QUADCOIL

Welcome to the github page of QUADCOIL, the stellarator coil complexity proxy/global coil optimizer!

Developer contact: Lanke Fu, LF2869@nyu.edu

QUADCOIL is a global coil optimization code that approximates coils with a smooth sheet current. In other words, it's a "winding surface" code. However, unlike other winding surface codes, QUADCOIL:

• Supports constrained optimization.
• Supports non-convex quadratic penalties/constraints, such as curvatures and Lorentz force.
• Includes robust winding surface generators that do not produce self-intersections.
• Calculates derivatives with respect to plasma shape, winding surface shape, objective weights, and constraint thresholds.

Read QUADCOIL documentations here.

Changelog

2026/06/22

• Adding xyz surface (SurfaceXYZFourierJAX)
• Adding xyz tensor surface (SurfaceXYZTensorFourierJAX)
• Adding exact uniform offset surface (SurfaceOffsetJAX), produced by calling Surface.uniform_offset() on any surface that's not a SurfaceOffsetJAX.
• Overhauling winding surface generators to work for all 3 surfaces types (a SurfaceOffsetJAX cannot be offset again by design). Improving smoothness. Winding surface can now simply generated by calling Surface.gen_winding_surface().
• Changing the signature of quadcoil() to reflect the updates in winding surface generators
• Overhauling adjoint differentiation to use the KKT condition, instead of the unconstrained auglag objective
• Adding preconditioners
  • precond='svd' for svd of induction matrix
  • precond='svd_K' to use the svd of the r-phi-z current K matrix
  • precond='ess' to use exponential spectral scaling
• Adding nescoil(). It takes largely the same parameters as quadcoil but solves the nescoil problem
• Adding the option to initialize a quadcoil run from a nescoil solution
• Adding interior point method (solver='ipm') and slsqp-jax solver (solver='slsqp'). However, benchmark shows that auglag still works the best.

Please see documentations for more details.