# Noise in Stochastic Partial Differential Equations (SPDEs)
Noise in Stochastic Partial Differential Equations (SPDEs) is a fundamental concept that represents randomness influencing the behavior of systems modeled by PDEs. Unlike deterministic PDEs, SPDEs incorporate stochastic processes to account for unpredictable phenomena, such as thermal fluctuations or environmental randomness. Noise in SPDEs can vary in type: additive, where it directly influences the equation's output; or multiplicative, where it depends on the solution state. Commonly, noise is modeled using Wiener processes or Lévy noise. The analysis of SPDEs involving noise is crucial for understanding complex systems in fields like fluid dynamics, population models, and financial mathematics, where uncertainty is inherent.

- [Stochastic partial differential equations in surface growth models](https://viadean.notion.site/Stochastic-partial-differential-equations-in-surface-growth-models-13b1ae7b9a328085a848da3e02e8f498?)

---
config:
 sankey:
  showValues: false
---
sankey-beta
Code snippets,MATLAB,4
MATLAB,KPZ Model,1
MATLAB,EW Model,1
MATLAB,MH Model,1
MATLAB,noise in SPDEs,1

Code snippets,Python,4
Python,KPZ Model,1
Python,EW Model,1
Python,MH Model,1
Python,noise in SPDEs,1

Code snippets,Julia,3
Julia,EW Model,1
Julia,MH Model,1
Julia,noise in SPDEs,1

Code snippets,R,1
R,KPZ Model,1

Code snippets,C++,1
C++,KPZ Model,1

### :cactus:MATLAB snippet

Handling noise in Stochastic Partial Differential Equations (SPDEs) is an essential aspect of modeling real-world phenomena where randomness plays a significant role. MATLAB provides powerful tools for simulating SPDEs with various noise structures. Below is an outline of approaches to incorporate noise in SPDEs using MATLAB.

### 1. **Types of Noise in SPDEs**

- **Additive Noise**: The noise term is independent of the solution \( $h(x, t)$ \) and is added directly to the equation. It has the form:
$$
\frac{\partial h(x, t)}{\partial t} = \mathcal{L}[h(x, t)] + \eta(x, t),
$$

where \($\eta(x, t)$\) is a space-time white noise.

- **Multiplicative Noise**: The noise depends on the solution and can amplify or dampen the fluctuations depending on the state of \( h(x, t) \). It takes the form:
$$
\frac{\partial h(x, t)}{\partial t} = \mathcal{L}[h(x, t)] + f(h(x, t)) \eta(x, t),
$$
where \( $f(h)$ \) is a function that modulates the noise.

### 2. **Simulating SPDEs in MATLAB**

MATLAB can be used to numerically solve SPDEs using finite difference or finite element methods, along with stochastic integration techniques. Here's how to get started:

#### a. **Discretization of the SPDE**

To simulate an SPDE, the domain needs to be discretized both in space and time:

- **Spatial Discretization**: Use finite difference methods (e.g., forward, backward, or central differences) or finite element methods to discretize the Laplacian or higher-order derivatives.
- **Time Discretization**: Use explicit or implicit time-stepping schemes such as Euler-Maruyama for stochastic systems.

#### b. **Generating Noise**

- **White Noise**: Generate white noise using random number functions such as `randn` in MATLAB for space-time noise.
- **Colored Noise**: To generate colored noise (e.g., spatially correlated noise), apply filtering techniques to `randn` output using functions like `conv2`.

### 3. **Example: Simulating the KPZ Equation**

The KPZ equation is an archetype of non-linear SPDEs:
$$
\frac{\partial h(x, t)}{\partial t} = \nu \nabla^2 h(x, t) + \frac{\lambda}{2} (\nabla h(x, t))^2 + \eta(x, t),
$$
where \($\eta(x, t)$\) represents Gaussian white noise.

**MATLAB Code Outline**:

% Parameters
L = 10; % Domain length
N = 100; % Number of spatial grid points
dx = L / (N-1); % Spatial step
dt = 0.01; % Time step
T = 1; % Total simulation time
nu = 1; % Diffusion coefficient
lambda = 1; % Nonlinearity coefficient

% Initial condition
h = zeros(N, 1);

% Precompute spatial differential operator (Laplacian)
Laplacian = diag(ones(N-1, 1), 1) + diag(ones(N-1, 1), -1) - 2*eye(N);
Laplacian = Laplacian / dx^2;

% Simulation loop
for t = 0:dt:T
    % Compute non-linear term
    grad_h = diff(h) / dx;
    grad_h_sq = grad_h.^2;
    
    % Add noise
    noise = sqrt(dt) * randn(N, 1); % White noise scaled for dt

    % Update equation (Euler-Maruyama method)
    h = h + dt * (nu * Laplacian * h + (lambda / 2) * [grad_h_sq; 0]) + noise;
    
    % Apply boundary conditions (e.g., periodic)
    h(1) = h(N); h(N) = h(2);
end

% Plotting results
plot(h);
title('Surface Height Evolution');
xlabel('Position');
ylabel('Height');

### 4. **Key Considerations**

- **Time Step Selection**: Ensure that the time step \( $dt$ \) satisfies stability conditions, especially when using explicit methods for numerical integration.
- **Noise Scaling**: The noise should be scaled appropriately to simulate realistic physical systems. For instance, the noise term \( $\eta(x, t)$ \) is often scaled by \( $\sqrt{dt}$ \) to maintain consistent variance over time.
- **Boundary Conditions**: Implement suitable boundary conditions (e.g., periodic, Neumann, or Dirichlet) as per the physical model requirements.

### 5. **MATLAB Functions and Toolboxes**

- **`randn`**: Generate Gaussian white noise.
- **`conv2`**: Apply 2D convolution for creating spatially correlated noise.
- **`ode45`, `ode15s`**: For deterministic parts if needed in hybrid models.

### **Extensions and Advanced Topics**

- **Higher-Dimensional Models**: Extend the approach for 2D or 3D surfaces by adjusting the discretization and noise generation.
- **Stochastic Finite Element Method**: Use MATLAB's `pdepe` or third-party toolboxes for handling more complex SPDEs with finite elements.
- **Parallel Computing**: Utilize MATLAB’s parallel computing tools for larger simulations that require high computational power.

By integrating these methods, MATLAB can effectively simulate various SPDEs with different types of noise, providing insights into complex systems with inherent randomness.

### :cactus:Python snippet

Implementing noise in Stochastic Partial Differential Equations (SPDEs) using Python involves simulating noise that acts as a driving force in these equations. The most common type of noise used in SPDEs is **Gaussian white noise** or **colored noise**. Here’s how we can handle noise in Python for SPDEs, with practical examples:

### 1. **Understanding Noise in SPDEs**

- **White Noise**: Spatial and temporal Gaussian noise with zero mean and constant variance. It’s represented as \($\eta(x, t)$\) and is often assumed to be independent in space and time.
- **Colored Noise**: Noise with a correlation structure, where its value depends on neighboring points.

### 2. **Generating Noise in Python**

To simulate noise for an SPDE, libraries such as `numpy`, `scipy`, and specialized tools like `stochastic` and `sdeint` can be used. Here’s a simple guide:

#### White Noise Simulation:

- **1D White Noise**: Use `numpy.random.normal` for generating Gaussian noise.
- **2D White Noise**: Create a 2D grid of Gaussian random numbers.

#### Python Code Example for Noise Generation:

import numpy as np
import matplotlib.pyplot as plt

# Define grid size and time step
Nx = 100  # Number of spatial points
Nt = 100  # Number of time points
dt = 0.01  # Time step

# Generate white noise (1D case)
noise_1d = np.random.normal(0, 1, (Nt, Nx))

# Visualize noise
plt.imshow(noise_1d, aspect='auto', cmap='gray', extent=[0, Nx, 0, Nt])
plt.colorbar(label='Noise Amplitude')
plt.xlabel('Spatial Coordinate')
plt.ylabel('Time Step')
plt.title('1D White Noise')
plt.show()

### 3. **Example of Integrating Noise into an SPDE**

Let’s illustrate with the **Edwards–Wilkinson (EW) equation** as an example:
$$
\frac{\partial h(x, t)}{\partial t} = \nu \nabla^2 h(x, t) + \eta(x, t),
$$

where \($\eta(x, t)$\) is white noise.

#### Numerical Scheme:

1. Discretize the spatial domain with a grid.
2. Use an explicit Euler scheme for the time-stepping.
3. Apply a finite-difference method for the Laplacian (\($\nabla^2$\)).

#### Python Code for the EW Equation:

# Import necessary libraries
import numpy as np
import matplotlib.pyplot as plt

# Parameters
Nx = 100  # Number of spatial points
Nt = 200  # Number of time steps
dx = 1.0  # Spatial step
dt = 0.01  # Time step
nu = 0.1  # Diffusion coefficient

# Initialize height profile and noise
h = np.zeros(Nx)
noise = np.random.normal(0, 1, (Nt, Nx))

# Discretized Laplacian operator
def laplacian(h, dx):
    return (np.roll(h, -1) - 2 * h + np.roll(h, 1)) / dx**2

# Time evolution
for t in range(1, Nt):
    h += nu * laplacian(h, dx) * dt + np.sqrt(dt) * noise[t]

# Plot final profile
plt.plot(h, label='Height profile at final time step')
plt.xlabel('Spatial Coordinate')
plt.ylabel('Height')
plt.title('Surface Evolution (EW Equation)')
plt.legend()
plt.show()

### 4. **Adding Colored Noise**

Colored noise can be introduced by filtering white noise:

- **Ornstein–Uhlenbeck Process**: Generates temporally correlated noise.
- **Gaussian Filter**: Smooths spatial white noise to create spatial correlation.

#### Python Code for Colored Noise:

from scipy.ndimage import gaussian_filter

# Generate 1D white noise
white_noise = np.random.normal(0, 1, Nx)

# Apply Gaussian filter for spatial correlation (colored noise)
colored_noise = gaussian_filter(white_noise, sigma=2)

plt.plot(white_noise, label='White Noise')
plt.plot(colored_noise, label='Colored Noise', linestyle='--')
plt.xlabel('Spatial Coordinate')
plt.title('Comparison of White and Colored Noise')
plt.legend()
plt.show()

### 5. **Advanced Libraries for SPDEs**

- **`stochastic` library**: Offers tools for simulating different types of stochastic processes.
- **`sdeint` library**: Useful for numerical integration of stochastic differential equations.
- **`pde` Python package**: Specialized for solving partial differential equations, including SPDEs, with built-in noise functionality.

### 6. **Tips for Implementing SPDEs**

- Ensure time and space discretization satisfies stability conditions, especially when dealing with numerical schemes involving noise.
- For higher-dimensional SPDEs, use libraries such as `NumPy`, `SciPy`, or `TensorFlow` for efficient array operations.
- Consider the **CFL (Courant-Friedrichs-Lewy) condition** for stable integration when using explicit time-stepping methods.

This setup helps create simulations for models like the KPZ equation or other SPDEs incorporating complex noise structures.