Skip to content

Gate BFGS gradient-convergence denominator fix on RDKit >= 2026.03 #183

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
scal444 merged 1 commit into from
May 28, 2026
Merged

Gate BFGS gradient-convergence denominator fix on RDKit >= 2026.03 #183

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
12 changes: 7 additions & 5 deletions src/minimizer/bfgs_minimize.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -908,11 +908,13 @@ __global__ void updateDGradKernel(const double gradTol,
double blockMax = cub::BlockReduce<double, 128>(tempStorage).Reduce(localMax, cubMax());

if (idxWithinSystem == 0) {
// Matches RDKit's signed-energy convergence denominator in ForceField::minimize.
// Negative-energy geometries can clamp this to 1, artificially tightening the
// check — but fixing it unconditionally diverges from reference behavior.
// TODO: file upstream RDKit bug; gate fix on RDKit version once merged there.
const double term = max(energies[sysIdx] * gradScales[sysIdx], 1.0);
// rdkit/rdkit#9298 (merged RDKit 2026.03) fixed the signed-energy denominator bug:
// raw negative energy clamped the denominator to 1, artificially tightening gradTol.
// Use |energy| when linked against a fixed RDKit; keep signed otherwise for parity.
constexpr bool kRdkitHasGradDenomFix =
RDKIT_VERSION_MAJOR > 2026 || (RDKIT_VERSION_MAJOR == 2026 && RDKIT_VERSION_MINOR >= 3);
const double energyMag = kRdkitHasGradDenomFix ? fabs(energies[sysIdx]) : energies[sysIdx];
const double term = max(energyMag * gradScales[sysIdx], 1.0);
blockMax /= term;
if (blockMax < gradTol) {
// Converged
Expand Down
7 changes: 6 additions & 1 deletion src/minimizer/bfgs_minimize_permol_kernels.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -292,7 +292,12 @@ __device__ void updateDGrad(const int
float blockMax = cub::BlockReduce<double, BLOCK_SIZE>(tempStorage).Reduce(localMax, cubMax());

if (threadIdx.x == 0) {
const float term = max(energy * gradScale, 1.0);
// rdkit/rdkit#9298 (merged RDKit 2026.03): use |energy| to avoid clamping the
// denominator to 1 when energy is negative; match signed behavior on older RDKit.
constexpr bool kRdkitHasGradDenomFix =
RDKIT_VERSION_MAJOR > 2026 || (RDKIT_VERSION_MAJOR == 2026 && RDKIT_VERSION_MINOR >= 3);
const double energyMag = kRdkitHasGradDenomFix ? fabs(energy) : energy;
const float term = max(energyMag * gradScale, 1.0);
blockMax /= term;
if (blockMax < gradTol) {
converged = true;
Expand Down
Loading