DHARA is a general solver for partial differential equations (PDEs) written in Python, designed to run efficiently on both CPUs and GPUs. It combines the flexibility of Python with high-performance libraries like NumPy for CPUs and CuPy for GPUs, enabling large-scale scientific simulations with the same codebase. Built with a modular and object-oriented design, DHARA is easy to extend and adapt to new physical models or numerical methods. It supports parallelization with MPI, efficient memory use, and time integration schemes, making it a versatile tool for studying fluid dynamics and related problems. Currently, DHARA is closed-source; for more details and collaboration opportunities, please visit Vayusoft Labs.

Libraries required: NumPy, CuPy (for GPUs), h5py, mpi4py, matplotlib

Problems DHARA can solve

1. Compressible flows: DHARA has built-in modules for finite volume and finite difference methods. For hyperbolic PDEs, the Kurganov and Tadmor semi-discrete method is implemented with various reconstructions: linear, weighted essentially non-oscillatory (WENO), central WENO (CWENO), and targeted WENO (TENO) up to 7th order. DHARA also includes the TVD-MacCormack scheme, which is well-suited for low-turbulent Mach number compressible convection studies and is highly optimized for performance. It can solve problems in both 2D and 3D, including the Euler equations, compressible magneto-hydrodynamics (MHD), moist convection, and other complex compressible flow systems.
2. Incompressible flows: DHARA solves the incompressible Navier–Stokes equations using finite difference methods with a projection scheme. The pressure–Poisson equation is handled efficiently using multigrid solvers, ensuring scalability for large 2D and 3D problems. Standard Runge–Kutta (RK) time integration schemes are used, making the solver robust and flexible for studying vortical flows, shear layers, and buoyancy-driven convection such as Rayleigh–Bénard convection.
3. Quantum fluids: DHARA includes modules for simulating quantum fluids by solving the Gross–Pitaevskii equation (GPE). It uses a pseudo-spectral method combined with the time-splitting spectral (TSSP) scheme for accurate and stable evolution of Bose–Einstein condensates and related quantum systems. These features make it suitable for studying quantum turbulence, vortex dynamics, and ground-state computations in both 2D and 3D settings.

DHARA architecture

Modular architecture of DHARA

• Object-oriented design:

  DHARA is developed using an object-oriented programming paradigm, where functionalities are organized into modular and extensible components. This design makes it simple to maintain, reuse, and extend the solver for new physical models or numerical schemes.

• Problem class:

  At the core, a dedicated problem class (e.g., Hydro3D) serves as the central interface for setting up and evolving simulations. It handles grid initialization, array allocation, boundary conditions, and external forcing, while also coordinating numerical flux computations.

• Flux solvers:

  Numerical fluxes are computed through specialized solver classes such as KTConvFlux for convective terms and ViscousFlux for viscous terms. Different high-order reconstructions (e.g., WENO, CWENO, TENO) are available as modular plug-ins.

• Time integration:

  Time advancement is managed by the TimeEvolution class, which supports explicit schemes like eSSPRK2, eSSPRK3, and standard Runge–Kutta methods. A solve_single_step routine updates the solution for one timestep, ensuring flexibility to switch schemes easily.

• I/O management:

  Data handling is performed by the DataIO class, which uses the h5py library to store simulation fields and global statistics in HDF5 format. This ensures portability, parallel I/O support, and compatibility with analysis tools.

• Extensibility:

  Thanks to its modular structure, new solvers, numerical methods, or physical models can be included with minimal changes. This makes DHARA not just a solver but a framework for PDE-based research.

Performance analysis