Rader's algorithm Imaginary Part Sign Issue

[Danc2050]

Comparing Rader's algorithm to the Jax's FFT (jnp.fft.fft), the arrays are nearly the same, except for the sign in the imaginary part is often flipped. Here is an example for two arrays:

rader: [ 5.0450296 +0.j         -0.88408756+1.5039028j  -0.25362217+0.112124j
 -0.19672948+0.22919299j  0.38429442-0.5026276j  -0.09411531-0.15777384j
 -0.09411531+0.15777384j  0.38429442+0.5026276j  -0.19672948-0.22919299j
 -0.25362217-0.112124j   -0.88408756-1.5039028j ]
jax.numpy.fft.fft: [ 5.04503   +0.j         -0.88408756-1.5039027j  -0.25362217-0.112124j
 -0.19672951-0.22919303j  0.3842943 +0.5026276j  -0.09411524+0.15777388j
 -0.09411524-0.15777388j  0.3842943 -0.5026276j  -0.19672951+0.22919303j
 -0.25362217+0.112124j   -0.88408756+1.5039027j ]

Here is a GPT produced description of the differences: