"""Exact rational-exponent examples of finite-jet recovery and layer peeling.

An exponential polynomial is {exponent: coefficient}, interpreted as
sum(coefficient*exp(exponent*z)). This test implementation uses exact SymPy
rational arithmetic. The manuscript's uniqueness theorem allows arbitrary
positive real layer parameters; this implementation deliberately does not
claim robust inversion of noisy data or arbitrary algebraic root finding.
"""
from __future__ import annotations
import sympy as sp


def clean(f):
    return {sp.Rational(k): sp.cancel(v) for k, v in f.items() if sp.cancel(v) != 0}


def add(f, g, scale=1):
    out = dict(f)
    for x, v in g.items():
        out[x] = out.get(x, 0) + scale * v
    return clean(out)


def mul(f, g):
    out = {}
    for x, a in f.items():
        for y, b in g.items():
            out[x+y] = out.get(x+y, 0) + a*b
    return clean(out)


def scaled(f, a):
    return clean({x: a*v for x, v in f.items()})


def cell(c, tau):
    c, tau = sp.Rational(c), sp.Rational(tau)
    if c <= 0 or tau <= 0:
        raise ValueError('Layer speeds and durations must be positive.')
    ell = c*tau
    co = {ell: sp.Rational(1, 2), -ell: sp.Rational(1, 2)}
    si = {ell: sp.Rational(1, 2), -ell: sp.Rational(-1, 2)}
    return [[co, scaled(si, 1/c)], [scaled(si, c), co]]


def matmul(A, B):
    return [[add(mul(A[i][0], B[0][j]), mul(A[i][1], B[1][j]))
             for j in range(2)] for i in range(2)]


def layer_product(layers):
    H = [[{sp.Rational(0): sp.Rational(1)}, {}],
         [{}, {sp.Rational(0): sp.Rational(1)}]]
    for c, tau in layers:
        H = matmul(cell(c, tau), H)
    return H


def moments(f, count):
    return [sp.cancel(sum(a*x**m for x, a in f.items())) for m in range(count)]


def recover_exponential_polynomial(mu, max_terms):
    """Prony recovery from 2*max_terms exact moments; handles cancellations."""
    if len(mu) < 2*max_terms:
        raise ValueError('At least 2*max_terms moments are required.')
    if all(v == 0 for v in mu[:2*max_terms]):
        return {}
    d = 0
    for j in range(1, max_terms+1):
        Hj = sp.Matrix(j, j, lambda a, b: mu[a+b])
        if Hj.det() != 0:
            d = j
    if not d:
        raise ValueError('The supplied moments are inconsistent with the model.')
    H = sp.Matrix(d, d, lambda a, b: mu[a+b])
    coeffs = H.inv() * -sp.Matrix(mu[d:2*d])
    x = sp.Symbol('x')
    poly = sp.Poly(x**d + sum(coeffs[j]*x**j for j in range(d)), x)
    roots = sp.roots(poly)
    if sum(roots.values()) != d or len(roots) != d or any(not r.is_Rational for r in roots):
        raise NotImplementedError('This exact demonstration handles distinct rational exponents.')
    nodes = sorted(roots)
    V = sp.Matrix(d, d, lambda a, b: nodes[b]**a)
    amplitudes = V.inv() * sp.Matrix(mu[:d])
    f = clean(dict(zip(nodes, amplitudes)))
    if moments(f, 2*max_terms) != list(mu[:2*max_terms]):
        raise ValueError('Moment consistency check failed.')
    return f


def peel_first_row(f, g, max_layers):
    """Recover ordered (c, tau) by leading-exponent cancellation, not limits."""
    f, g = clean(f), clean(g)
    layers = []
    for _ in range(max_layers):
        if f == {sp.Rational(0): sp.Rational(1)} and not g:
            return layers
        if not f or not g or max(f) != max(g):
            raise ValueError('First row is inconsistent with a positive finite-layer medium.')
        c = sp.cancel(f[max(f)] / g[max(g)])
        minus, plus = add(f, g, -c), add(f, g, c)
        if not minus or not plus:
            raise ValueError('Invalid reflection numerator or denominator.')
        ell = sp.cancel((max(plus) - max(minus))/2)
        if c <= 0 or ell <= 0:
            raise ValueError('Nonpositive recovered speed or optical duration.')
        tau = sp.cancel(ell/c)
        co = {ell: sp.Rational(1, 2), -ell: sp.Rational(1, 2)}
        si = {ell: sp.Rational(1, 2), -ell: sp.Rational(-1, 2)}
        new_f = add(mul(f, co), mul(g, si), -c)
        new_g = add(mul(g, co), mul(f, si), -1/c)
        f, g = new_f, new_g
        layers.append((c, tau))
    if f == {sp.Rational(0): sp.Rational(1)} and not g:
        return layers
    raise ValueError('Residual medium remains after max_layers peeling steps.')