamilton, Philosophical Transactions of the Royal Society of London 124, 247 (1834).
  3. [4]W. R. Hamilton, Philosophical Transactions of the Royal Society of London 125, 95 (1835).
  4. [5]V. I. Arnold, Mathematical Methods of Classical Mechanics, 2nd ed., Graduate Texts in Mathematics, Vol. 60 (Springer, New York, 1989).
  5. [6]H. Goldstein, C. Poole, and J. Safko, Classical Mechanics, 3rd ed. (Addison-Wesley, San Francisco, 2002).
  6. [7]E. Noether, Nachrichten von der Gesellschaft der Wissenschaften zu Göttingen, Mathematisch-Physikalische Klasse , 235 (1918).
  7. [8]A. Rothkopf and W. A. Horowitz, Physical Review E (2024), 10.1103/rch9-wsbw, arXiv:2409.11063 [physics.class-ph].DOI
  8. [9]H. Goldstein, C. Poole, and J. Safko, Classical Mechanics, 3rd ed. (Addison Wesley, 2002).
  9. [10]M. R. Flannery, Journal of Mathematical Physics 52, 032705 (2011).
  10. [11]I. M. Gelfand and S. V. Fomin, Calculus of Variations (Dover, Mineola, NY, 2000).
  11. [12]A. Einstein, Annalen der Physik 17, 891 (1905).
  12. [13]K. R. Popper, The Logic of Scientific Discovery (Routledge, London, 1959).
  13. [14]M. R. Flannery, American Journal of Physics 73, 265 (2005).
  14. [15]M. de León, M. Lainz, A. López-Gordón, and J. C. Marrero, Journal of Geometry and Physics 198, 105116 (2024).
  15. [16]W. A. Horowitz and A. Rothkopf, (2024), arXiv:2408.14420 [math-ph].arxiv.org/abs/2408.14420
  16. [17]O. E. Fernandez and M. L. Radhakrishnan, Scientific Reports 8, 14878 (2018).
  17. [18]B. Bert and W. A. Horowitz, (2025), arXiv:2505.20684 [hep-th].arxiv.org/abs/2505.20684
  18. [19]C. R. Galley, Phys. Rev. Lett. 110, 174301 (2013), arXiv:1210.2745 [gr-qc].DOI
  19. [20]A. Rothkopf and J. Nordström, J. Comput. Phys. 498, 112652 (2024), arXiv:2307.04490 [math.NA].arxiv.org/abs/2307.04490
  20. [21]J. Schwinger, Journal of Mathematical Physics 2, 407 (1961).
  21. [22]L. V. Keldysh, Soviet Physics JETP 20, 1018 (1965).
  22. [23]J. Berges, AIP Conf. Proc. 739, 3 (2004), arXiv:hep-ph/0409233.DOI
  23. [24]J. Rammer, Quantum Field Theory of Non-equilibrium States (Cambridge University Press, Cambridge, 2007).
  24. [25]K. Balzer and M. Bonitz, Nonequilibrium Green’s Functions Approach to Inhomogeneous Systems, Lecture Notes in Physics, Vol. 867 (Springer, Berlin, Heidelberg, 2013).
  25. [26]A. Kamenov, Field Theory of Non-Equilibrium Systems (Cambridge University Press, Cambridge, 2011).
  26. [27]N. P. Landsman and C. G. van Weert, Phys. Rept. 145, 141 (1987).
  27. [28]A. K. Das, Finite Temperature Field Theory (World Scientific, New York, 1997).
  28. [29]M. Laine and A. Vuorinen, Basics of Thermal Field Theory, Vol. 925 (Springer, 2016) arXiv:1701.01554 [hep-ph].
  29. [30]R. D. Jordan, Physical Review D 33, 444 (1986).
  30. [31]S. Weinberg, Physical Review D 72, 043514 (2005).
  31. [32]J. Casalderrey-Solana, H. Liu, D. Mateos, K. Rajagopal, and U. Achim Wiedemann, Gauge/String Duality, Hot QCD and Heavy Ion Collisions (Cambridge University Press, 2014) arXiv:1101.0618 [hep-th].
  32. [33]R. P. Feynman and F. L. Vernon, Jr., Annals Phys. 24, 118 (1963).
  33. [34]A. O. Caldeira and A. J. Leggett, Physica A 121, 587 (1983).
  34. [35]P. M. Morse and H. Feshbach, Methods of Theoretical Physics (McGraw-Hill, New York, 1953) section 3.2.
  35. [36]H. F. Trotter, Proc. Am. Math. Soc. 10, 545 (1959).
  36. [37]G. Strang, SIAM Journal on Numerical Analysis 5, 506 (1968).
  37. [38]L. Hörmander, The Analysis of Linear Partial Differential Operators I: Distribution Theory and Fourier Analysis, 2nd ed., Classics in Mathematics (Springer Berlin Heidelberg, 2003) softcover ISBN: 978-3-540-00662-6.
  38. [39]T. Kugo and I. Ojima, Prog. Theor. Phys. Suppl. 66, 1 (1979).
  39. [40]J. Berges and T. Gasenzer, Phys. Rev. A 76, 033604 (2007), arXiv:cond-mat/0703163.DOI
  40. [41]R. A. Horn and C. R. Johnson, Matrix Analysis, 2nd ed. (Cambridge University Press, Cambridge, 2013).
  41. [42]G. V. Dunne, J. Phys. A 41, 304006 (2008), arXiv:0711.1178 [hep-th].DOI
  42. [43]R. P. Feynman and A. R. Hibbs, Quantum Mechanics and Path Integrals (McGraw–Hill, New York, 1965).
  43. [44]L. s. Schulman, TECHNIQUES AND APPLICATIONS OF PATH INTEGRATION (1981).
  44. [45]A. E. Brouwer and W. H. Haemers, Spectra of Graphs, Universitext (Springer Science & Business Media, 2012).
  45. [46]R. Wong, Asymptotic Approximations of Integrals, Classics in Applied Mathematics, Vol. 34 (Society for Industrial and Applied Mathematics (SIAM), Philadelphia, PA, 2001) corrected reprint of the 1989 edition.DOI
  46. [47]A. Rothkopf, W. A. Horowitz, and J. Nordström, J. Comput. Phys. 524, 113686 (2025), arXiv:2404.18676 [math.NA].DOI
  47. [48]A. Rothkopf, W. A. Horowitz, and J. Nordström, In preparation (2026).
  48. [49]T. H. Wolff, Lectures on Harmonic Analysis, edited by I. Laba and C. Shubin, University Lecture Series, Vol. 29 (American Mathematical Society, Providence, RI, 2003) p. 137.DOI
  49. [50]B. S. DeWitt, Rev. Mod. Phys. 29, 377 (1957).
  50. [51]M. S. Marinov, Phys. Rept. 60, 1 (1980).
  51. [52]H. Kleinert, Path Integrals in Quantum Mechanics, Statistics, Polymer Physics, and Financial Markets (World Scientific, Singapore, 2009).

Paper details

vlnc:W_b5vkbhac
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

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