CP-PAW Code: Integrating CP Dynamics & PAW

• CP-PAW is a computational framework that integrates Car–Parrinello molecular dynamics with the all-electron PAW method for accurate simulations of solids, liquids, and molecules.
• It employs on-the-fly PAW augmentation and G-dependent fictitious masses to harmonize electronic and nuclear dynamics while ensuring correct vibrational spectra.
• The code, written in Fortran 2008 and parallelized with MPI, offers robust input workflows, advanced integration techniques, and postprocessing tools for comprehensive atomistic studies.

CP-PAW is a computational framework integrating first-principles electronic structure theory with ab initio molecular dynamics, specifically combining the all-electron Projector Augmented-Wave (PAW) method with Car–Parrinello (CP) dynamics for the simulation of atomistic condensed phase systems. The design enables simultaneous determination of electronic and nuclear ground states as well as dynamical properties in solids, liquids, and molecular assemblies, supporting both quantum and classical components in a unified formalism (Blöchl et al., 17 Jan 2026).

1. Theoretical Underpinnings

CP-PAW centers on the Car–Parrinello Lagrangian, which enables the coupled propagation of nuclear positions $\{R_I\}$ and Kohn–Sham pseudo-wavefunctions $\{|\tilde\psi_n(t)\rangle\}$: $$\mathcal{L}[R, \psi, \dot{R}, \dot{\psi}] = \sum_I \frac{1}{2}M_I \dot{R}_I^2 + \sum_n \frac{1}{2}\mu_n \int d^3r |\partial_t \psi_n(r,t)|^2 - E_{DFT}[\{\psi_n\}, \{R_I\}] +\sum_{mn}\Lambda_{mn} \left[\langle \psi_m|\psi_n\rangle - \delta_{mn}\right]$$ where $M_I$ are ionic masses, $\mu_n$ are fictitious electron masses to separate timescales, $E_{DFT}$ is the DFT total energy, and $\Lambda_{mn}$ enforce orthonormality. The resulting Euler–Lagrange equations include an overlap operator $O$ (arising from augmentation) and yield explicit coupled equations for the ions and electrons.

The PAW method, as implemented, reconstructs all-electron wavefunctions via a transformation $\mathcal{T}$ composed of atom-centered corrections: $$|\psi_n\rangle = \mathcal{T}|\tilde{\psi}_n\rangle = |\tilde{\psi}_n\rangle + \sum_\alpha (|\phi_\alpha\rangle - |\tilde{\phi}_\alpha\rangle) \langle \tilde{p}_\alpha | \tilde{\psi}_n \rangle$$ with precalculated all-electron and pseudo partial waves, and projector functions $\langle\tilde{p}_\alpha|$.

The Car–Parrinello equations and PAW corrections are unified within CP-PAW, ensuring consistent forces and Hamiltonians through the action principle. Mass renormalization subtracts the wavefunction cloud's inertia from ionic masses to recover correct vibrational spectra, while $G$-dependent fictitious masses $\mu(G)$ are deployed to homogenize plane-wave timescales (Blöchl et al., 17 Jan 2026).

2. Code Architecture and Data Handling

CP-PAW is implemented in Fortran 2008 and parallelized with MPI, leveraging BLAS/LAPACK and FFTW3 libraries. Its modular structure is as follows:

Module	Functionality	Special Features
BuildTools	Build system, compiler and defaults	Parameter file editing
PAW_Setup	On-the-fly partial wave & projector construction	No external datasets required
PW_Core	Plane-wave and FFT routines	$G$-space management
PAW_OneCenter	Log-grid one-center (radial, Bessel) computations	Efficient atomic augmentation
CP_Dynamics	Verlet integrators, thermostats, constraints	Dual thermostat, mass renormalization
Static SCF	Wavefunction optimization as damped CP dynamics	Simulated annealing, robust convergence
Forces_Stress	Ionic forces/stress with PAW corrections	All-electron accuracy
IO & Analysis	Input, output, parsing, tools (paw_dos, paw_tra, etc)	Projected DOS, trajectory, structure dump

Wavefunctions are stored as coefficients $\psi_n(G)$ in the plane-wave basis, while augmentation spheres $\Omega_R$ encapsulate site-local data. Input uses a hierarchical tagged-tree format, with flexible block order and case-insensitivity. All PAW augmentation is generated on the fly from minimal per-species parameters (e.g., RAD, NPRO, POW).

3. Installation and Build Workflow

Supported on Unix/Linux and MacOS environments, prerequisites include a Fortran 2008 compiler, BLAS/LAPACK, FFTW3, MPI (optional), LibXC (for advanced functionals), with build orchestrated by bash scripts and GNU Make (≥4.3). Installation involves:

1. Cloning the source repository.
2. Executing paw_install.sh, which auto-detects libraries and builds three executables: paw_db.x (debug), paw_fast.x (serial), and paw_fast_parallel.x (MPI).
3. Compiler flags and library paths may be customized in src/BuildTools/defaultparmfile and passed via -f parameter.

Environment configuration adds ${CPPAW_ROOT}/bin/*</code> to <code>PATH</code> and, if needed, loads relevant module environments for <a href="https://www.emergentmind.com/topics/additive-parallel-correction" title="" rel="nofollow" data-turbo="false" class="assistant-link" x-data x-tooltip.raw="">parallel</a> execution or advanced functionals.</p> <h2 class='paper-heading' id='user-workflow-and-standard-use-cases'>4. User Workflow and Standard Use Cases</h2> <p>Typical CP-PAW projects revolve around four files: the structure (<code>.strc</code>), control (<code>.cntl</code>), protocol/output log (<code>.prot</code>), and binary restart (<code>.rstrt</code>). Inputs follow a block-hierarchical, tag-based format, e.g.,</p> <p>

!STRUCTURE !LATTICE ... !ATOM ... !SPECIES ... !KPOINTS ... !END !CONTROL START=T NSTEP=500 !PSIDYN ... !RDYN ... !END !EOB

Ground-state SCF and Car–Parrinello MD simulations are supported out-of-the-box. For example, malonaldehyde&#39;s equilibrium and finite-temperature MD are implemented by modifying thermostat blocks and constraints and invoking the appropriate executable (<code>paw_fast.x</code> or <code>paw_fast_parallel.x</code>). Analysis of dynamics leverages tools like <code>paw_tra</code> (for mode tracking) and <code>paw_dos</code> / <code>paw_dosplot</code> (for DOS, COHP).</p> <p>Solid-state workflows include k-point formation, occupation settings (e.g., TETRA+ for metals, Mermin functionals for finite temperature), and cell/volume optimizations with subsequent data extraction for equations-of-state or phase transitions.</p> <h2 class='paper-heading' id='distinctive-features-and-best-practice-recommendations'>5. Distinctive Features and Best-Practice Recommendations</h2> <p>CP-PAW&#39;s unique contributions are:</p> <ul> <li><strong>On-the-fly PAW construction:</strong> No dependence on external pseudo potential libraries; all augmentation data is built from minimal parameters directly by the user within species blocks.</li> <li><strong>$G$-Dependent Fictitious Mass:</strong> Tailoring$\mu(G)$$accelerates electronic wavefunction convergence and expands integration stability regions.</li> <li><strong>Dual Thermostat Schemes:</strong> Ions employ standard Nosé–Hoover, while electronic degrees of freedom utilize a wavefunction thermostat that removes fictitious energy without injecting heat.</li> <li><strong>Damped CP Dynamics:</strong> Ground-state optimization employs simulated annealing, with adaptive friction control ($$a_{opt}=\omega\Delta$$).</li> <li><strong>Local <a href="https://www.emergentmind.com/topics/hg-tnet-hybrid" title="" rel="nofollow" data-turbo="false" class="assistant-link" x-data x-tooltip.raw="">Hybrid</a> Functionals (e.g., PBE0r):</strong> Hybrid functionals are rendered computationally feasible in plane-wave codes by expressing exchange integrals in a local atomic basis, eschewing four-center operations.</li> <li><strong>Advanced Integration and Occupation Techniques:</strong> Iterative tetrahedron k-point integration with curvature correction for metals; Mermin functional with dynamic occupations ($$f_n$variables) for explicit finite-$T$ DFT ensembles.</li> <li><strong>Physical Constraints:</strong> Suppression of global translation/rotation (&quot;flying ice cube&quot;) and auxiliary charge models for periodic electrostatics.</li> <li><strong>Tooling and Analysis:</strong> Postprocessing capabilities encompass structure analysis (<code>paw_strc</code>), vibrational mode tracking (<code>paw_tra</code>), projected DOS (<code>paw_dos</code>, <code>paw_dosplot</code>), and extraction of scalar or tensor observables.</li> </ul> <p>Parameter selection for robust simulations follows empirical guidelines: plane-wave cutoff 15–30 Ry (preliminary), 50–100 Ry (production); k-point spacing $R \geq 30$(molecules),$40–60$(metals, with TETRA+); time-step$\Delta$below$2/\omega_{max}$$; and occupation strategy appropriate to system type.</p> <h2 class='paper-heading' id='applications-and-system-flexibility'>6. Applications and System Flexibility</h2> <p>CP-PAW is equipped to address a broad range of condensed matter simulations, from molecular clusters to bulk metals and surfaces. It supports ground-state and finite-temperature studies, structural optimization, vibrational analyses, and excited-state electronic structure via post-hoc methods. The combination of all-electron accuracy with efficient dynamical integration permits investigations of complex oxides, phase transitions, and structure–property relationships across diverse chemical and physical environments (<a href="/papers/2601.12004" title="" rel="nofollow" data-turbo="false" class="assistant-link" x-data x-tooltip.raw="">Blöchl et al., 17 Jan 2026</a>).</p> <h2 class='paper-heading' id='impact-and-development-outlook'>7. Impact and Development Outlook</h2> <p>By integrating Car–Parrinello molecular dynamics with all-electron PAW formalism in a unified, flexible codebase featuring on-the-fly augmentation,$$G$-dependent masses, mass renormalization, and advanced analysis tools, CP-PAW provides a powerful and extensible platform for high-accuracy atomistic simulations. Its open, extensible design, together with a documented workflow and postprocessing suite, positions CP-PAW as a versatile tool for first-principles studies of condensed matter phenomena and complex materials (Blöchl et al., 17 Jan 2026).