sparse-ir 1.1.7__py3-none-any.whl → 2.0.0a2__py3-none-any.whl

sparse_ir/poly.py

@@ -1,14 +1,137 @@
-# Copyright (C) 2020-2022 Markus Wallerberger, Hiroshi Shinaoka, and others
-# SPDX-License-Identifier: MIT
+"""
+Piecewise polynomial functionality for SparseIR.
+
+This module provides piecewise Legendre polynomial representation and
+their Fourier transforms, which serve as core mathematical infrastructure
+for IR basis functions.
+"""
+
+from ctypes import c_int, POINTER
 import numpy as np
-from warnings import warn
-import numpy.polynomial.legendre as np_legendre
-import scipy.special as sp_special
+import weakref
+import threading
+
+from pylibsparseir.core import _lib
+from pylibsparseir.core import funcs_eval_single_float64, funcs_eval_single_complex128
+from pylibsparseir.core import funcs_get_size, funcs_get_roots
+
+# Global registry to track pointer usage
+_pointer_registry = weakref.WeakSet()
+_registry_lock = threading.Lock()
+
+def funcs_get_slice(funcs_ptr, indices):
+    status = c_int()
+    indices = np.asarray(indices, dtype=np.int32)
+    funcs = _lib.spir_funcs_get_slice(funcs_ptr, len(indices), indices.ctypes.data_as(POINTER(c_int)), status)
+    if status.value != 0:
+        raise RuntimeError(f"Failed to get basis function {indices}: {status.value}")
+    return FunctionSet(funcs)
+
+def funcs_ft_get_slice(funcs_ptr, indices):
+    status = c_int()
+    indices = np.asarray(indices, dtype=np.int32)
+    funcs = _lib.spir_funcs_get_slice(funcs_ptr, len(indices), indices.ctypes.data_as(POINTER(c_int)), status)
+    if status.value != 0:
+        raise RuntimeError(f"Failed to get basis function {indices}: {status.value}")
+    return FunctionSetFT(funcs)
+
+class FunctionSet:
+    """Wrapper for basis function evaluation."""
+
+    def __init__(self, funcs_ptr):
+        self._ptr = funcs_ptr
+        self._released = False
+        # Register this object for safe cleanup
+        with _registry_lock:
+            _pointer_registry.add(self)
 
-from . import _util
-from . import _roots
-from . import _gauss
+    def __call__(self, x):
+        """Evaluate basis functions at given points."""
+        if self._released:
+            raise RuntimeError("Function set has been released")
+        if not isinstance(x, np.ndarray):
+            o = funcs_eval_single_float64(self._ptr, x)
+            if len(o) == 1:
+                return o[0]
+            else:
+                return o
+        else:
+            o = np.stack([funcs_eval_single_float64(self._ptr, e) for e in x]).T
+            if len(o) == 1:
+                return o[0]
+            else:
+                return o
+
+    def __getitem__(self, index):
+        """Get a single basis function."""
+        if self._released:
+            raise RuntimeError("Function set has been released")
+        sz = funcs_get_size(self._ptr)
+        return funcs_get_slice(self._ptr, [index % sz])
+
+    def release(self):
+        """Manually release the function set."""
+        if not self._released and self._ptr:
+            try:
+                _lib.spir_funcs_release(self._ptr)
+            except:
+                pass
+            self._released = True
+            self._ptr = None
+
+    def __del__(self):
+        # Only release if we haven't been released yet
+        if not self._released:
+            self.release()
+
+class FunctionSetFT:
+    """Wrapper for basis function evaluation."""
+
+    def __init__(self, funcs_ptr):
+        self._ptr = funcs_ptr
+        self._released = False
+        # Register this object for safe cleanup
+        with _registry_lock:
+            _pointer_registry.add(self)
 
+    def __call__(self, x):
+        """Evaluate basis functions at given points."""
+        if self._released:
+            raise RuntimeError("Function set has been released")
+        if not isinstance(x, np.ndarray):
+            o = funcs_eval_single_complex128(self._ptr, x)
+            if len(o) == 1:
+                return o[0]
+            else:
+                return o
+        else:
+            o = np.stack([funcs_eval_single_complex128(self._ptr, e) for e in x]).T
+            if len(o) == 1:
+                return o[0]
+            else:
+                return o
+
+    def __getitem__(self, index):
+        """Get a single basis function."""
+        if self._released:
+            raise RuntimeError("Function set has been released")
+        sz = funcs_get_size(self._ptr)
+        return funcs_ft_get_slice(self._ptr, [index % sz])
+
+    def release(self):
+        """Manually release the function set."""
+        if not self._released and self._ptr:
+            try:
+                _lib.spir_funcs_release(self._ptr)
+            except:
+                pass
+            self._released = True
+            self._ptr = None
+
+    def __del__(self):
+        # Only release if we haven't been released yet
+        if not self._released:
+            self.release()
 
 class PiecewiseLegendrePoly:
     """Piecewise Legendre polynomial.
@@ -17,92 +140,34 @@ class PiecewiseLegendrePoly:
     intervals ``S[i] = [a[i], a[i+1]]``, where on each interval the function
     is expanded in scaled Legendre polynomials.
     """
-    def __init__(self, data, knots, dx=None, symm=None):
-        """Piecewise Legendre polynomial"""
-        if np.isnan(data).any():
-            raise ValueError("PiecewiseLegendrePoly: data contains NaN!")
-        if isinstance(knots, self.__class__):
-            if dx is not None or symm is None:
-                raise RuntimeError("wrong arguments")
-            self.__dict__.update(knots.__dict__)
-            self.data = data
-            self.symm = symm
-            return
-
-        data = np.array(data)
-        knots = np.array(knots)
-        polyorder, nsegments = data.shape[:2]
-        if knots.shape != (nsegments+1,):
-            raise ValueError("Invalid knots array")
-        if not (knots[1:] >= knots[:-1]).all():
-            raise ValueError("Knots must be monotonically increasing")
-        if symm is None:
-            # TODO: infer symmetry from data
-            symm = np.zeros(data.shape[2:])
-        else:
-            symm = np.array(symm)
-            if symm.shape != data.shape[2:]:
-                raise ValueError("shape mismatch")
-        if dx is None:
-            dx = knots[1:] - knots[:-1]
-        else:
-            dx = np.array(dx)
-            if not np.allclose(dx, knots[1:] - knots[:-1]):
-                raise ValueError("dx must work with knots")
-
-        self.nsegments = nsegments
-        self.polyorder = polyorder
-        self.xmin = knots[0]
-        self.xmax = knots[-1]
-
-        self.knots = knots
-        self.dx = dx
-        self.data = data
-        self.symm = symm
-        self._xm = .5 * (knots[1:] + knots[:-1])
-        self._inv_xs = 2/dx
-        self._norm = np.sqrt(self._inv_xs)
-
-    def __getitem__(self, l):
-        """Return part of a set of piecewise polynomials"""
-        new_symm = self.symm[l]
-        if isinstance(l, tuple):
-            new_data = self.data[(slice(None), slice(None), *l)]
-        else:
-            new_data = self.data[:,:,l]
-        return self.__class__(new_data, self, symm=new_symm)
+
+    def __init__(self, funcs: FunctionSet, xmin: float, xmax: float):
+        self._funcs = funcs
+        self._xmin = xmin
+        self._xmax = xmax
 
     def __call__(self, x):
-        """Evaluate polynomial at position x"""
-        i, xtilde = self._split(np.asarray(x))
-        data = self.data[:, i]
-
-        # Evaluate for all values of l.  x and data array must be
-        # broadcast'able against each other, so we append dimensions here
-        func_dims = self.data.ndim - 2
-        datashape = i.shape + (1,) * func_dims
-        res = np_legendre.legval(xtilde.reshape(datashape), data, tensor=False)
-        res *= self._norm[i.reshape(datashape)]
-
-        # Finally, exchange the x and vector dimensions
-        order = tuple(range(i.ndim, i.ndim + func_dims)) + tuple(range(i.ndim))
-        return res.transpose(*order)
-
-    def value(self, l, x):
-        """Return value for l and x."""
-        if self.data.ndim != 3:
-            raise ValueError("Only allowed for vector of data")
-
-        l, x = np.broadcast_arrays(l, x)
-        i, xtilde = self._split(x)
-        data = self.data[:, i, l]
-
-        # This should now neatly broadcast against each other
-        res = np_legendre.legval(xtilde, data, tensor=False)
-        res *= self._norm[i]
-        return res
-
-    def overlap(self, f, *, rtol=2.3e-16, return_error=False, points=None):
+        """Evaluate basis functions at given points."""
+        return self._funcs(x)
+
+
+class PiecewiseLegendrePolyVector:
+    """Piecewise Legendre polynomial vector."""
+
+    def __init__(self, funcs: FunctionSet, xmin: float, xmax: float):
+        self._funcs = funcs
+        self._xmin = xmin
+        self._xmax = xmax
+
+    def __call__(self, x):
+        """Evaluate basis functions at given points."""
+        return self._funcs(x)
+
+    def __getitem__(self, index):
+        """Get a single basis function."""
+        return PiecewiseLegendrePoly(self._funcs[index], self._xmin, self._xmax)
+
+    def overlap(self, f, n_points=100):
         r"""Evaluate overlap integral of this polynomial with function ``f``.
 
         Given the function ``f``, evaluate the integral::
@@ -116,86 +181,59 @@ class PiecewiseLegendrePoly:
             f (callable):
                 function that is called with a point ``x`` and returns ``f(x)``
                 at that position.
-
-            points (sequence of floats)
-                A sequence of break points in the integration interval
-                where local difficulties of the integrand may occur
-                (e.g., singularities, discontinuities)
+            n_points (int):
+                Number of quadrature points per integration segment.
 
         Return:
             array-like object with shape (poly_dims, f_dims)
             poly_dims are the shape of the polynomial and f_dims are those
             of the function f(x).
         """
-        int_result, int_error = _compute_overlap(self, f, rtol=rtol, points=points)
-        if return_error:
-            return int_result, int_error
-        else:
-            return int_result
-
-    def deriv(self, n=1):
-        """Get polynomial for the n'th derivative"""
-        ddata = np_legendre.legder(self.data, n)
-
-        _scale_shape = (1, -1) + (1,) * (self.data.ndim - 2)
-        scale = self._inv_xs ** n
-        ddata *= scale.reshape(_scale_shape)
-        return self.__class__(ddata, self, symm=(-1)**n * self.symm)
-
-    def roots(self, alpha=2):
-        """Find all roots of the piecewise polynomial
-
-        Assume that between each two knots (pieces) there are at most ``alpha``
-        roots.
-        """
-        if self.data.ndim > 2:
-            raise ValueError("select single polynomial before calling roots()")
-
-        grid = self.knots
-        xmid = (self.xmax + self.xmin) / 2
-        if self.symm:
-            if grid[self.nsegments // 2] == xmid:
-                grid = grid[self.nsegments//2:]
-            else:
-                grid = np.hstack((xmid, grid[grid > xmid]))
-
-        grid = _refine_grid(grid, alpha)
-        roots = _roots.find_all(self, grid)
-
-        if self.symm == 1:
-            revroots = (self.xmax + self.xmin) - roots[::-1]
-            roots = np.hstack((revroots, roots))
-        elif self.symm == -1:
-            # There must be a zero at exactly the midpoint, but we may either
-            # slightly miss it or have a spurious zero
-            if roots.size:
-                if roots[0] == xmid or self(xmid) * self.deriv()(xmid) < 0:
-                    roots = roots[1:]
-            revroots = (self.xmax + self.xmin) - roots[::-1]
-            roots = np.hstack((revroots, xmid, roots))
-
-        return roots
-
-    @property
-    def shape(self): return self.data.shape[2:]
-
-    @property
-    def size(self): return self.data[:1,:1].size
-
-    @property
-    def ndim(self): return self.data.ndim - 2
-
-    def _split(self, x):
-        """Split segment"""
-        x = _util.check_range(x, self.xmin, self.xmax)
-        i = self.knots.searchsorted(x, 'right').clip(None, self.nsegments)
-        i -= 1
-        xtilde = x - self._xm[i]
-        xtilde *= self._inv_xs[i]
-        return i, xtilde
-
-
-class PiecewiseLegendreFT:
+        from scipy.integrate import fixed_quad
+
+        xmin = self._xmin
+        xmax = self._xmax
+        roots = funcs_get_roots(self._funcs._ptr).tolist()
+        roots.sort()
+
+        # Create integration segments
+        segments = [xmin] + roots + [xmax]
+        segments = sorted(list(set(segments)))  # Remove duplicates and sort
+
+        # Collect all quadrature points and weights
+        all_x = []
+        all_weights = []
+        
+        for j in range(len(segments) - 1):
+            a, b = segments[j], segments[j+1]
+            if abs(b - a) < 1e-14:  # Skip zero-length segments
+                continue
+            
+            # Get Gauss-Legendre quadrature points and weights
+            from scipy.special import roots_legendre
+            x_quad, w_quad = roots_legendre(n_points)
+            # Scale to actual interval
+            x_scaled = (b - a) / 2 * x_quad + (a + b) / 2
+            w_scaled = w_quad * (b - a) / 2
+            
+            all_x.extend(x_scaled)
+            all_weights.extend(w_scaled)
+
+        # Convert to numpy arrays for batch processing
+        all_x = np.array(all_x)
+        all_weights = np.array(all_weights)
+
+        # Evaluate function and polynomials at all points
+        f_values = f(all_x)  # This should work with array input
+        poly_values = self._funcs(all_x)  # Shape: (n_polys, n_points)
+
+        # Compute overlap integrals
+        output = np.sum(poly_values * f_values * all_weights, axis=1)
+
+        return output
+
+
+class PiecewiseLegendrePolyFT:
     """Fourier transform of a piecewise Legendre polynomial.
 
     For a given frequency index ``n``, the Fourier transform of the Legendre
@@ -207,319 +245,26 @@ class PiecewiseLegendreFT:
     case ``n`` must be even, or antiperiodically (``freq='odd'``), in which case
     ``n`` must be odd.
     """
-    _DEFAULT_GRID = np.hstack([np.arange(2**6),
-                        (2**np.linspace(6, 35, 16*(35-6)+1)).astype(int)])
-
-    def __init__(self, poly, freq='even', n_asymp=None, power_model=None):
-        if poly.xmin != -1 or poly.xmax != 1:
-            raise NotImplementedError("Only interval [-1, 1] supported")
-        self.poly = poly
-        self.freq = freq
-        self.zeta = {'any': None, 'even': 0, 'odd': 1}[freq]
-        if n_asymp is None:
-            self.n_asymp = np.inf
-            self._model = None
-        else:
-            self.n_asymp = n_asymp
-            if power_model is None:
-                self._model = _power_model(freq, poly)
-            else:
-                self._model = power_model
-
-    @property
-    def shape(self): return self.poly.shape
-
-    @property
-    def size(self): return self.poly.size
-
-    @property
-    def ndim(self): return self.poly.ndim
-
-    def __getitem__(self, l):
-        model = self._model if self._model is None else self._model[l]
-        return self.__class__(self.poly[l], self.freq, self.n_asymp, model)
-
-    @_util.ravel_argument(last_dim=True)
-    def __call__(self, n):
-        """Obtain Fourier transform of polynomial for given frequencies"""
-        n = _util.check_reduced_matsubara(n, self.zeta)
-        result = _compute_unl_inner(self.poly, n)
-
-        # We use the asymptotics at frequencies larger than conv_radius
-        # since it has lower relative error.
-        cond_outer = np.abs(n) >= self.n_asymp
-        if cond_outer.any():
-            n_outer = n[cond_outer]
-            result[..., cond_outer] = self._model.giw(n_outer).T
-
-        return result
-
-    def extrema(self, *, part=None, grid=None, positive_only=False):
-        """Obtain extrema of Fourier-transformed polynomial."""
-        if self.poly.shape:
-            raise ValueError("select single polynomial")
-        if grid is None:
-            grid = self._DEFAULT_GRID
-
-        f = self._func_for_part(part)
-        x0 = _roots.discrete_extrema(f, grid)
-        x0 = 2 * x0 + self.zeta
-        if not positive_only:
-            x0 = _symmetrize_matsubara(x0)
-        return x0
-
-    def sign_changes(self, *, part=None, grid=None, positive_only=False):
-        """Obtain sign changes of Fourier-transformed polynomial."""
-        if self.poly.shape:
-            raise ValueError("select single polynomial")
-        if grid is None:
-            grid = self._DEFAULT_GRID
-
-        f = self._func_for_part(part)
-        x0 = _roots.find_all(f, grid, type='discrete')
-        x0 = 2 * x0 + self.zeta
-        if not positive_only:
-            x0 = _symmetrize_matsubara(x0)
-        return x0
-
-    def _func_for_part(self, part=None):
-        if part is None:
-            parity = self.poly.symm
-            if np.allclose(parity, 1):
-                part = 'real' if self.zeta == 0 else 'imag'
-            elif np.allclose(parity, -1):
-                part = 'imag' if self.zeta == 0 else 'real'
-            else:
-                raise ValueError("cannot detect parity.")
-        if part == 'real':
-            return lambda n: self(2*n + self.zeta).real
-        elif part == 'imag':
-            return lambda n: self(2*n + self.zeta).imag
-        else:
-            raise ValueError("part must be either 'real' or 'imag'")
 
+    def __init__(self, funcs: FunctionSetFT):
+        assert isinstance(funcs, FunctionSetFT), "funcs must be a FunctionSetFT"
+        self._funcs = funcs
 
+    def __call__(self, x):
+        """Evaluate basis functions at given points."""
+        return self._funcs(x)
 
-def _imag_power(n):
-    """Imaginary unit raised to an integer power without numerical error"""
-    n = np.asarray(n)
-    if not np.issubdtype(n.dtype, np.integer):
-        raise ValueError("expecting set of integers here")
-    cycle = np.array([1, 0+1j, -1, 0-1j], complex)
-    return cycle[n % 4]
-
-
-def _get_tnl(l, w):
-    r"""Fourier integral of the l-th Legendre polynomial::
-
-        T_l(w) == \int_{-1}^1 dx \exp(iwx) P_l(x)
-    """
-    # spherical_jn gives NaN for w < 0, but since we know that P_l(x) is real,
-    # we simply conjugate the result for w > 0 in these cases.
-    result = 2 * _imag_power(l) * sp_special.spherical_jn(l, np.abs(w))
-    np.conjugate(result, out=result, where=w < 0)
-    return result
-
-
-def _shift_xmid(knots, dx):
-    r"""Return midpoint relative to the nearest integer plus a shift.
-
-    Return the midpoints ``xmid`` of the segments, as pair ``(diff, shift)``,
-    where shift is in ``(0,1,-1)`` and ``diff`` is a float such that
-    ``xmid == shift + diff`` to floating point accuracy.
-    """
-    dx_half = dx / 2
-    xmid_m1 = dx.cumsum() - dx_half
-    xmid_p1 = -dx[::-1].cumsum()[::-1] + dx_half
-    xmid_0 = knots[1:] - dx_half
-
-    shift = np.round(xmid_0).astype(int)
-    diff = np.choose(shift+1, (xmid_m1, xmid_0, xmid_p1))
-    return diff, shift
-
+class PiecewiseLegendrePolyFTVector:
+    """Fourier transform of a piecewise Legendre polynomial vector."""
 
-def _phase_stable(poly, wn):
-    """Phase factor for the piecewise Legendre to Matsubara transform.
+    def __init__(self, funcs: FunctionSetFT):
+        assert isinstance(funcs, FunctionSetFT), "funcs must be a FunctionSetFT"
+        self._funcs = funcs
 
-    Compute the following phase factor in a stable way::
+    def __call__(self, x: np.ndarray) -> np.ndarray:
+        """Evaluate basis functions at given points."""
+        return self._funcs(x)
 
-        np.exp(1j * np.pi/2 * wn[:,None] * poly.dx.cumsum()[None,:])
-    """
-    # A naive implementation is losing precision close to x=1 and/or x=-1:
-    # there, the multiplication with `wn` results in `wn//4` almost extra turns
-    # around the unit circle.  The cosine and sines will first map those
-    # back to the interval [-pi, pi) before doing the computation, which loses
-    # digits in dx.  To avoid this, we extract the nearest integer dx.cumsum()
-    # and rewrite above expression like below.
-    #
-    # Now `wn` still results in extra revolutions, but the mapping back does
-    # not cut digits that were not there in the first place.
-    xmid_diff, extra_shift = _shift_xmid(poly.knots, poly.dx)
-
-    if np.issubdtype(wn.dtype, np.integer):
-        shift_arg = wn[None,:] * xmid_diff[:,None]
-    else:
-        delta_wn, wn = np.modf(wn)
-        wn = wn.astype(int)
-        shift_arg = wn[None,:] * xmid_diff[:,None]
-        shift_arg += delta_wn[None,:] * (extra_shift + xmid_diff)[:,None]
-
-    phase_shifted = np.exp(0.5j * np.pi * shift_arg)
-    corr = _imag_power((extra_shift[:,None] + 1) * wn[None,:])
-    return corr * phase_shifted
-
-
-def _compute_unl_inner(poly, wn):
-    """Compute piecewise Legendre to Matsubara transform."""
-    dx_half = poly.dx / 2
-
-    data_flat = poly.data.reshape(*poly.data.shape[:2], -1)
-    data_sc = data_flat * np.sqrt(dx_half/2)[None,:,None]
-    p = np.arange(poly.polyorder)
-
-    wred = np.pi/2 * wn
-    phase_wi = _phase_stable(poly, wn)
-    t_pin = _get_tnl(p[:,None,None], wred[None,:] * dx_half[:,None]) * phase_wi
-
-    # Perform the following, but faster:
-    #   resulth = einsum('pin,pil->nl', t_pin, data_sc)
-    npi = poly.polyorder * poly.nsegments
-    result_flat = (t_pin.reshape(npi,-1).T @ data_sc.reshape(npi,-1)).T
-    return result_flat.reshape(*poly.data.shape[2:], wn.size)
-
-
-class _PowerModel:
-    """Model from a high-frequency series expansion::
-
-        A(iw) == sum(A[n] / (iw)**(n+1) for n in range(1, N))
-
-    where ``iw == 1j * pi/2 * wn`` is a reduced imaginary frequency, i.e.,
-    ``wn`` is an odd/even number for fermionic/bosonic frequencies.
-    """
-    def __init__(self, moments):
-        """Initialize model"""
-        if moments.ndim == 1:
-            moments = moments[:, None]
-        self.moments = np.asarray(moments)
-        self.nmom, self.nl = self.moments.shape
-
-    @_util.ravel_argument()
-    def giw(self, wn):
-        """Return model Green's function for vector of frequencies"""
-        wn = _util.check_reduced_matsubara(wn)
-        result_dtype = np.result_type(1j, wn, self.moments)
-        result = np.zeros((wn.size, self.nl), result_dtype)
-        inv_iw = 1j * np.pi/2 * wn
-        np.reciprocal(inv_iw, out=inv_iw, where=(wn != 0))
-        for mom in self.moments[::-1]:
-            result += mom
-            result *= inv_iw[:, None]
-        return result
-
-    def __getitem__(self, l):
-        return self.__class__(self.moments[:,l])
-
-
-def _derivs(ppoly, x):
-    """Evaluate polynomial and its derivatives at specific x"""
-    yield ppoly(x)
-    for _ in range(ppoly.polyorder-1):
-        ppoly = ppoly.deriv()
-        yield ppoly(x)
-
-
-def _power_moments(stat, deriv_x1):
-    """Return moments"""
-    statsign = {'odd': -1, 'even': 1}[stat]
-    mmax, lmax = deriv_x1.shape
-    m = np.arange(mmax)[:,None]
-    l = np.arange(lmax)[None,:]
-    coeff_lm = ((-1.0)**(m+1) + statsign * (-1.0)**l) * deriv_x1
-    return -statsign/np.sqrt(2.0) * coeff_lm
-
-
-def _power_model(stat, poly):
-    deriv_x1 = np.asarray(list(_derivs(poly, x=1)))
-    if deriv_x1.ndim == 1:
-        deriv_x1 = deriv_x1[:,None]
-    moments = _power_moments(stat, deriv_x1)
-    return _PowerModel(moments)
-
-
-def _refine_grid(knots, alpha):
-    """Linear refinement of grid"""
-    result = np.linspace(knots[:-1], knots[1:], alpha, endpoint=False)
-    return np.hstack((result.T.ravel(), knots[-1]))
-
-
-def _symmetrize_matsubara(x0):
-    if not x0.size:
-        return x0
-    if not (x0[1:] >= x0[:-1]).all():
-        raise ValueError("set of Matsubara points not ordered")
-    if not (x0[0] >= 0):
-        raise ValueError("points must be non-negative")
-    if x0[0] == 0:
-        x0 = np.hstack([-x0[::-1], x0[1:]])
-    else:
-        x0 = np.hstack([-x0[::-1], x0])
-    return x0
-
-
-def _compute_overlap(poly, f, rtol=2.3e-16, radix=2, max_refine_levels=40,
-                     max_refine_points=2000, points=None):
-    base_rule = _gauss.kronrod_31_15()
-    if points is None:
-        knots = poly.knots
-    else:
-        points = np.asarray(points)
-        knots = np.unique(np.hstack((poly.knots, points)))
-    xstart = knots[:-1]
-    xstop = knots[1:]
-
-    f_shape = None
-    res_value = 0
-    res_error = 0
-    res_magn = 0
-    for _ in range(max_refine_levels):
-        #print(f"Level {_}: {xstart.size} segments")
-        if xstart.size > max_refine_points:
-            warn("Refinement is too broad, aborting (increase rtol)")
-            break
-
-        rule = base_rule.reseat(xstart[:, None], xstop[:, None])
-
-        fx = np.array(list(map(f, rule.x.ravel())))
-        if f_shape is None:
-            f_shape = fx.shape[1:]
-        elif fx.shape[1:] != f_shape:
-            raise ValueError("inconsistent shapes")
-        fx = fx.reshape(rule.x.shape + (-1,))
-
-        valx = poly(rule.x).reshape(-1, *rule.x.shape, 1) * fx
-        int21 = (valx[:, :, :, :] * rule.w[:, :, None]).sum(2)
-        int10 = (valx[:, :, rule.vsel, :] * rule.v[:, :, None]).sum(2)
-        intdiff = np.abs(int21 - int10)
-        intmagn = np.abs(int10)
-
-        magn = res_magn + intmagn.sum(1).max(1)
-        relerror = intdiff.max(2) / magn[:, None]
-
-        xconverged = (relerror <= rtol).all(0)
-        res_value += int10[:, xconverged].sum(1)
-        res_error += intdiff[:, xconverged].sum(1)
-        res_magn += intmagn[:, xconverged].sum(1).max(1)
-        if xconverged.all():
-            break
-
-        xrefine = ~xconverged
-        xstart = xstart[xrefine]
-        xstop = xstop[xrefine]
-        xedge = np.linspace(xstart, xstop, radix + 1, axis=-1)
-        xstart = xedge[:, :-1].ravel()
-        xstop = xedge[:, 1:].ravel()
-    else:
-        warn("Integration did not converge after refinement")
-
-    res_shape = poly.shape + f_shape
-    return res_value.reshape(res_shape), res_error.reshape(res_shape)
+    def __getitem__(self, index):
+        """Get a single basis function."""
+        return PiecewiseLegendrePolyFT(self._funcs[index])