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

sparse_ir/kernel.py

@@ -1,139 +1,33 @@
-# Copyright (C) 2020-2022 Markus Wallerberger, Hiroshi Shinaoka, and others
-# SPDX-License-Identifier: MIT
-import numpy as np
-from typing import Callable
-
-
-class AbstractKernel:
-    r"""Integral kernel ``K(x, y)``.
-
-    Abstract base class for an integral kernel, i.e., a real binary function
-    ``K(x, y)`` used in a Fredhold integral equation of the first kind:
-
-    .. math::           u(x) = \int K(x, y) v(y) dy
-
-    where ``x ∈ [xmin, xmax]`` and ``y ∈ [ymin, ymax]``.  For its SVE to exist,
-    the kernel must be square-integrable, for its singular values to decay
-    exponentially, it must be smooth.
-
-    In general, the kernel is applied to a scaled spectral function ρ'(y) as:
-
-    .. math::            \int K(x, y) \rho'(y) dy,
-
-    where ρ'(y) = w(y) ρ(y).
-    """
-    def __call__(self, x, y, x_plus=None, x_minus=None):
-        """Evaluate kernel at point (x, y)
-
-        For given ``x, y``, return the value of ``K(x, y)``. The arguments may
-        be numpy arrays, in which case the function shall be evaluated over
-        the broadcast arrays.
-
-        The parameters ``x_plus`` and ``x_minus``, if given, shall contain the
-        values of ``x - xmin`` and ``xmax - x``, respectively.  This is useful
-        if either difference is to be formed and cancellation expected.
-        """
-        raise NotImplementedError()
-
-    def sve_hints(self, eps):
-        """Provide discretisation hints for the SVE routines.
-
-        Advises the SVE routines of discretisation parameters suitable in
-        tranforming the (infinite) SVE into an (finite) SVD problem.
-
-        See: :class:``AbstractSVEHints``.
-        """
-        raise NotImplementedError()
-
-    @property
-    def xrange(self):
-        """Tuple ``(xmin, xmax)`` delimiting the range of allowed x values"""
-        return -1, 1
+"""
+Kernel classes for SparseIR.
 
-    @property
-    def yrange(self):
-        """Tuple ``(ymin, ymax)`` delimiting the range of allowed y values"""
-        return -1, 1
-
-    @property
-    def is_centrosymmetric(self):
-        """Kernel is centrosymmetric.
-
-        Returns true if and only if ``K(x, y) == K(-x, -y)`` for all values of
-        ``x`` and ``y``.  This allows the kernel to be block-diagonalized,
-        speeding up the singular value expansion by a factor of 4.  Defaults
-        to false.
-        """
-        return False
-
-    def get_symmetrized(self, sign):
-        """Return symmetrized kernel ``K(x, y) + sign * K(x, -y)``.
+This module provides Python wrappers for kernel objects from the C library.
+"""
 
-        By default, this returns a simple wrapper over the current instance
-        which naively performs the sum.  You may want to override this to
-        avoid cancellation.
-        """
-        return ReducedKernel(self, sign)
-
-    @property
-    def ypower(self):
-        """Power with which the y coordinate scales."""
-        return 0
-
-    @property
-    def conv_radius(self):
-        """Convergence radius of the Matsubara basis asymptotic model.
-
-        For improved relative numerical accuracy, the IR basis functions on the
-        Matsubara axis ``basis.uhat(n)`` can be evaluated from an asymptotic
-        expression for ``abs(n) > conv_radius``.  If ``conv_radius`` is
-        ``None``, then the asymptotics are unused (the default).
-        """
-        return None
-
-    def weight_func(self, statistics: str) -> Callable[[np.ndarray], np.ndarray]:
-        """Return the weight function for given statistics"""
-        if statistics not in 'FB':
-            raise ValueError("statistics must be 'F' for fermions or 'B' for bosons")
-        return lambda x: np.ones_like(x)
-
-
-class AbstractSVEHints:
-    """Discretization hints for singular value expansion of a given kernel."""
-    @property
-    def segments_x(self):
-        """Segments for piecewise polynomials on the ``x`` axis.
-
-        List of segments on the ``x`` axis for the associated piecewise
-        polynomial.  Should reflect the approximate position of roots of a
-        high-order singular function in ``x``.
-        """
-        raise NotImplementedError()
-
-    @property
-    def segments_y(self):
-        """Segments for piecewise polynomials on the ``y`` axis.
+import ctypes
+from ctypes import c_int, c_double, byref
+import numpy as np
 
-        List of segments on the ``y`` axis for the associated piecewise
-        polynomial.  Should reflect the approximate position of roots of a
-        high-order singular function in ``y``.
-        """
-        raise NotImplementedError()
+from pylibsparseir.core import _lib
+from pylibsparseir.core import logistic_kernel_new, reg_bose_kernel_new
+from pylibsparseir.constants import COMPUTATION_SUCCESS
+from .abstract import AbstractKernel
 
-    @property
-    def ngauss(self):
-        """Gauss-Legendre order to use to guarantee accuracy"""
-        raise NotImplementedError()
 
-    @property
-    def nsvals(self):
-        """Upper bound for number of singular values
+def kernel_domain(kernel: AbstractKernel):
+    """Get the domain boundaries of a kernel."""
+    xmin = c_double()
+    xmax = c_double()
+    ymin = c_double()
+    ymax = c_double()
 
-        Upper bound on the number of singular values above the given
-        threshold, i.e., where ``s[l] >= eps * s[0]``.
-        """
-        raise NotImplementedError()
+    status = _lib.spir_kernel_domain(
+        kernel._ptr, byref(xmin), byref(xmax), byref(ymin), byref(ymax)
+    )
+    if status != COMPUTATION_SUCCESS:
+        raise RuntimeError(f"Failed to get kernel domain: {status}")
 
+    return xmin.value, xmax.value, ymin.value, ymax.value
 
 class LogisticKernel(AbstractKernel):
     r"""Fermionic/bosonic analytical continuation kernel.
@@ -155,97 +49,27 @@ class LogisticKernel(AbstractKernel):
     i.e., a rescaling of the spectral function with the weight function:
 
     .. math::  w(y) = \frac1{\tanh(\Lambda y/2)}.
-    """
-    def __init__(self, lambda_):
-        self.lambda_ = lambda_
-
-    def __call__(self, x, y, x_plus=None, x_minus=None):
-        x, y = _check_domain(self, x, y)
-        u_plus, u_minus, v = _compute_uv(self.lambda_, x, y, x_plus, x_minus)
-        return self._compute(u_plus, u_minus, v)
-
-    def sve_hints(self, eps):
-        return _SVEHintsLogistic(self, eps)
-
-    def _compute(self, u_plus, u_minus, v):
-        # By introducing u_\pm = (1 \pm x)/2 and v = lambda * y, we can write
-        # the kernel in the following two ways:
-        #
-        #    k = exp(-u_+ * v) / (exp(-v) + 1)
-        #      = exp(-u_- * -v) / (exp(v) + 1)
-        #
-        # We need to use the upper equation for v >= 0 and the lower one for
-        # v < 0 to avoid overflowing both numerator and denominator
-        abs_v = np.abs(v)
-        enum = np.exp(-abs_v * np.where(v > 0, u_plus, u_minus))
-        denom = 1 + np.exp(-abs_v)
-        return enum / denom
-
-    @property
-    def is_centrosymmetric(self):
-        return True
-
-    def get_symmetrized(self, sign):
-        if sign == -1:
-            return _LogisticKernelOdd(self, sign)
-        return super().get_symmetrized(sign)
-
-    @property
-    def conv_radius(self): return 40 * self.lambda_
-
-    def weight_func(self, statistics: str) -> Callable[[np.ndarray], np.ndarray]:
-        """
-        Return the weight function for given statistics.
-
-         - Fermion: `w(x) == 1`
-         - Boson: `w(y) == 1/tanh(Λ*y/2)`
-        """
-        if statistics not in "FB":
-            raise ValueError("invalid value of statistics argument")
-        if statistics == "F":
-            return lambda y: np.ones_like(y)
-        else:
-            return lambda y: 1/np.tanh(0.5*self.lambda_*y)
-
-
-class _SVEHintsLogistic(AbstractSVEHints):
-    def __init__(self, kernel, eps):
-        self.kernel = kernel
-        self.eps = eps
 
-    @property
-    def ngauss(self): return 10 if self.eps >= 1e-8 else 16
+    Parameters
+    ----------
+    lambda_ : float
+        Kernel cutoff Λ = β * ωmax
+    """
 
-    @property
-    def segments_x(self):
-        nzeros = max(int(np.round(15 * np.log10(self.kernel.lambda_))), 1)
-        diffs = 1./np.cosh(.143 * np.arange(nzeros))
-        zeros_pos = diffs.cumsum()
-        zeros_pos /= zeros_pos[-1]
-        return np.concatenate((-zeros_pos[::-1], [0], zeros_pos))
+    def __init__(self, lambda_):
+        """Initialize logistic kernel with cutoff lambda."""
+        self._lambda = float(lambda_)
+        self._ptr = logistic_kernel_new(self._lambda)
 
     @property
-    def segments_y(self):
-        # Zeros around -1 and 1 are distributed asymptotically identical
-        leading_diffs = np.array([
-            0.01523, 0.03314, 0.04848, 0.05987, 0.06703, 0.07028, 0.07030,
-            0.06791, 0.06391, 0.05896, 0.05358, 0.04814, 0.04288, 0.03795,
-            0.03342, 0.02932, 0.02565, 0.02239, 0.01951, 0.01699])
+    def lambda_(self):
+        """Kernel cutoff."""
+        return self._lambda
 
-        nzeros = max(int(np.round(20 * np.log10(self.kernel.lambda_))), 2)
-        if nzeros < 20:
-            leading_diffs = leading_diffs[:nzeros]
-        diffs = .25 / np.exp(.141 * np.arange(nzeros))
-        diffs[:leading_diffs.size] = leading_diffs
-        zeros = diffs.cumsum()
-        zeros = zeros[:-1] / zeros[-1]
-        zeros -= 1
-        return np.concatenate(([-1], zeros, [0], -zeros[::-1], [1]))
-
-    @property
-    def nsvals(self):
-        log10_lambda = max(1, np.log10(self.kernel.lambda_))
-        return int(np.round((25 + log10_lambda) * log10_lambda))
+    def __del__(self):
+        """Clean up kernel resources."""
+        if hasattr(self, '_ptr') and self._ptr:
+            _lib.spir_kernel_release(self._ptr)
 
 
 class RegularizedBoseKernel(AbstractKernel):
@@ -259,271 +83,24 @@ class RegularizedBoseKernel(AbstractKernel):
         K(x, y) = \frac{y \exp(-\Lambda y(x + 1)/2)}{\exp(-\Lambda y) - 1}
 
     Care has to be taken in evaluating this expression around ``y == 0``.
-    """
-    def __init__(self, lambda_):
-        self.lambda_ = lambda_
 
-    def __call__(self, x, y, x_plus=None, x_minus=None):
-        x, y = _check_domain(self, x, y)
-        u_plus, u_minus, v = _compute_uv(self.lambda_, x, y, x_plus, x_minus)
-        return self._compute(u_plus, u_minus, v)
-
-    def _compute(self, u_plus, u_minus, v):
-        # With "reduced variables" u, v we have:
-        #
-        #   K = -1/lambda * exp(-u_+ * v) * v / (exp(-v) - 1)
-        #     = -1/lambda * exp(-u_- * -v) * (-v) / (exp(v) - 1)
-        #
-        # where we again need to use the upper equation for v >= 0 and the
-        # lower one for v < 0 to avoid overflow.
-        abs_v = np.abs(v)
-        enum = np.exp(-abs_v * np.where(v >= 0, u_plus, u_minus))
-        dtype = v.dtype
-
-        # The expression ``v / (exp(v) - 1)`` is tricky to evaluate: firstly,
-        # it has a singularity at v=0, which can be cured by treating that case
-        # separately.  Secondly, the denominator loses precision around 0 since
-        # exp(v) = 1 + v + ..., which can be avoided using expm1(...)
-        not_tiny = abs_v >= 1e-200
-        denom = -np.ones_like(abs_v)
-        np.divide(abs_v, np.expm1(-abs_v, where=not_tiny),
-                  out=denom, where=not_tiny)
-        return -1/dtype.type(self.lambda_) * enum * denom
-
-    def sve_hints(self, eps):
-        return _SVEHintsRegularizedBose(self, eps)
-
-    @property
-    def is_centrosymmetric(self):
-        return True
-
-    def get_symmetrized(self, sign):
-        if sign == -1:
-            return _RegularizedBoseKernelOdd(self, sign)
-        return super().get_symmetrized(sign)
-
-    @property
-    def ypower(self): return 1
-
-    @property
-    def conv_radius(self): return 40 * self.lambda_
-
-    def weight_func(self, statistics: str) -> Callable[[np.ndarray], np.ndarray]:
-        """ Return the weight function for given statistics """
-        if statistics != "B":
-            raise ValueError("Kernel is designed for bosonic functions")
-        return lambda y: 1/y
-
-
-class _SVEHintsRegularizedBose(AbstractSVEHints):
-    def __init__(self, kernel, eps):
-        self.kernel = kernel
-        self.eps = eps
-
-    @property
-    def ngauss(self): return 10 if self.eps >= 1e-8 else 16
-
-    @property
-    def segments_x(self):
-        # Somewhat less accurate ...
-        nzeros = max(int(np.round(15 * np.log10(self.kernel.lambda_))), 15)
-        diffs = 1./np.cosh(.18 * np.arange(nzeros))
-        zeros_pos = diffs.cumsum()
-        zeros_pos /= zeros_pos[-1]
-        return np.concatenate((-zeros_pos[::-1], [0], zeros_pos))
-
-    @property
-    def segments_y(self):
-        # Zeros around -1 and 1 are distributed asymptotically identical
-        leading_diffs = np.array([
-            0.01363, 0.02984, 0.04408, 0.05514, 0.06268, 0.06679, 0.06793,
-            0.06669, 0.06373, 0.05963, 0.05488, 0.04987, 0.04487, 0.04005,
-            0.03553, 0.03137, 0.02758, 0.02418, 0.02115, 0.01846])
-
-        nzeros = max(int(np.round(20 * np.log10(self.kernel.lambda_))), 20)
-        i = np.arange(nzeros)
-        diffs = .12/np.exp(.0337 * i * np.log(i+1))
-        #diffs[:leading_diffs.size] = leading_diffs
-        zeros = diffs.cumsum()
-        zeros = zeros[:-1] / zeros[-1]
-        zeros -= 1
-        return np.concatenate(([-1], zeros, [0], -zeros[::-1], [1]))
-
-    @property
-    def nsvals(self):
-        log10_lambda = max(1, np.log10(self.kernel.lambda_))
-        return int(28 * log10_lambda)
-
-
-class ReducedKernel(AbstractKernel):
-    """Restriction of centrosymmetric kernel to positive interval.
-
-    For a kernel ``K`` on ``[-1, 1] x [-1, 1]`` that is centrosymmetric, i.e.
-    ``K(x, y) == K(-x, -y)``, it is straight-forward to show that the left/right
-    singular vectors can be chosen as either odd or even functions.
-
-    Consequentially, they are singular functions of a reduced kernel ``K_red``
-    on ``[0, 1] x [0, 1]`` that is given as either::
-
-        K_red(x, y) == K(x, y) + sign * K(x, -y)
-
-    This kernel is what this class represents.  The full singular functions can
-    be reconstructed by (anti-)symmetrically continuing them to the negative
-    axis.
+    Parameters
+    ----------
+    lambda_ : float
+        Kernel cutoff Λ = β * ωmax
     """
-    def __init__(self, inner, sign=1):
-        if not inner.is_centrosymmetric:
-            raise ValueError("inner kernel must be centrosymmetric")
-        if np.abs(sign) != 1:
-            raise ValueError("sign must square to one")
-
-        self.inner = inner
-        self.sign = sign
-
-    def __call__(self, x, y, x_plus=None, x_minus=None):
-        x, y = _check_domain(self, x, y)
-
-        # The reduced kernel is defined only over the interval [0, 1], which
-        # means we must add one to get the x_plus for the inner kernels.  We
-        # can compute this as 1 + x, since we are away from -1.
-        x_plus = 1 + x_plus
-
-        K_plus = self.inner(x, y, x_plus, x_minus)
-        K_minus = self.inner(x, -y, x_plus, x_minus)
-        return K_plus + K_minus if self.sign == 1 else K_plus - K_minus
-
-    @property
-    def xrange(self):
-        _, xmax = self.inner.xrange
-        return 0, xmax
-
-    @property
-    def yrange(self):
-        _, ymax = self.inner.yrange
-        return 0, ymax
-
-    def sve_hints(self, eps):
-        return _SVEHintsReduced(self.inner.sve_hints(eps))
-
-    @property
-    def is_centrosymmetric(self):
-        """True iff K(x,y) = K(-x, -y)"""
-        return False
-
-    def get_symmetrized(self, sign):
-        raise RuntimeError("cannot symmetrize twice")
-
-    @property
-    def ypower(self): return self.inner.ypower
-
-    @property
-    def conv_radius(self): return self.inner.conv_radius
-
-
-class _SVEHintsReduced(AbstractSVEHints):
-    def __init__(self, inner_hints):
-        self.inner_hints = inner_hints
-
-    @property
-    def ngauss(self): return self.inner_hints.ngauss
-
-    @property
-    def segments_x(self): return _symm_segments(self.inner_hints.segments_x)
 
-    @property
-    def segments_y(self): return _symm_segments(self.inner_hints.segments_y)
+    def __init__(self, lambda_):
+        """Initialize regularized bosonic kernel with cutoff lambda."""
+        self._lambda = float(lambda_)
+        self._ptr = reg_bose_kernel_new(self._lambda)
 
     @property
-    def nsvals(self): return (self.inner_hints.nsvals + 1) // 2
-
-
-class _LogisticKernelOdd(ReducedKernel):
-    """Fermionic analytical continuation kernel, odd.
-
-    In dimensionless variables ``x = 2*τ/β - 1``, ``y = β*ω/Λ``, the fermionic
-    integral kernel is a function on ``[-1, 1] x [-1, 1]``::
-
-        K(x, y) == -sinh(Λ/2 * x * y) / cosh(Λ/2 * y)
-    """
-    def __call__(self, x, y, x_plus=None, x_minus=None):
-        result = super().__call__(x, y, x_plus, x_minus)
-
-        # For x * y around 0, antisymmetrization introduces cancellation, which
-        # reduces the relative precision.  To combat this, we replace the
-        # values with the explicit form
-        v_half = self.inner.lambda_/2 * y
-        xy_small = x * v_half < 1
-        cosh_finite = v_half < 85
-        np.divide(-np.sinh(v_half * x, where=xy_small),
-                  np.cosh(v_half, where=cosh_finite),
-                  out=result, where=np.logical_and(xy_small, cosh_finite))
-        return result
-
-
-class _RegularizedBoseKernelOdd(ReducedKernel):
-    """Bosonic analytical continuation kernel, odd.
-
-    In dimensionless variables ``x = 2*τ/β - 1``, ``y = β*ω/Λ``, the fermionic
-    integral kernel is a function on ``[-1, 1] x [-1, 1]``::
-
-            K(x, y) = -y * sinh(Λ/2 * x * y) / sinh(Λ/2 * y)
-    """
-    def __call__(self, x, y, x_plus=None, x_minus=None):
-        result = super().__call__(x, y, x_plus, x_minus)
-
-        # For x * y around 0, antisymmetrization introduces cancellation, which
-        # reduces the relative precision.  To combat this, we replace the
-        # values with the explicit form
-        v_half = self.inner.lambda_/2 * y
-        xv_half = x * v_half
-        xy_small = xv_half < 1
-        sinh_range = np.logical_and(v_half > 1e-200, v_half < 85)
-        np.divide(
-            np.multiply(-y, np.sinh(xv_half, where=xy_small), where=xy_small),
-            np.sinh(v_half, where=sinh_range),
-            out=result, where=np.logical_and(xy_small, sinh_range))
-        return result
-
-
-def matrix_from_gauss(kernel, gauss_x, gauss_y):
-    """Compute matrix for kernel from Gauss rule"""
-    # (1 +- x) is problematic around x = -1 and x = 1, where the quadrature
-    # nodes are clustered most tightly.  Thus we have the need for the
-    # matrix method.
-    return kernel(gauss_x.x[:,None], gauss_y.x[None,:],
-                  gauss_x.x_forward[:,None], gauss_x.x_backward[:,None])
-
-
-def _check_domain(kernel, x, y):
-    """Check that arguments lie within the correct domain"""
-    x = np.asarray(x)
-    xmin, xmax = kernel.xrange
-    if not (x >= xmin).all() or not (x <= xmax).all():
-        raise ValueError("x values not in range [{:g},{:g}]".format(xmin, xmax))
-
-    y = np.asarray(y)
-    ymin, ymax = kernel.yrange
-    if not (y >= ymin).all() or not (y <= ymax).all():
-        raise ValueError("y values not in range [{:g},{:g}]".format(ymin, ymax))
-    return x, y
-
-
-def _symm_segments(x):
-    x = np.asarray(x)
-    if not np.allclose(x, -x[::-1]):
-        raise ValueError("segments must be symmetric")
-    xpos = x[x.size // 2:]
-    if xpos[0] != 0:
-        xpos = np.hstack([0, xpos])
-    return xpos
-
+    def lambda_(self):
+        """Kernel cutoff."""
+        return self._lambda
 
-def _compute_uv(lambda_, x, y, x_plus=None, x_minus=None):
-    if x_plus is None:
-        x_plus = 1 + x
-    if x_minus is None:
-        x_minus = 1 - x
-    u_plus = .5 * x_plus
-    u_minus = .5 * x_minus
-    v = lambda_ * y
-    return u_plus, u_minus, v
+    def __del__(self):
+        """Clean up kernel resources."""
+        if hasattr(self, '_ptr') and self._ptr:
+            _lib.spir_kernel_release(self._ptr)