研究人员对IBM的铁硫簇量子化学基准测试结果进行了自旋态审计，发现重大偏差。[1]该审计发现，SQD计算所收敛的电子自旋态（⟨S²⟩=4.7-7.0）与目标基态（⟨S²⟩=0的单态）明显不符。[1]审计人员利用IBM提供的原始硬件测量数据重现了计算过程，确认[4Fe-4S]系统收敛到自旋三态而非预期的基态。[1]
在能量精度方面，计算结果同样存在显著偏离。[1][2Fe-2S]系统的偏离值为248毫哈特里，而[4Fe-4S]系统的偏离值达到1,438毫哈特里。[1]IBM采用的自旋缓解方法虽然应用了额外处理，但改善有限——该方法将子空间扩大了4倍（194,481行列式对比原有的48,600）。[1]审计人员已向IBM提交相关问题，并发表了包含完整分析轨迹和复现方法的预印本。[1]在审计过程中，Claude AI在72小时内完成了初步审计工作，并在经历多轮对抗性审查后自行撤回了错误结论。[1]
A researcher has conducted a spin-state audit of IBM's SQD (Symmetry-Preserving Quantum Downfolding) quantum chemistry calculations on iron–sulfur clusters and identified significant deviations from expected results [1]. The computed electronic spin states, measured as ⟨S²⟩ values ranging from 4.7 to 7.0, diverge substantially from the target ground state singlet configuration with ⟨S²⟩ = 0 [1].
Using raw hardware measurement data provided by IBM, the researcher reproduced the computational process and discovered that the [4Fe-4S] system converged to a spin triplet state (S = 1 eigenstate) rather than the anticipated ground state [1]. For the [2Fe-2S] system, deviations of 248 milliHartree from reference values were observed, while the [4Fe-4S] system exhibited a substantially larger discrepancy of 1,438 milliHartree [1]. IBM's spin-correction methodology shifted the ground-state energy by less than one nanoHartree while simultaneously expanding the computational subspace fourfold, from 48,600 to 194,481 determinants [1].
The researcher submitted detailed findings to IBM and published a preprint containing complete analytical traces and reproduction methods [1]. A GitHub issue documenting the technical problems has been filed at https://github.com/Qiskit/qiskit-addon-sqd/issues/337 [1].