NFoil

High-performance subsonic airfoil analysis with Numba JIT acceleration.

NFoil is a fork of mfoil by Krzysztof J. Fidkowski, rewritten with Numba JIT compilation for ~18x faster subsonic airfoil analysis. It retains full numerical compatibility with the original while dramatically reducing solve times with the addition of a Graphical User Interface (GUI).

Developed and optimized by Cayetano Martínez-Muriel with assistance from Google Antigravity. Based on the original mfoil by Krzysztof J. Fidkowski, which uses mostly the same physical models as XFOIL, with some modifications.

---

Performance

Test: Polar computation consisting of 16 angles of attack for the viscous flow around a NACA 2412 airfoil at $Re=10^6$

Panels	Original (s)	NFoil (s)	Speedup
100	26.5	1.4	18.8x
200	67.7	4.0	17.0x
300	115.6	6.1	18.8x
400	180.8	11.3	16.0x

Both versions share the same algorithmic complexity of $\approx O(N^{1.35})$. The speedup comes entirely from eliminating Python/SciPy overhead via JIT compilation and dense array operations.

Results obtained after running both versions in the command line on a M1-Max MacBook Pro.

---

Interactive GUI

NFoil includes a full-featured GUI (gui.py) built with CustomTkinter:

• Real-time analysis: Single-point solve with live Cp distribution, airfoil geometry (with physical BL thickness $\delta$), and aerodynamic coefficients.
• Robust polar computations: Multi-$\alpha$ sweep with automatic continuation past non-converged points. If a point fails, the solver retries with a cold BL restart; if it still fails, it skips that point and continues the sweep. All computed cases from multiple runs are stored with descriptive labels (NACA/$Re$/$\alpha$/$M$) in a persistent dropdown. It is possible to overlay polars from multiple Reynolds numbers for comparison.
• Target Cl (Inverse Mode): A dedicated toggle and input field to find the angle of attack for a specific lift coefficient ($C_L$), integrated into both single-point solves and polar sweeps.
• Load/Unload Airfoil: Directly load custom coordinates from .dat or .txt files (TE-to-TE, CCW/CW). Includes an "Unload" button to instantly revert to the parameterized NACA generator.
• BL properties tab: Skin friction ($C_f$), displacement thickness ($\delta^*$), momentum thickness ($\theta$), and shape parameter ($H_k$) across upper, lower, and wake surfaces.
• GPU flow fields (optional): Real-time velocity/pressure contour maps via Taichi Lang (taichi_fields.py), leveraging Metal/CUDA (and other) GPUs.
• ASCII export: $C_p$ curves, geometry, skin friction arrays, and polar data.

python gui.py

---

Main Changes from Original mfoil

1. Dense Array Assembly (Eliminating SciPy Sparse Overhead)

The #1 bottleneck in the original code was SciPy sparse matrix overhead inside the coupled Newton solver:

• build_glob_sys: Replaced scipy.sparse.dok_matrix allocation for R_U and R_x with dense np.zeros arrays. This eliminated ~70% of total runtime that was spent in DOK indexing, _validate_indices, __setitem__, and COO conversion.

• solve_glob: Replaced scipy.sparse.lil_matrix assembly + scipy.sparse.linalg.spsolve with dense np.zeros array + np.linalg.solve. Includes a LinAlgError fallback to np.linalg.lstsq for near-singular Jacobians at extreme operating conditions.

2. Numba JIT-Compiled Physics Kernels

All core boundary-layer functions have been rewritten as @njit(cache=True, fastmath=True) kernels.

3. Vectorized Sensitivity Assembly

calc_ue_m (inviscid edge velocity sensitivity matrix) was rebuilt with parallelised Numba kernels: - build_B_bulk: Vectorized source-panel influence coefficient assembly over all panels simultaneously, replacing an O($N^2$) Python loop. - build_Csig_bulk: Vectorized wake source-influence assembly.

4. Robust Solver Improvements

• Singular matrix fallback: solve_glob catches np.linalg.LinAlgError and falls back to np.linalg.lstsq for near-singular Jacobians (common at very low Reynolds numbers or high angles of attack).

5.Flow Field Calculation

• Velocity and pressure fields: taichi_fields.py computes the velocity and pressure fields on a background grid using a Taichi Lang, rendering the flow field in real-time using the available GPU.

---

Dependencies

numpy
scipy
matplotlib
numba
customtkinter   # for GUI only
taichi           # optional, for GPU flow fields

Quick Start

import nfoil as nf

# Create and solve
N = nf.nfoil(naca='2412', npanel=200)
N.setoper(alpha=5, Re=1e6, Ma=0.3)
N.solve()

# Results
print(f"Cl = {N.post.cl:.4f}")
print(f"Cd = {N.post.cd:.6f}")
print(f"Cm = {N.post.cm:.4f}")

Citation

If you use or modify this tool, cite this work as:

• Martínez-Muriel, C. NFoil (2026), GitHub Repository, https://github.com/cayetanomarmur/NFoil

License

MIT License — Copyright (C) 2026 Cayetano Martínez-Muriel.

Acknowledgments

Many thanks to Krzysztof J. Fidkowski for his work on mfoil, and to the XFOIL team for the original tool.

Thanks to the Google Antigravity team for the IDE.