Research-Stack/6-Documentation/docs/recovered/deep-research-report.md

# Nitrogenase N2-Binding Source Audit and Regression Dataset

## Source audit

The core numeric point checks out. The four Bjornsson values you flagged from the later Ryde paper — **+69, +41, +8, and −17 kJ/mol** for the best \(N_2\)-binding energies of \(E_0\), \(E_1\), \(E_2\), and \(E_4\) — do appear in Jiang and Ryde’s introduction as their summary of Pang and Bjornsson’s newer study, and those numbers match the more explicit Pang/Bjornsson values once the reported kcal/mol numbers are converted to kJ/mol. One source entry in your list, however, needs correction: the 2005 Kästner/Hemmen/Blöchl paper is **10.1063/1.2008227**, while **10.1063/1.2042456** is an unrelated paper on quantum control. citeturn54view0turn32search2turn31search1

## Pang and Bjornsson values

For regression purposes, Pang and Bjornsson’s terminology matters. In that paper, **“binding energy”** means binding relative to the **lowest-energy isomer of a given \(E_n\) manifold**, while **“single-step \(N_2\) binding energy”** means binding to a **specific precursor isomer**. They also state explicitly that these are **0 K electronic energies** without ZPVE or entropy, so they are not free energies of binding. citeturn18search0turn27search0

| Redox state | Regression-safe Pang/Bjornsson value | Pang/Bjornsson structural reference |
|---|---:|---|
| \(E_0\) | **+69.45 kJ/mol** | best QM-VI \(N_2\)-bound minimum at Fe6, \(E_0\)-\(N_2\)@Fe6-BS147 |
| \(E_1\) | **+41.42 kJ/mol** | best QM-VI \(N_2\)-bound minimum at Fe6 |
| \(E_2\) | **+7.95 kJ/mol** | **best-isomer** \(E_2\)-hyd value, referenced to the most stable \(E_2\)-hyd precursor |
| \(E_4\) | **−17.15 kJ/mol** | **best-isomer** \(E_4\)-SP value, referenced to the most stable \(E_4\) precursor |

These values come directly from Pang and Bjornsson’s large-QM discussion: \(E_0\) binding to Fe6 is **16.6 kcal/mol**, \(E_1\) binding to Fe6 is **9.9 kcal/mol**, \(E_2\)-hyd best-isomer binding is **+1.9 kcal/mol**, and \(E_4\)-SP best-isomer binding is **−4.1 kcal/mol**. Converting with \(1\ \text{kcal mol}^{-1}=4.184\ \text{kJ mol}^{-1}\) gives the four regression values above. citeturn24search0turn23search0turn27search0

Two caveats are especially important before fitting anything. First, the alternative **single-step** values are much more favorable than the **best-isomer** values for \(E_2\) and \(E_4\): Pang and Bjornsson report **−9.8 kcal/mol** for direct binding to the alternative \(E_2\)-hyd-SH\(^{-}\)@Fe2 precursor and **−15.2 kcal/mol** for \(E_4\)-SP-SH\(^{-}\)@Fe2 \(\rightarrow E_4\)-SP-\(N_2\)@Fe6. Second, the paper stresses that these energies are meant to compare electronic trends, not room-temperature binding free energies. Mixing those single-step values with the best-isomer series would create a category error in any regression. citeturn18search0turn27search0

## Jiang and Ryde comparison

A key clarification: in the **published** Dalton Transactions paper, **Table 2 is not the cross-state summary table**. Table 2 is the set of **ten \(E_2\) structures without \(N_2\)**. The cross-state \(N_2\)-binding comparison is spread across **Table 1** (\(E_0/E_1\)), **Table 3** (\(E_2\)), **Table 5** (\(E_3\)), and **Table 7** (\(E_4\)), with additional best-structure checks in **ESI Tables S1–S3**. If you want the paper’s most systematic state-by-state comparison in the final published version, those are the tables to read together. citeturn47view0turn49view0turn50view0turn48view0turn45view0

Ryde’s methodological framework is broader than Pang’s. The paper compares **four functionals** — TPSS, r2SCAN, TPSSh, and B3LYP — and distinguishes different energetic measures, including the ordinary \(N_2\)-binding energy \(\Delta E_{N_2}\) and a **direct binding energy** \(\Delta E_{db}\) referenced to the same structure with \(N_2\) in the second coordination sphere. That means a Pang-style regression and a Ryde-style cross-functional survey are not interchangeable unless the metric is harmonized up front. citeturn49view0turn54view0

The broad conclusion of Jiang and Ryde is unusually clear even though the tables are method-sensitive. In the abstract and the conclusion, they state that **TPSS gives the strongest bonding and is the only functional that reproduces the experimental pattern of unfavorable binding for \(E_0\)–\(E_2\) and favorable binding for \(E_3\) and \(E_4\)**. **r2SCAN** gives favorable binding only to \(E_4\), while **TPSSh** and **B3LYP** do not yield favorable binding to any \(E_n\) state in the final comparison. They also emphasize that **B3LYP strongly favors triply protonated carbide states** and that structures with **two hydrides bridging Fe2/Fe6 and a partially dissociated S2B ligand** are the best \(E_4\)-type models for TPSS, r2SCAN, and TPSSh. citeturn12search3turn54view0turn3view2

The paper also gives an important warning for anyone doing quantitative fitting. When Jiang and Ryde add entropy corrections, the picture softens materially: with the **larger** entropy correction, **no** \(E_n\) state gives favorable \(N_2\) binding; with the **smaller** correction, **TPSS** still gives favorable binding for \(E_3\) and \(E_4\), whereas **r2SCAN** gives favorable binding only to \(E_4\). In the ESI they further note that some basis-set changes alter relative energies by more than **20 kJ/mol**, and Table S2 shows cases where the **electronic structure itself changes extensively** on the larger basis. That is exactly the sort of hidden heterogeneity that can flatten or distort a regression if all points are pooled indiscriminately. citeturn3view2turn36view1

## Hallmen and Kästner in context

For the paywalled 2015 paper, the abstract-level record is still useful. Hallmen and Kästner state that they examined \(N_2\) binding to a **reduced and protonated FeMo-cofactor including the central carbon ligand**, found that the **central ligand stabilizes the cluster**, and concluded that \(N_2\) can bind either **Fe** or **Mo**, with **Fe preferred** and **exo** modes more stable than **endo** modes. That makes the paper mechanistically relevant as a bridge between the older pre-carbide DFT literature and the later carbide-aware Fe-site binding work. citeturn29search3turn29search4

What the accessible sources do **not** safely provide is a trustworthy numeric extraction comparable to Pang’s or Ryde’s tabulated series. For that reason, Hallmen and Kästner is a sound **context source**, but not a safe **regression datapoint source** unless the full paper is opened and its energetic definitions are checked directly. citeturn29search4

## Earlier Kästner trajectory

The older Kästner/Blöchl sequence is still useful, but it sits on an older structural footing. In the 2003 JACS study, Schimpl, Petrilli, and Blöchl modeled the FeMo cofactor with a **central nitrogen ligand**, found that the **Fe–Mo cage opens** on \(N_2\) binding, identified both **axial** and **bridged** binding modes, and concluded that **Mo binding is less favorable than Fe-site binding**. That paper is historically important because it already places sulfur-bridge opening and Fe-site coordination at the center of the mechanism, but it predates the 2011 structural assignment of the interstitial **carbide** in modern FeMoco descriptions. citeturn29search1turn53search0turn53search1

The corrected 2005 JCP paper by Kästner, Hemmen, and Blöchl then pushed the mechanism one step further. Its central quantitative result is that **bridging** \(N_2\), activated by bonding to **two Fe sites**, lowers the energy for the first hydrogen transfer by **123 kJ/mol**, while an **axial** mode with an open sulfur bridge is **30 kJ/mol less reactive**. It also reports that the energetic ordering of axial and bridged binding modes reverses in favor of the **bridged** dinitrogen upon first protonation. citeturn32search2

The 2007 JACS paper continued the same DFT trajectory from **monoprotonated bound dinitrogen** all the way to ammonia release. Its abstract says that during the modeled catalytic conversion, nitrogen **bridges two Fe atoms**, a **cis-to-trans diazene** rearrangement occurs, **N–N bond cleavage is strongly exothermic**, and release of the **second ammonia** is facilitated by **re-closure of the sulfur bridge** after an intramolecular proton transfer. In other words, it is best read as a continuation of the 2003–2005 Fe-site, sulfur-lability, Fe-bridged pathway picture. citeturn34search0turn34search1

The short 2005 ChemPhysChem paper, *Towards an understanding of the workings of nitrogenase from DFT calculations*, is best treated as a **bibliographic waypoint** rather than a data source in this exercise. PubMed provides the citation but **no abstract**, so it is not a reliable place to mine numbers or detailed mechanistic constraints unless the full text is opened separately. citeturn52search0

## Regression-ready takeaways

If the immediate goal is a clean regression input, the most defensible numeric series is the **Pang/Bjornsson best-isomer QM/MM set**: \(E_0 = +69.45\), \(E_1 = +41.42\), \(E_2 = +7.95\), and \(E_4 = -17.15\ \text{kJ mol}^{-1}\). Those four points are internally consistent in definition, level of discussion, and structural framing. Jiang and Ryde can then be used as the **sensitivity analysis layer** that tells you how much those energies move with functional choice, alternative structural manifolds, and entropy treatment. citeturn24search0turn23search0turn27search0turn12search3turn3view2

The main thing not to do is pool together four different categories of numbers: Pang **best-isomer** values, Pang **single-step** values, Ryde **\(\Delta E_{N_2}\)** values, and Ryde **\(\Delta E_{db}\)** values. Those are different observables. A second thing not to do is treat Hallmen 2015 as numerically commensurate until the full text is in hand. A third is to ignore the structural regime change: the Pang sequence is not simply “binding gets linearly better with reduction.” It also crosses from resting-state-like manifolds (\(E_0/E_1\)) into **hydride-bearing, S2B-hemilabile manifolds** (\(E_2/E_4\)), which is exactly why both Pang and Ryde keep emphasizing hydride ligation, low-spin Fe configurations, and local coordination change as the real drivers of favorable binding. citeturn18search0turn12search3turn54view0

If a purely descriptive fit is still useful, an ordinary least-squares line through the four Pang best-isomer points gives approximately **\(E_{\text{bind}}(\text{kJ/mol}) = 63.26 - 21.63\,E_n\)** with **\(R^2 \approx 0.95\)**. That fit is fine as a compact summary of the monotonic trend in this one internally consistent series, but it should be treated as **descriptive only**, not mechanistic, because two of the four points already sit on different electronic and structural manifolds than the first two. citeturn24search0turn23search0turn27search0turn18search0turn12search3