Skip to content

implements "none" option #165

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
ryanmrichard merged 2 commits into from
Apr 7, 2026
Merged

implements "none" option #165

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
7286fe9
implements "none" option
ryanmrichard Apr 6, 2026
cadd072
try isolating check for uq
ryanmrichard Apr 7, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
53 changes: 39 additions & 14 deletions src/integrals/ao_integrals/uq_atom_symm_blocked_driver.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,33 +29,58 @@ namespace {
template<typename FloatType, typename T, typename Tensor>
auto average_error(T&& strides, T&& nbf, T&& ao_i, Tensor&& error,
utils::mean_type mean) {
std::vector<FloatType> buffer;
#ifdef ENABLE_SIGMA
using uq_type = sigma::Uncertain<FloatType>;
auto n_elements = nbf[0] * nbf[1] * nbf[2] * nbf[3];

if(mean == utils::mean_type::none) {
std::vector<uq_type> result;
result.reserve(n_elements);
for(std::size_t i = 0; i < nbf[0]; ++i) {
auto ioffset = (ao_i[0] + i) * strides[0];
for(std::size_t j = 0; j < nbf[1]; ++j) {
auto joffset = ioffset + (ao_i[1] + j) * strides[1];
for(std::size_t k = 0; k < nbf[2]; ++k) {
auto koffset = joffset + (ao_i[2] + k) * strides[2];
for(std::size_t l = 0; l < nbf[3]; ++l) {
auto loffset = koffset + (ao_i[3] + l) * strides[3];
result.push_back(uq_type{0.0, error[loffset]});
}
}
}
}
return result;
}

std::vector<FloatType> buffer;
buffer.reserve(n_elements);
for(std::size_t i = 0; i < nbf[0]; ++i) {
auto ioffset = (ao_i[0] + i) * strides[0];

for(std::size_t j = 0; j < nbf[1]; ++j) {
auto joffset = ioffset + (ao_i[1] + j) * strides[1];

for(std::size_t k = 0; k < nbf[2]; ++k) {
auto koffset = joffset + (ao_i[2] + k) * strides[2];

for(std::size_t l = 0; l < nbf[3]; ++l) {
auto loffset = koffset + (ao_i[3] + l) * strides[3];
buffer.push_back(error[loffset]);
}
}
}
}

return utils::compute_mean(mean, buffer);
auto mean_value = utils::compute_mean(mean, buffer);
return std::vector<uq_type>(n_elements, uq_type{0.0, mean_value});
#else
throw std::runtime_error("Sigma support not enabled!");
return std::vector<int>{};
#endif
}

#ifdef ENABLE_SIGMA
template<typename FloatType, typename T, typename Tensor>
auto compute_block(T&& strides, T&& nbf, T&& ao_i, Tensor&& value,
FloatType error) {
const std::vector<sigma::Uncertain<FloatType>>& errors) {
auto n_elements = nbf[0] * nbf[1] * nbf[2] * nbf[3];
std::vector<std::decay_t<FloatType>> buffer(n_elements, error);
std::vector<sigma::Uncertain<FloatType>> buffer(n_elements);

for(std::size_t i = 0; i < nbf[0]; ++i) {
auto ilocal = i * nbf[1] * nbf[2] * nbf[3];
Expand All @@ -70,15 +95,16 @@ auto compute_block(T&& strides, T&& nbf, T&& ao_i, Tensor&& value,
auto koffset = joffset + (ao_i[2] + k) * strides[2];

for(std::size_t l = 0; l < nbf[3]; ++l) {
auto llocal = klocal + l;
auto loffset = koffset + (ao_i[3] + l) * strides[3];
buffer[llocal] += value[loffset];
auto llocal = klocal + l;
auto loffset = koffset + (ao_i[3] + l) * strides[3];
buffer[llocal] = errors[llocal] + value[loffset];
}
}
}
}
return buffer;
}
#endif

template<typename FloatType, typename T>
void set_block(T&& strides, T&& nbf,
Expand Down Expand Up @@ -191,13 +217,12 @@ struct Kernel {
bool pair_gt = (c2eqc0 && centers[3] > centers[1]);
if(pair_gt && all_same) break;

auto block_error = average_error<float_type>(
auto block_errors = average_error<float_type>(
strides, nbf, ao_offsets, error, m_mean);
uq_type max_uq{0.0, block_error};

// Compute (ab|cd)
auto block = compute_block(strides, nbf, ao_offsets,
t, max_uq);
t, block_errors);

// Set all symmetry equivalent blocks to `block`
auto perms = get_permutations_with_sigma(
Expand Down
36 changes: 34 additions & 2 deletions tests/cxx/unit/integrals/ao_integrals/test_uq_atom_symm_blocked_driver.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -66,11 +66,36 @@ auto corr_answer(const simde::type::tensor& T) {
}
}

template<typename FloatType>
auto corr_answer_no_mean(const simde::type::tensor& T,
const simde::type::tensor& T_eri,
const simde::type::tensor& T_err) {
if constexpr(std::is_same_v<FloatType, double>) {
return true;
} else {
auto t_uq = eigen_tensor<4, FloatType>(T.buffer());
auto t_eri = eigen_tensor<4, double>(T_eri.buffer());
auto t_err = eigen_tensor<4, double>(T_err.buffer());

for(Eigen::Index i = 0; i < 2; ++i)
for(Eigen::Index j = 0; j < 2; ++j)
for(Eigen::Index k = 0; k < 2; ++k)
for(Eigen::Index l = 0; l < 2; ++l) {
REQUIRE(t_uq(i, j, k, l).mean() ==
Catch::Approx(t_eri(i, j, k, l)).margin(1E-6));
REQUIRE(t_uq(i, j, k, l).sd() ==
Catch::Approx(t_err(i, j, k, l)).margin(1E-6));
}
return true;
}
}

} // namespace

TEST_CASE("UQ Atom Symm Blocked Driver") {
using float_type = tensorwrapper::types::udouble;
using test_pt = simde::ERI4;
using error_pt = integrals::property_types::Uncertainty<test_pt>;
using tensorwrapper::operations::approximately_equal;

if constexpr(tensorwrapper::types::is_uncertain_v<float_type>) {
Expand All @@ -80,7 +105,7 @@ TEST_CASE("UQ Atom Symm Blocked Driver") {
integrals::set_defaults(mm);
REQUIRE(mm.count("UQ Atom Symm Blocked Driver"));
mm.change_input("ERI4", "Threshold", 1.0e-6);
auto mod = mm.at("UQ Atom Blocked Driver");
auto mod = mm.at("UQ Atom Symm Blocked Driver");

// Make Operator
simde::type::v_ee_type op{};
Expand All @@ -96,7 +121,14 @@ TEST_CASE("UQ Atom Symm Blocked Driver") {
// Make BraKet Input
chemist::braket::BraKet braket(aos_squared, op, aos_squared);

SECTION("No Mean") { REQUIRE_THROWS(mod.run_as<test_pt>(braket)); }
SECTION("No Mean") {
auto T = mod.run_as<test_pt>(braket);
auto tol = 1.0e-6;
auto T_eri = mm.at("ERI4").run_as<test_pt>(braket);
auto T_err =
mm.at("Primitive Error Model").run_as<error_pt>(braket, tol);
REQUIRE(corr_answer_no_mean<float_type>(T, T_eri, T_err));
}
SECTION("Max Error") {
mod.change_input("Mean Type", "max");
auto T = mod.run_as<test_pt>(braket);
Expand Down
Loading