Package neqsim.thermo.util.derivatives

package neqsim.thermo.util.derivatives

Automatic differentiation and gradient computation for thermodynamic calculations.

This package provides classes for computing and storing derivatives of thermodynamic properties with respect to temperature, pressure, and composition. These gradients enable:

• Gradient-based optimization of process conditions
• Integration with machine learning frameworks (JAX, PyTorch via custom backward passes)
• Sensitivity analysis of thermodynamic calculations
• Physics-informed neural network training

Key Classes:

• PropertyGradient - Gradients of scalar properties (density, enthalpy, etc.)
• FlashGradients - Gradients of flash calculation results (K-values, phase fractions)
• FugacityJacobian - Jacobian matrix of fugacity coefficients

Usage with Python/JAX:

# Define custom VJP for JAX
# Use @jax.custom_vjp decorator
def flash_density(T, P, z):
    # Forward: call NeqSim
    system = create_system(z)
    system.setTemperature(T)
    system.setPressure(P)
    ops.TPflash()
    return system.getDensity()

def flash_density_bwd(res, g):
    # Backward: use NeqSim's analytical gradients
    grads = DifferentiableFlash(system).computePropertyGradient("density")
    return (g * grads.dT, g * grads.dP, g * grads.dz)

Since:

3.0

Author:

ESOL

Related Packages

Package	Description
neqsim.thermo.util

Classes

Class	Description
DifferentiableFlash	Computes gradients of flash calculation results using the implicit function theorem.
FlashGradients	Container for gradients of flash calculation results with respect to input parameters.
FugacityJacobian	Container for Jacobian matrix of fugacity coefficients.
PropertyGradient	Container for gradients of a scalar thermodynamic property with respect to state variables.