Large deviations principles for symplectic discretizations of stochastic linear Schrödinger Equation

This paper establishes that symplectic discretizations, including spectral Galerkin spatial semi-discretization and temporal full discretization, weakly asymptotically preserve the large deviations principle of the stochastic linear Schrödinger equation, thereby providing an effective numerical approach for approximating the LDP rate function in infinite-dimensional spaces.

The Gist

Imagine you are trying to predict the weather, but instead of a single storm, you are dealing with a chaotic, infinite ocean of waves that are constantly being shaken by random gusts of wind. This is what a Stochastic Linear Schrödinger Equation represents: a mathematical model for how waves (like light or quantum particles) move through a messy, random environment.

The paper you are asking about is like a guidebook for building the best possible simulation (a computer model) of this chaotic ocean. The authors want to know: If we use a computer to simulate this ocean for a very long time, will our simulation tell us the truth about how rare, extreme events happen?

Here is the breakdown of their discovery using simple analogies:

1. The Problem: The "Rare Event" Lottery

In this chaotic ocean, most of the time, the waves behave normally. But occasionally, a "monster wave" (a rare event) appears.

• The Goal: We want to calculate the odds of these monster waves happening. In math, this is called the Large Deviations Principle (LDP). It's like trying to calculate the exact probability of winning the lottery jackpot, but for physical waves.
• The Challenge: The ocean is infinite (it has infinite dimensions). You can't simulate an infinite ocean on a computer. You have to chop it up into a finite grid (a digital approximation).
• The Risk: When you chop up the ocean, you might accidentally change the rules of the game. Your computer model might say monster waves are impossible, when in reality, they happen once in a blue moon.

2. The Solution: The "Symplectic" Compass

The authors tested two types of digital maps (numerical methods) to see which one preserves the truth about these rare events.

• The "Non-Symplectic" Map (The Broken Compass):
  Imagine you are navigating a boat. A standard map might drift slightly off course every time you turn a corner. Over a long journey, this drift adds up, and you end up in a completely different ocean.

  • In the paper: These standard methods fail. They distort the "rare event" probabilities so badly that the computer says the monster waves have a 0% chance of happening, even though they do. They lose the "geometric structure" of the original equation.

• The "Symplectic" Map (The Perfect Compass):
  This is a special type of navigation tool designed specifically for systems that conserve energy and structure (like Hamiltonian systems). It's like a GPS that knows the terrain is curved and adjusts its calculations to stay perfectly aligned with the laws of physics, no matter how long you travel.

  • In the paper: The authors prove that Symplectic Discretizations are the winners. Even though they are just an approximation, they preserve the "fingerprint" of the rare events. If the real ocean has a 1-in-a-million chance of a monster wave, the symplectic computer model also calculates a 1-in-a-million chance (as the simulation gets more precise).

3. The Analogy: The "Infinite Dice"

Think of the Schrödinger equation as a giant, infinite set of dice being rolled every second.

• The Real System: The dice are fair, but the pattern of "snake eyes" (rare events) follows a very specific, complex mathematical rule (the Rate Function).
• The Simulation: You try to simulate this with a computer that can only roll a finite number of dice at a time.
  • Bad Method: You use a method that accidentally "loads" the dice. Over time, the computer thinks "snake eyes" never happen. The simulation is useless for studying rare events.
  • Symplectic Method: You use a method that respects the physics of the dice. Even though you are rolling fewer dice, the pattern of the rare "snake eyes" remains mathematically identical to the real infinite system.

4. The Big Discovery

The paper proves two main things:

1. Existence: They mathematically proved that the "monster waves" (rare events) in the real infinite ocean follow a specific, predictable rule (an LDP).
2. Preservation: They proved that if you use Symplectic methods to simulate this ocean, your computer model will eventually converge to that same rule.

Why does this matter?
In fields like finance, engineering, and physics, we often care most about the "black swan" events—the crashes, the failures, the extreme storms. If your simulation tool (like a standard computer model) tells you these events are impossible because it distorted the math, you might build a bridge that collapses or a portfolio that crashes.

This paper tells us: "If you want to predict the impossible, you must use Symplectic methods." It provides a way to accurately calculate the odds of rare, catastrophic events in complex, infinite-dimensional systems using computers.

Summary

• The Equation: A model for waves in a random world.
• The Question: Can we simulate this on a computer without losing the truth about rare, extreme events?
• The Answer: Yes, but only if you use a special "Symplectic" algorithm.
• The Metaphor: Standard algorithms are like a blurry camera that misses the rare details. Symplectic algorithms are like a high-definition lens that keeps the rare details sharp, no matter how long you zoom in.

This is the first time anyone has proven that these specific "Symplectic" digital tools can accurately mimic the probability of rare events in infinite-dimensional spaces, opening the door to much safer and more accurate predictions in science and engineering.

Technical Summary

Here is a detailed technical summary of the paper "Large Deviations Principles for Symplectic Discretizations of Stochastic Linear Schrödinger Equation" by Chen, Hong, Jin, and Sun.

1. Problem Statement

The paper addresses the long-time asymptotic behavior and probabilistic characteristics of the Stochastic Linear Schrödinger Equation (SLSE) and its numerical discretizations. Specifically, the authors investigate the Large Deviations Principle (LDP) for the observable $B_T = \frac{u(T)}{T}$ as $T \to \infty$, where $u(T)$ is the solution to the SLSE.

The core research questions are:

1. Does the discrete approximation of the observable $B_T$, derived from numerical discretizations of the SLSE, satisfy an LDP?
2. If so, which types of numerical discretizations can asymptotically preserve the LDP rate function of the original infinite-dimensional system?
3. Can symplectic discretizations provide an effective approach to approximating the LDP rate function in infinite-dimensional spaces?

The SLSE is given by:
$$du = i\Delta u dt + i\alpha dW(t), \quad u(0) = u_0$$
where $\Delta$ is the Laplacian with Dirichlet boundary conditions, $\alpha > 0$, and $W$ is a $Q$-Wiener process.

2. Methodology

The authors employ a combination of stochastic analysis, large deviation theory, and geometric numerical integration techniques.

• Theoretical Framework: The primary tool is the abstract Gärtner–Ellis theorem, which establishes LDPs based on the existence of the logarithmic moment generating function (LMGF) and exponential tightness.
• Spatial Discretization: The infinite-dimensional system is first approximated using the Spectral Galerkin method. This projects the solution onto a finite-dimensional subspace $H_M$ spanned by the first $M$ eigenfunctions of the covariance operator $Q$.
• Temporal Discretization: The resulting finite-dimensional stochastic Hamiltonian system is discretized in time using:
  • Symplectic schemes: Including the Midpoint scheme and Exponential Euler method.
  • Non-symplectic schemes: Including the Backward Euler–Maruyama method.
• Dimensionality Reduction: The high-dimensional system is decomposed into $M$ independent 2-dimensional subsystems to facilitate the analysis of the LDP for the full discretization.
• Rate Function Analysis: The authors derive explicit expressions for the rate functions of the continuous system, the semi-discrete system, and the full discrete system. They compare these to define "weak asymptotic preservation."

3. Key Contributions

A. LDP for the Exact Solution

The authors prove that the observable $B_T = u(T)/T$ of the exact SLSE satisfies an LDP on the Hilbert space $H_0 = L^2(0, \pi; \mathbb{C})$.

• Rate Function: The good rate function $I(x)$ is explicitly derived as:
  $$I(x) = \begin{cases} \frac{1}{\alpha^2} \|Q^{-1/2}x\|_{H_0}^2 & \text{if } x \in Q^{1/2}(H_0) \\ +\infty & \text{otherwise} \end{cases}$$
• Exponential Tightness: They establish exponential tightness by utilizing the regularity of the solution in $H^1$ and the Fernique theorem to bound the probability of the solution escaping compact sets.

B. LDP for Spatial Semi-Discretization

For the spectral Galerkin approximation $\{u^M\}$, the authors prove that the discrete observable $B^M_T$ satisfies an LDP on $H_0$ with a rate function $I_M$.

• Weak Asymptotic Preservation: They introduce a definition of weak asymptotic preservation, meaning that for any $x$ in the domain of the exact rate function, there exists a sequence of approximations $x_M$ such that $I_M(x_M) \to I(x)$ as $M \to \infty$.
• Result: The spatial semi-discretization weakly asymptotically preserves the LDP of the original system.

C. LDP for Full Discretization (Spatio-Temporal)

This is the paper's most significant contribution. The authors analyze full discretizations combining spectral Galerkin (spatial) and symplectic/non-symplectic methods (temporal).

• Symplectic Discretizations: When using temporal symplectic schemes (satisfying specific determinant and trace conditions), the modified rate function $I_{M, \tau}^{mod}$ converges pointwise to the semi-discrete rate function $I_M$ as the time step $\tau \to 0$. Consequently, the full discretization weakly asymptotically preserves the LDP of the original system.
• Non-Symplectic Discretizations: When using non-symplectic schemes (e.g., Backward Euler–Maruyama), the rate function degenerates. Specifically, the rate function becomes $0$at$x=0$and$+\infty$ elsewhere. This indicates a failure to preserve the LDP structure; the numerical method cannot capture the rare event statistics of the original system.

D. Extension to Complex-Valued Noises

The paper extends the LDP results to the case where the driving noise is complex-valued (composed of two correlated real Wiener processes), deriving the corresponding rate function involving the sum of the covariance operators.

4. Main Results

1. Existence of LDP: The observable $B_T$ for the SLSE satisfies an LDP with a convex, good rate function derived via the Gärtner–Ellis theorem.
2. Preservation by Symplectic Methods: Full discretizations based on symplectic temporal schemes and spectral Galerkin spatial approximation weakly asymptotically preserve the LDP rate function of the continuous system.
   • Mathematical Statement: $\lim_{\tau \to 0} I_{M, \tau}^{mod}(x) = I_M(x)$, and $I_M \to I$ as $M \to \infty$.
3. Failure of Non-Symplectic Methods: Full discretizations using non-symplectic temporal schemes do not preserve the LDP. Their rate functions collapse to a trivial form, failing to describe the exponential decay of rare events accurately.
4. Explicit Rate Functions: The paper provides explicit formulas for the rate functions of various numerical schemes (Midpoint, Exponential Euler, Backward Euler), demonstrating how the symplectic structure maintains the correct dependence on the noise covariance $Q$.

5. Significance

• Geometric Numerical Integration: The results provide a rigorous theoretical justification for the superiority of symplectic integrators in long-time simulations of stochastic Hamiltonian PDEs, extending beyond stability and energy conservation to the preservation of large deviation statistics.
• Approximation of Infinite-Dimensional LDPs: This work offers the first effective approach to approximating LDP rate functions in infinite-dimensional spaces using numerical discretizations. It suggests that symplectic discretizations can be used as a computational tool to estimate the rate functions of complex stochastic systems where analytical solutions are intractable.
• Open Problem Resolution: The paper partially answers an open problem proposed in previous literature (referenced as [3]) regarding the preservation of LDPs by numerical methods for infinite-dimensional systems.
• Practical Implication: For applications in physics (e.g., wave propagation in random media) and finance where rare events are critical, using symplectic discretizations ensures that the statistical properties of rare events are correctly captured, whereas non-symplectic methods may yield misleading results.

In conclusion, the paper establishes that symplectic discretizations are essential for the faithful numerical approximation of the Large Deviation Principles of stochastic linear Schrödinger equations, ensuring that the exponential decay rates of rare events are preserved asymptotically.