expectation_compute_with_gradients_backward#

cuquantum.bindings.cutensornet.expectation_compute_with_gradients_backward(
intptr_t handle,
intptr_t tensor_network_expectation,
int32_t accumulate_gradients,
intptr_t expectation_value_adjoint,
intptr_t state_norm_adjoint,
intptr_t work_desc,
intptr_t expectation_value,
intptr_t state_norm,
intptr_t cuda_stream,
)[source]#

Computes the tensor network state expectation value and its gradients together.

Parameters:

  • handle (intptr_t) – cuTensorNet library handle.

  • tensor_network_expectation (intptr_t) – Tensor network state expectation value representation.

  • accumulate_gradients (int32_t) – If non-zero, add to existing gradient values; otherwise overwrite.

  • expectation_value_adjoint (intptr_t) – Upstream gradient scalar for chain rule computation (host-accessible pointer, same type as state). Set to 1 for direct gradient .

  • state_norm_adjoint (intptr_t) – Upstream gradient for state norm in chain rule (host-accessible pointer, same type as state).

  • work_desc (intptr_t) – The workspace descriptor with SCRATCH workspace set.

  • expectation_value (intptr_t) – Computed expectation value (host-accessible pointer, same type as state).

  • state_norm (intptr_t) – Host-accessible pointer to store the squared 2-norm of the state.

  • cuda_stream (intptr_t) – The CUDA stream on which the computation is performed.

See also

cutensornetExpectationComputeWithGradientsBackward()