Hamilton Revised: The Action Principle for Initial Value Problems

This paper rigorously derives a variational action principle for classical initial value problems by taking the classical limit of the Schwinger-Keldysh formalism, revealing that both forward and backward paths possess classical solutions and fluctuations, with the backward path uniquely propagating backwards in time to naturally enforce zero deviation without manual constraint.

The Gist

Imagine you are trying to predict where a ball will be tomorrow.

In the old days (Newton's way), you'd say: "If I know where the ball is right now and how fast it's moving, I can calculate exactly where it will be." This is the Initial Value Problem. It's how we actually do physics in the real world.

But for about 200 years, the "fancy" way to do this (Hamilton's Principle) was a bit weird. It asked a strange question: "If I know where the ball is now AND where it will be in the future, what path did it take?"

This is like saying, "To figure out how you drove to work this morning, I need to know your starting point and your destination." It feels backwards! It's like having a crystal ball that tells you the future, which breaks the rules of cause and effect.

The Problem with the "Fancy" Way
The paper argues that this "fancy" way has a hidden flaw. When physicists tried to fix it using a method called "Galley's Principle" (which tries to double the number of paths to fix the math), they ran into a glitch. The math kept forcing one of the paths to be zero, and the solutions would jump around weirdly at the end. It was like trying to balance a scale, but the weights kept disappearing.

The Solution: The "Time-Traveling" Shadow
The authors of this paper, Horowitz and Rothkopf, decided to look at this problem through the lens of Quantum Mechanics. Specifically, they used a tool called the Schwinger-Keldysh formalism.

Think of the quantum world as a foggy morning. You don't see a single clear path; you see a thousand possible paths the particle could take, all overlapping.

• The "Plus" Path: This is the average, the main path, the one that looks like the classical ball we see.
• The "Minus" Path: This is the difference between the forward path and the backward path. In the quantum fog, this represents the "wobble" or the fluctuations.

Usually, physicists say, "Okay, when we get to the classical world (where things are big and solid), we just pretend the 'Minus' path doesn't exist. We just set it to zero by hand."

The Big Discovery
The authors say: "No, don't set it to zero by hand! Let the math do it for you."

They ran the quantum math all the way down to the classical limit (where Planck's constant, $\hbar$, goes to zero). Here is what they found, using a simple analogy:

Imagine the "Minus" path is a shadow cast by the ball.

1. In the quantum world, the shadow is wiggly and real.
2. As you move toward the classical world, the shadow doesn't just vanish; it gets pulled tight.
3. The math shows that the shadow has a "rule": It must start at zero at the end of the movie.
4. Because it starts at zero at the end, and the laws of physics are strict, the shadow is forced to be zero all the way back to the beginning.

So, the "Minus" path doesn't need to be deleted by a human hand. The equations of motion naturally force it to be zero, propagating backwards in time from the future to the past. It's like a domino effect running in reverse: if the last domino is standing still, the math proves all the previous ones must have been standing still too.

The "Revised" Action
Because of this discovery, the authors propose a new, corrected version of Hamilton's Principle (which they call Hamilton's Revised Action).

• Old Way: You had to guess the start and the end, and the math was messy about how velocity and position changed.
• New Way: You only need to know the start (where the ball is and how fast it's going). The math automatically handles the rest.
  • It introduces a "ghost" path (the minus path) that starts at zero in the future and forces itself to be zero in the past.
  • It adds a small "kick" at the beginning to account for the initial momentum (how fast the ball was thrown).

Why This Matters
This isn't just about balls and toys. This is a fundamental fix to how we write the laws of physics.

1. It fixes the "Time Travel" problem: We no longer need to know the future to predict the present.
2. It fixes the "Velocity" confusion: It clearly separates the position from the speed in the math, which has been a headache for a long time, especially for complex systems like rolling wheels (non-holonomic constraints).
3. It bridges the gap: It shows a perfect, clean line from the fuzzy quantum world to the solid classical world, proving that the "classical" rules are just the natural result of the quantum rules when the "wiggles" die out.

In a Nutshell
The paper says: "Stop guessing the future to find the past. Instead, use the quantum 'shadow' method. Let the shadow naturally collapse to zero as you move from the quantum world to the classical world. This gives us a perfect, unambiguous set of rules for how everything moves, from a single atom to a rolling car."

Technical Summary

1. Problem Statement

The paper addresses fundamental conceptual and practical limitations in the standard formulation of classical mechanics derived from Hamilton's Principle of Extremized Action. Specifically, it targets three main issues:

• The Boundary Value vs. Initial Value Paradox: Standard derivations of the Euler-Lagrange equations require fixing the path variations at both the initial ($t_i$) and final ($t_f$) times ($\delta q(t_i) = \delta q(t_f) = 0$). This implies a boundary value problem (BVP), requiring "acausal" knowledge of the final state to determine the trajectory. However, physical reality and experiments operate as Initial Value Problems (IVP), where only initial position and velocity are known.
• The Independence of Position and Velocity Variations: In the standard derivation, the variation of velocity $\delta \dot{q}$ is treated as the time derivative of the position variation $\frac{d}{dt}(\delta q)$ (the transposition rule). The authors argue this is conceptually ambiguous, particularly for systems with non-holonomic constraints (where constraints depend on velocities), leading to confusion regarding the symplectic structure of phase space.
• The Ad Hoc Nature of the Classical Limit in Schwinger-Keldysh Formalism: The Schwinger-Keldysh (closed-time-path) formalism in quantum mechanics doubles degrees of freedom into "forward" ($1$) and "backward" ($2$) paths, often redefined as average ($+$) and difference ($-$) paths. The standard approach to recovering classical mechanics involves setting the "$-$" degrees of freedom to zero by hand in the $\hbar \to 0$ limit. The paper questions whether this is a rigorous consequence of the limit or an arbitrary imposition.

The paper also critiques a recent proposal by Galley (2013) which attempted to resolve the IVP issue by doubling degrees of freedom but failed to correctly handle boundary conditions, leading to numerical instabilities (jumps) and unphysical constraints on the "$-$" path.

2. Methodology

The authors rigorously derive the classical limit of classical mechanics by starting from the Schwinger-Keldysh path integral formulation of quantum mechanics for a non-relativistic point particle.

• Starting Point: The expectation value of the position operator $\langle \hat{x} \rangle(t_f)$ for a Gaussian wavepacket initial state.
• Discretization (Trotterization): The time evolution operator is discretized into $N$ steps. The authors carefully insert complete sets of position and momentum eigenstates. Crucially, they place momentum insertions at half-time steps ($k+1/2$) relative to position steps ($k$) to ensure numerical accuracy and correct operator ordering (Strang splitting).
• Coordinate Transformation: The forward ($1$) and backward ($2$) paths are transformed into average ($+$) and difference ($-$) coordinates:
  $$x_+ = \frac{x_1 + x_2}{2}, \quad x_- = x_1 - x_2$$
  (and similarly for momenta).
• Introduction of Lagrange Multipliers: To handle the mismatch in the number of position and momentum integrals and to rigorously define the relationship between velocity and position, the authors introduce Lagrange multipliers ($\lambda$). These act as new momentum-like variables that enforce constraints (e.g., relating the discrete difference of positions to the velocity variable).
• Stationary Phase Approximation: The classical limit is taken via the Method of Stationary Phase as $\hbar \to 0$. The authors do not simply set $\hbar=0$; they analyze the critical points of the action functional.
• Generalization: The derivation is performed first in 1D Cartesian coordinates and then generalized to $d$-dimensional generalized coordinates with a metric $g_{ij}(q)$, avoiding ad hoc operator ordering prescriptions.

3. Key Contributions

A. Derivation of "Hamilton's Revised Action" ($S_{HR}$)

The authors derive a new variational action principle for initial value problems. The action is defined as:
$$S_{HR} = \vec{p}_0 \cdot \vec{x}_-(t_i) + \int_{t_i}^{t_f} dt \left[ L(\vec{x}_1, \dot{\vec{x}}_1) - L(\vec{x}_2, \dot{\vec{x}}_2) \right]$$
where $\vec{p}_0$ is the initial momentum.
Crucially, the boundary conditions for the variation are:
$$\delta x_-(t_f) = 0, \quad \delta x_+(t_i) = 0$$
Note: Unlike standard formulations or Galley's proposal, $\delta x_-(t_i)$ is not set to zero. This allows the variation to yield the initial momentum condition naturally.

B. Rigorous Emergence of the Classical Limit

The paper demonstrates that the classical limit is not achieved by manually setting the "$-$" variables to zero. Instead:

1. The equations of motion derived from $\delta S_{HR} = 0$ for the "$-$" coordinates constitute an Initial Value Problem with "initial" conditions at the final time ($t_f$): $x_-(t_f) = 0$ and $p_-(t_f) = 0$.
2. Because the equations of motion for the "$-$" sector propagate backwards in time from $t_f$ to $t_i$, and the unique solution to this homogeneous system with zero final conditions is the trivial solution, the dynamics force $x_-(t) = 0$ and $p_-(t) = 0$ for all $t$.
3. This proves that the "$-$" degrees of freedom vanish dynamically, not by hand.

C. Resolution of the Transposition Rule Ambiguity

By treating the velocity $\dot{x}$ as an independent integration variable in the path integral (enforced by Lagrange multipliers), the authors show that $\delta \dot{x}$ and $\delta x$ are independent variations before the equations of motion are applied. The relationship $\dot{x} = \frac{dx}{dt}$ is enforced dynamically by the Lagrange multipliers. This resolves the conceptual confusion regarding the transposition rule $\delta \dot{x} = \frac{d}{dt}(\delta x)$, particularly for non-holonomic constraints where this rule is often debated.

D. Correction to Galley's Principle

The paper identifies that Galley's original formulation incorrectly set $\delta x_-(t_i) = 0$, leading to an overdetermined system and numerical jumps. The authors show that the correct formulation requires injecting the initial momentum $p_0$ into the action (via the term $p_0 x_-(t_i)$) and only fixing $\delta x_-(t_f)$ and $\delta x_+(t_i)$.

4. Key Results

• Recovery of Newton/Hamilton/Euler-Lagrange: The variation of $S_{HR}$ yields the standard Euler-Lagrange equations and Hamilton's equations as Initial Value Problems.
  • Initial conditions: $x(t_i) = x_0$ and $p(t_i) = p_0$ (derived naturally from the variation).
  • Equations of motion: Standard form.
• Backward Propagation of Fluctuations: The "$-$" path (often interpreted as quantum fluctuations) is shown to propagate backwards in time from the final time slice. The unique solution to the "$-$" equations of motion with zero final boundary conditions is identically zero.
• Normalization: The paper explicitly calculates the Hessian of the action and shows that the functional determinants from the Gaussian fluctuations in the numerator and denominator of the expectation value cancel exactly, yielding the correct classical result $\langle \hat{x} \rangle = x_{cl}$.
• Generalized Coordinates: The derivation successfully extends to curved manifolds (generalized coordinates) without relying on ambiguous midpoint rules for operator ordering, providing a rigorous Trotterization for non-commuting operators.

5. Significance

• Foundational Clarity: It resolves a century-old conceptual tension between the boundary-value nature of the variational principle and the initial-value nature of physical dynamics. It proves that Hamilton's Principle can be rigorously formulated as an IVP without "acausal" assumptions.
• Numerical Stability: By correcting the boundary conditions (removing the constraint on $\delta x_-(t_i)$), the paper provides a stable variational principle for numerical simulations of classical and dissipative systems, fixing the "jumps" observed in previous implementations of Galley's principle.
• Non-Holonomic Constraints: The explicit treatment of velocity variations as independent variables offers a promising pathway to resolve long-standing open questions regarding the variational formulation of systems with non-holonomic constraints (e.g., rolling without slipping) and their quantization.
• Quantum-Classical Correspondence: It provides a rigorous, non-ad-hoc derivation of the classical limit from the Schwinger-Keldysh formalism, showing that the vanishing of the "difference" path is a dynamical consequence of the $\hbar \to 0$ limit and the specific boundary conditions of the expectation value, rather than an arbitrary truncation.

In summary, the paper establishes a new, rigorous variational framework ("Hamilton's Revised Action") that unifies the quantum path integral formalism with the classical initial value problem, correcting previous misconceptions about the role of the "minus" path and the independence of velocity variations.