Changelog

v0.8.6

  • Fixed a bug in quadrupole formula generation, see this issue on github. Users should rerun any code involving quadrupole SHG following v0.8.0.

v0.8.1

  • Added multiprocessing compatibility to model and cost function generation.

v0.8.0

  • Rewrote the Fourier formula generation process. Previously sympy had a bug which made computing the Fourier transforms of typical SHG formulas incredibly slow. The previous workaroud was to compute the Fourier transforms of a general SHG tensor ahead of time (i.e. the “uncontracted Fourier transforms”), and then having the user generate a full Fourier formula for their specific system using a contraction of this tensor. A different workaround which I was not able to implement until now is to enumerate all the possible terms that could appear in a viable SHG formula (i.e. arbitrary products of sines and cosines), and then create a lookup table for their respective Fourier transforms for access at runtime. This is now implemented in shgpy.core.data_handler.form_to_fform(). My initial tests suggest that, while it is still slightly slower than the previous workaround, not only is it much simpler for the user, but it also makes it trivial to compute the Fourier transforms of higher multiple SHG formulas (i.e. magnetic dipole and electric quadrupole contributions). This is now documented in the tutorials and examples.

  • Added the ability to generate the raw cost functions or model functions from the Fourier formulas. This is useful in case the user wants to use a different wrapper around scipy.optimize like LMFIT (which I highly recommend!).

  • Real-space formula generation is now implemented as a single function: shgpy.formgen.formgen(). This function handles electric dipole, magnetic dipole, and electric quadrupole tensors through three arguments (t_eee, t_mee, and t_qee, respectively).

    • Note: this is the first release in which users are able to calculate the magnetic dipole contribution.

  • Added copy and as_pandas methods to containers

  • Added filter method to shgpy.core.data_handler.DataContainer().

  • Added apply_arbitrary_scale method to shgpy.core.data_handler.FormContainer() and shgpy.core.data_handler.fFormContainer().

  • Added multiple example files demonstrating the above.

v0.7.10

v0.7.8

  • Added the ability to apply Kleinman symmetry to a particular SHG tensor (i.e. enforce full permutation symmetry of the susceptibility indices). Use shgpy.particularize() with the option permute_all_indices=True

v0.7.6

v0.7.5

  • Fixed a number of tensors which were incorrectly defined due to typos in Boyd. The affected tensors were

    • D_6 (dipole)

    • C_4 (quadrupole)

    • C_4h (quadrupole)

    • S_4 (quadrupole)

v0.7.0

  • Officially transitioned to supporting only tensors for which all the involved symbols are purely real, as defined by sympy assumptions. This got rid of a lot of redundancy in function definitions, such as shgpy.fformgen.generate_contracted_fourier_transforms versus shgpy.fformgen.generate_contracted_fourier_transforms_complex, shgpy.formgen.formgen_dipole_quadrupole_real and shgpy.formgen.formgen_dipole_quadrupole_complex, ect. In all cases, these functions have been replaced by a single function, e.g. shgpy.formgen.formgen_dipole_quadrupole, and you will receive a NotImplementedError if you try to use any of the replaced definitions.

  • To aid in explicitly defining the reality of SHG tensors, added shgpy.make_tensor_real to complement shgpy.make_tensor_complex.

  • Transitioned to compiling cost functions at runtime by generating C code with sympy.ulities.codegen. This is a workaround to the fact that complicated sympy.lambdify functions are very slow to evaluate.

  • Added the ability to generate a cost function independently with shgpy.fformfit.gen_cost_func() and use it in one of the fitting routines by the load_cost_func_filename argument.

  • Added shgpy.fformfit.dual_annealing_fit and shgpy.fformfit.dual_annealing_fit_with_bounds.

v0.6.1

  • Added the ability to optionally send arguments to the scipy.optimize.basinhopping function. This is useful e.g. for debugging – use

>>> basinhopping_kwargs = {'disp':True}

to send disp=True to scipy.optimize.basinhopping, which initializes verbose output.

v0.5.1

  • Fixed a bug related to the change in v0.5.0 in which shgpy.load_fform() wasn’t compatible with the new pickling scheme.

v0.5.0

  • In .p file handling, switched from pickling pure sympy expressions in shgpy.fformgen to pickling string represetations of those expressions generated using sympy.srepr. This is a workaround to a sympy / pickle bug in which unpickling sympy expressions will cause sympy to conflate objects like sympy.Symbol('x') with sympy.Symbol('x', real=True).

  • To use v0.5.0, you will have to remake your .p files.