Merge pull request #281 from evaleev/kshitij/feature/dfbs_generator

Automatic Density Fitting basis generator

.github/workflows/cmake.yml

@@ -105,7 +105,7 @@ jobs:
       shell: bash
       working-directory: ${{github.workspace}}/build/compiler
       run: |
-        CPPFLAGS="-I$EIGEN3_INCLUDE_DIR" CXXFLAGS="-std=c++11 -Wno-enum-compare" ${{github.workspace}}/configure --with-max-am=2,1 --with-eri-max-am=2,2 --with-eri3-max-am=3,2 --enable-eri=1 --enable-eri3=1 --enable-1body=1 --disable-1body-property-derivs --with-multipole-max-order=2
+        CPPFLAGS="-I$EIGEN3_INCLUDE_DIR" CXXFLAGS="-std=c++11 -Wno-enum-compare" ${{github.workspace}}/configure --with-max-am=2,1 --with-eri-max-am=2,2 --with-eri3-max-am=3,2 --with-eri2-max-am=3,2 --enable-eri=1 --enable-eri3=1 --enable-eri2=0 --enable-1body=1 --disable-1body-property-derivs --with-multipole-max-order=2
         make -j3
         make check
         cd src/bin/test_eri && ./stdtests.pl && cd ../../..

configure.ac

@@ -341,6 +341,7 @@ esac
 )
 
 LIBINT_HARD_MAX_AM=12
+AC_DEFINE_UNQUOTED(LIBINT_HARD_MAX_AM,$LIBINT_HARD_MAX_AM)
 LIBINT_MAX_AM=4
 AC_ARG_WITH(max-am,
 AS_HELP_STRING([--with-max-am=N],[Support Gaussians of angular momentum up to N. Can specify values for each derivative levels as a list N0,N1,N2...]),
@@ -528,7 +529,7 @@ AS_HELP_STRING([--with-eri3-max-am=N],[Support 3-center ERIs for Gaussians of an
      else
       ERI3_MAX_AM=$withval
      fi
-     if test $ERI3_MAX_AM -ge $LIBINT_HARD_MAX_AM; then
+     if test $ERI3_MAX_AM -gt $LIBINT_HARD_MAX_AM; then
       AC_MSG_ERROR([Value for --with-eri3-max-am too high ($withval). Are you sure you know what you are doing?])
      fi
      AC_SUBST(ERI3_MAX_AM)
@@ -599,7 +600,7 @@ AS_HELP_STRING([--with-eri2-max-am=N],[Support 2-center ERIs for Gaussians of an
      else
       ERI2_MAX_AM=$withval
      fi
-     if test $ERI2_MAX_AM -ge $LIBINT_HARD_MAX_AM; then
+     if test $ERI2_MAX_AM -gt $LIBINT_HARD_MAX_AM; then
       AC_MSG_ERROR([Value for --with-eri2-max-am too high ($withval). Are you sure you know what you are doing?])
      fi
      AC_SUBST(ERI2_MAX_AM)

export/cmake/CMakeLists.txt.export

@@ -446,6 +446,7 @@ if (LIBINT_HAS_CXX_API)
             tests/unit/test-c-api.cc
             tests/unit/test-core.cc
             tests/unit/test-core-ints.cc
+            tests/unit/test-df-autogen.cc
             tests/unit/test-permute.cc
             tests/unit/test-precision.cc
             tests/unit/test-shell-order.cc

include/libint2/config.h.in

@@ -47,6 +47,9 @@
 /* Prefix for all names in API */
 #undef LIBINT_API_PREFIX
 
+/* Max AM supported by Libint *in principle* */
+#undef LIBINT_HARD_MAX_AM
+
 /* Max AM (same for all derivatives; if not defined see LIBINT_MAX_AM_LIST) */
 #undef LIBINT_MAX_AM
 

include/libint2/dfbs_generator.h

@@ -0,0 +1,321 @@
+/*
+ *  Copyright (C) 2004-2024 Edward F. Valeev
+ *
+ *  This file is part of Libint.
+ *
+ *  Libint is free software: you can redistribute it and/or modify
+ *  it under the terms of the GNU Lesser General Public License as published by
+ *  the Free Software Foundation, either version 3 of the License, or
+ *  (at your option) any later version.
+ *
+ *  Libint is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *  GNU Lesser General Public License for more details.
+ *
+ *  You should have received a copy of the GNU Lesser General Public License
+ *  along with Libint.  If not, see <http://www.gnu.org/licenses/>.
+ *
+ */
+
+#ifndef _libint2_src_lib_libint_dfbs_generator_h_
+#define _libint2_src_lib_libint_dfbs_generator_h_
+
+#include <libint2.h>
+#include <libint2/atom.h>
+#include <libint2/basis.h>
+#include <libint2/boys.h>
+#include <libint2/pivoted_cholesky.h>
+
+#include <Eigen/Dense>
+#include <Eigen/Eigenvalues>
+#include <algorithm>
+#include <cmath>
+
+namespace libint2 {
+
+namespace detail {
+
+/// @brief Uncontracts a set of shells
+/// @param[in] shells a vector of shells to be uncontracted
+/// @return a vector of uncontracted shells
+inline std::vector<Shell> uncontract(const std::vector<Shell> &shells) {
+  std::vector<Shell> primitive_shells;
+  for (const auto &contracted_shell : shells) {
+    for (size_t p = 0; p < contracted_shell.nprim(); p++) {
+      const auto prim_shell = contracted_shell.extract_primitive(p, true);
+      // if dealing with generally contracted basis (e.g., cc-pvxz) represented
+      // as a segmented basis need to remove duplicates
+      if (std::find(primitive_shells.begin(), primitive_shells.end(),
+                    prim_shell) == primitive_shells.end())
+        primitive_shells.emplace_back(std::move(prim_shell));
+    }
+  }
+  return primitive_shells;
+}
+
+/// @brief returns \f$ \Gamma(x) \f$
+inline double gamma_function(const double x) { return std::tgamma(x); }
+
+/// @brief Computes an effective exponent for a product of two primitive
+/// gaussians as as described in J. Chem. Theory Comput. 2021, 17, 6886−6900.
+/// \f$ \alpha_{eff} = \left[ \frac{\Gamma(L+2)\Gamma(l_\mu +l_\nu +
+/// \frac{3}{2})}{\Gamma(L+ \frac{3}{2})\Gamma(l_\mu +l_\nu + 2)} \right]^2
+/// (\alpha_\mu + \alpha_\nu) \f$
+/// @param shell1 first primitive shell
+/// @param shell2 second primitive shell
+/// @param L total angular momentum of product function
+/// @return effective exponent of product function
+inline double alpha_eff(const Shell &shell1, const Shell &shell2, const int L) {
+  const auto alpha1 = shell1.alpha[0];
+  const auto alpha2 = shell2.alpha[0];
+  const auto l1 = shell1.contr[0].l;
+  const auto l2 = shell2.contr[0].l;
+  const auto prefactor =
+      std::pow((gamma_function(L + 2.) * gamma_function(l1 + l2 + 1.5)) /
+                   (gamma_function(l1 + l2 + 2.) * gamma_function(L + 1.5)),
+               2.);
+  return prefactor * (alpha1 + alpha2);
+}
+
+/// @brief Creates a set of product functions from a set of primitive shells.
+/// Each pair of primitive shells produces a set of product functions with an
+/// effective exponent (\f$ \alpha_{eff} \f$ as described above) and angular
+/// momentum ranging from \f$ |l1-l2| \leq L \leq l1+l2 \f$ as described in J.
+/// Chem. Theory Comput. 2021, 17, 6886−6900.
+/// @param primitive_shells set of primitive shells
+/// @return set of product functions
+inline std::vector<Shell> product_functions(
+    const std::vector<Shell> &primitive_shells) {
+  std::vector<Shell> product_functions;
+  for (size_t i = 0; i < primitive_shells.size(); ++i) {
+    for (size_t j = 0; j <= i; ++j) {
+      const auto li = primitive_shells[i].contr[0].l;
+      const auto lj = primitive_shells[j].contr[0].l;
+      for (auto L = std::abs(li - lj); L <= li + lj; L++) {
+        const auto alpha = libint2::svector<double>(
+            {alpha_eff(primitive_shells[i], primitive_shells[j], L)});
+        libint2::svector<Shell::Contraction> contr_;
+        Shell::Contraction contr1;
+        contr1.l = L;
+        contr1.pure = true;  // libint2 needs solid harmonics for 2c2b integrals
+        contr1.coeff = {1.0};
+        contr_.push_back(contr1);
+        assert(primitive_shells[i].O == primitive_shells[j].O);
+        const auto shell = Shell(alpha, contr_, primitive_shells[i].O);
+        if (std::find(product_functions.begin(), product_functions.end(),
+                      shell) == product_functions.end())
+          product_functions.emplace_back(shell);
+      }
+    }
+  }
+  return product_functions;
+}
+
+/// @brief returns a hash map of shell indices to basis function indices
+inline std::vector<size_t> map_shell_to_basis_function(
+    const std::vector<libint2::Shell> &shells) {
+  std::vector<size_t> result;
+  result.reserve(shells.size());
+
+  size_t n = 0;
+  for (auto &&shell : shells) {
+    result.push_back(n);
+    n += shell.size();
+  }
+
+  return result;
+}
+
+/// @brief Computes the Coulomb matrix \f$ (\mu|rij^{-1}|\nu) \f$  for a set of
+/// shells
+/// @param shells set of shells
+/// @return Coulomb matrix
+inline Eigen::MatrixXd compute_coulomb_matrix(
+    const std::vector<Shell> &shells) {
+  const auto n = nbf(shells);
+  Eigen::MatrixXd result = Eigen::MatrixXd::Zero(n, n);
+  using libint2::Engine;
+  const auto oper = libint2::Operator::coulomb;
+  const auto braket = libint2::BraKet::xs_xs;
+  Engine engine(oper, max_nprim(shells), max_l(shells), 0,
+                std::numeric_limits<scalar_type>::epsilon(),
+                libint2::default_params(oper), braket);
+  engine.set(ScreeningMethod::Conservative);
+  const auto shell2bf = map_shell_to_basis_function(shells);
+  const auto &buf = engine.results();
+  for (size_t s1 = 0; s1 != shells.size(); ++s1) {
+    auto bf1 = shell2bf[s1];
+    auto n1 = shells[s1].size();
+    for (size_t s2 = 0; s2 <= s1; ++s2) {
+      auto bf2 = shell2bf[s2];
+      auto n2 = shells[s2].size();
+      engine.compute(shells[s1], shells[s2]);
+      Eigen::Map<const Eigen::MatrixXd> buf_mat(buf[0], n1, n2);
+      result.block(bf1, bf2, n1, n2) = buf_mat;
+      if (s1 != s2) result.block(bf2, bf1, n2, n1) = buf_mat.transpose();
+    }
+  }
+  return result;
+}
+
+/// @brief Splits a set of shells by angular momentum
+/// @param shells set of shells
+/// @return vector of vectors of shells split by angular momentum L. The i-th
+///  vector contains all shells with angular momentum i
+inline std::vector<std::vector<Shell>> split_by_L(
+    const std::vector<Shell> &shells) {
+  int lmax = max_l(shells);
+  std::vector<std::vector<Shell>> sorted_shells;
+  sorted_shells.resize(lmax + 1);
+  for (auto &&shell : shells) {
+    auto l = shell.contr[0].l;
+    sorted_shells[l].push_back(shell);
+  }
+  return sorted_shells;
+}
+
+/// @brief Computes the reduced set of product functions via pivoted Cholesky
+/// decomposition as described in J. Chem. Theory Comput. 2021, 17, 6886−6900.
+/// Cholesky decomposition is performed on the Coulomb matrix of the product
+/// functions and the pivot indices are sorted in ascending order of column sums
+/// of the Coulomb matrix. The reduced set of product functions is then
+/// constructed by selecting the product functions corresponding to the pivot
+/// indices.
+/// @param shells set of shells
+/// @param cholesky_threshold threshold for choosing a product function via
+/// pivoted Cholesky decomposition
+/// @return reduced set of product functions
+inline std::vector<Shell> pivoted_cholesky_in_L(
+    const std::vector<Shell> &shells, const double cholesky_threshold) {
+  const auto n = shells.size();  // number of shells
+  std::vector<size_t>
+      shell_indices;  // hash map of basis function indices to shell indices
+  const auto L = shells[0].contr[0].l;  // all shells must have same L
+  const auto nbf = libint2::nbf(
+      shells);  // total number of basis functions in vector of shells
+  for (size_t i = 0; i < n; ++i) {
+    for (size_t j = 0; j < 2 * L + 1;
+         ++j)  // 2L+1 since libint2 strictly uses solid harmonics for 2c2b
+      // integrals
+      shell_indices.push_back(i);
+  }
+  assert(shell_indices.size() == nbf);
+  const auto C = compute_coulomb_matrix(shells);
+  std::vector<size_t> pivot(nbf);
+  for (auto i = 0; i < nbf; ++i) {
+    pivot[i] = i;
+  }
+  // set pivot indices in ascending order of off diagonal elements of Coulomb
+  // matrix see Phys. Rev. A 101, 032504 (Accurate reproduction of strongly
+  // repulsive interatomic potentials)
+  Eigen::MatrixXd C_off_diag = C;
+  auto col_sum = C_off_diag.colwise().sum();
+  // sort pivot indices in ascending order of column sums
+  std::sort(pivot.begin(), pivot.end(), [&col_sum](size_t i1, size_t i2) {
+    return col_sum[i1] < col_sum[i2];
+  });
+  // compute Cholesky decomposition
+  const auto reduced_pivots = pivoted_cholesky(C, cholesky_threshold, pivot);
+
+  std::vector<Shell> reduced_shells;
+  for (size_t i = 0; i < reduced_pivots.size(); ++i) {
+    // check if the reduced shell is already in reduced shells
+    if (std::find(reduced_shells.begin(), reduced_shells.end(),
+                  shells[shell_indices[reduced_pivots[i]]]) ==
+        reduced_shells.end())
+      reduced_shells.push_back(shells[shell_indices[reduced_pivots[i]]]);
+  }
+  return reduced_shells;
+}
+}  // namespace detail
+
+/// @brief This class produces density fitting basis sets for an atom from
+/// products of AO basis functions and eliminates linearly dependent functions
+/// via pivoted Cholesky decomposition see: J. Chem. Theory Comput. 2021, 17,
+/// 6886−6900 (Straightforward and Accurate Automatic Auxiliary Basis Set
+/// Generation for Molecular Calculations with Atomic Orbital Basis Sets)
+class DFBasisSetGenerator {
+ public:
+  /// @brief constructor for DFBasisSetGenerator class, generates density
+  /// fitting basis set from products of AO basis functions of and atom see: J.
+  /// Chem. Theory Comput. 2021, 17, 6886−6900 (Straightforward and Accurate
+  /// Automatic Auxiliary Basis Set Generation for Molecular Calculations with
+  /// Atomic Orbital Basis Sets)
+  /// @param obs_name name of basis set for AO functions
+  /// @param atoms vector of atoms
+  /// @param cholesky_threshold threshold for threshold for  pivoted Cholesky
+  /// decomposition to be performed on the Coulomb matrix of the product
+  /// functions
+  DFBasisSetGenerator(std::string obs_name, const Atom &atom,
+                      const double cholesky_threshold = 1e-7) {
+    // get AO basis shells for each atom
+    const auto atom_bs = BasisSet(obs_name, {atom});
+    const auto obs_shells = atom_bs.shells();
+    // get primitive shells from AO functions
+    const auto primitive_shells = detail::uncontract(obs_shells);
+    // compute candidate shells
+    candidate_shells_ = detail::product_functions(primitive_shells);
+    cholesky_threshold_ = cholesky_threshold;
+  }
+
+  /// @brief constructor for DFBasisSetGenerator class, generates density
+  /// fitting basis set from products of AO shells provided by user
+  /// @param shells vector of vector of shells for each atom
+  /// @param cholesky_threshold threshold for  pivoted Cholesky decomposition to
+  /// be performed on the Coulomb matrix of the product functions
+  DFBasisSetGenerator(const std::vector<Shell> &shells,
+                      const double cholesky_threshold = 1e-7) {
+    const auto primitive_shells = detail::uncontract(shells);
+    candidate_shells_ = detail::product_functions(primitive_shells);
+    cholesky_threshold_ = cholesky_threshold;
+  }
+
+  DFBasisSetGenerator() = default;
+
+  ~DFBasisSetGenerator() = default;
+
+  /// @brief returns the candidate shells (full set of product functions)
+  std::vector<Shell> candidate_shells() { return candidate_shells_; }
+
+  /// @brief returns a set of shells reduced via pivoted Cholesky
+  /// decomposition of the Coulomb matrix of the product functions for each
+  /// angular momentum L as described in J. Chem. Theory Comput. 2021, 17,
+  /// 6886−6900.
+  std::vector<Shell> reduced_shells() {
+    if (reduced_shells_computed_)
+      return reduced_shells_;
+    else {
+      const auto candidate_splitted_in_L =
+          detail::split_by_L(candidate_shells_);
+      for (size_t i = 0; i < candidate_splitted_in_L.size(); ++i) {
+        std::vector<Shell> reduced_shells_L;
+        if (candidate_splitted_in_L[i].size() > 1)
+          reduced_shells_L = detail::pivoted_cholesky_in_L(
+              candidate_splitted_in_L[i], cholesky_threshold_);
+        else
+          reduced_shells_L = candidate_splitted_in_L[i];
+        reduced_shells_.insert(reduced_shells_.end(), reduced_shells_L.begin(),
+                               reduced_shells_L.end());
+      }
+      reduced_shells_computed_ = true;
+    }
+    return reduced_shells_;
+  }
+
+  /// @brief returns a BasisSet of shells reduced via pivoted Cholesky
+  /// decomposition of the Coulomb matrix of the product functions for each
+  /// angular momentum L as described in J. Chem. Theory Comput. 2021, 17,
+  /// 6886−6900.
+  const BasisSet reduced_basis() { return BasisSet(reduced_shells()); }
+
+ private:
+  double cholesky_threshold_;
+  std::vector<Shell> candidate_shells_;  // full set of product functions
+  std::vector<Shell> reduced_shells_;    // reduced set of product functions
+  bool reduced_shells_computed_ = false;
+};
+
+}  // namespace libint2
+
+#endif /* _libint2_src_lib_libint_dfbs_generator_h_ */