{"id":"8d8ab6d2-1d98-4d85-b4b3-f94aab5c8afe","arxiv_id":"2505.12645","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Combining constrained orbital-freezing (FR) or ionized-fragment ALMO guesses with squared-gradient minimization makes orbital-optimized DFT reliably converge to charge-transfer excited states in large supramolecular systems.","lead":"This chemistry paper tests two ways to create good starting guesses for computing charge-transfer excited states in large molecules, then shows these guesses let a particular DFT method converge to the wanted excited state. The recipe makes it practical to calculate how electrons move between molecules in systems like dye-sensitized solar cells and supramolecular cages.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The large-system reliability claim is not independently validated, and the D102-TiO2 result in Table 4 shows a 1.97 eV FR-SGM/sTDA discrepancy that is never reconciled.","rationale":"The reader's conditional verdict is appropriate. The strongest support is the external EOM-CCSD(fT) benchmark on the dimer and the internally consistent applications to the cage and dye-TiO2 systems. The load-bearing gap is that the large-system half of the central claim is supported only by internal consistency (smooth DCT fits, density-difference plots) and not by any independent reference. The paper itself flags the relevant caveats: SGM's squared-gradient objective has undesired minima and cusps (Section 2), and the large-system SGM runs used a loose 10^-4 convergence threshold (Section 4.3). The D102-TiO2 row in Table 4 is the most concrete place where the method's behavior is anomalous relative to sTDA, and the paper gives no reference calculation to decide which method is correct. This does not invalidate the dimer benchmark or the JK2 and cage results, but it means the word 'reliably' in the central claim is not yet quantified or independently confirmed for large systems. A single targeted recomputation with an alternative guess and tighter convergence would settle whether the D102 endpoint is guess-dependent. Since this is an addressable validation gap and the paper already discusses several limitations, the conditional verdict stands without revision.","tokens_in":19066,"tokens_out":4713,"duration_ms":51964,"concrete_test":"Recompute the D102-TiO2 low-lying CT state with FR-SGM starting from the sTDA natural orbitals instead of the FR guess, and tighten the SGM convergence from 10^-4 to 10^-6. If the two initializations land on different stationary points, or if tightening convergence moves the energy by more than about 0.1 eV, the large-system convergence claim is not supported. As a complementary check, run full TDA with the same ωB97X-D functional rather than sTDA for this state; a value near 3.5 eV would indicate that FR-SGM converged to a different stationary point or an overestimated saddle point.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 4.2 establishes on the tetrafluoroethylene-ethylene dimer that FR and ALMO guesses put SGM in the correct basin, but the headline claim extends this reliability to large systems. In the large-system applications, the only evidence that the SGM endpoint is the targeted CT state is the DCT descriptor, density-difference plots, and smooth fits; there is no independent reference or state-overlap check. This matters because Section 2 states that the squared-gradient objective has undesired minima and cusps, and Section 4.3 admits SGM was run with a loose 10^-4 convergence criterion. The D102-TiO2 row of Table 4 is a concrete red flag: FR-SGM gives 5.45 eV versus 3.48 eV from sTDA for a dye-TiO2 CT state, opposite in sign to the roughly 1 eV red shifts reported for every JK2-TiO2 state. If sTDA is inaccurate, that should be demonstrated; if FR-SGM is inaccurate, the claim that it reliably converges to the targeted states fails for one of the two new chemical systems. The dimer benchmark cannot certify the large-system basin because it contains a single well-characterized low-lying ICT state, whereas the dye-TiO2 and cage systems have multiple close-lying CT and LMCT states.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents an improved initial-guess strategy for variational orbital-optimized DFT (OO-DFT) calculations of charge-transfer excitations, combining a frozen hole-and-electron constrained optimization (FR) with squared-gradient minimization (SGM). The FR guess is compared with an ALMO-based guess on the tetrafluoroethylene-ethylene dimer, where FR-SGM is shown to converge reliably to the targeted lowest ICT state with a mean error of 0.64 eV and very small variance against EOM-CCSD(fT). The method is then applied to two large supramolecular systems: a Pd coordination cage with PTZ-ANQ donor-acceptor ligands and two dye-TiO2 complexes. The paper also compares two range-separation tuning schemes for TD-DFT, the global density-dependent (GDD) tuning and a DCT-based tuning, recommending the former. The central claim is that with FR or ALMO guesses, SGM-based OO-DFT reliably converges to targeted CT states even in large systems.","tokens_in":19340,"tokens_out":7526,"duration_ms":78470,"significance":"If the large-system reliability claim holds, the FR-SGM method would be a practically useful tool for computing charge-transfer excitation energies in supramolecular and dye-semiconductor systems at DFT cost, with a clear prescription for generating good initial guesses. The dimer benchmark is a genuine strength: it provides an external reference (EOM-CCSD(fT)) and demonstrates that FR-SGM gives a small, systematic error with exceptionally low variance, while TDA errors are larger and more scattered. The comparison of tuning schemes is also useful for practitioners. However, the step from the dimer benchmark to the claim of reliable convergence in large systems is not fully backed by independent validation, and the unexplained discrepancy for D102-TiO2 in Table 4 is a concrete unresolved point. The conclusions about the cage system depend on an ad hoc shift parameter in the fit. These issues make the manuscript suitable for major revision rather than acceptance as is.","major_comments":[{"comment":"The FR-SGM result for the D102-TiO2 dye-TiO2 CT state is 5.45 eV, which is 1.97 eV higher than the sTDA value of 3.48 eV, while all JK2-TiO2 states are red-shifted relative to sTDA by up to about 1 eV. The manuscript does not explain this sign-reversed discrepancy or provide any independent reference for the dye-TiO2 states. Since this system is one of only two new large-system applications and the only one with an external comparison, the conclusion in Section 5 that 'FR-SGM reliably converged on the CT excitations also for these systems' is not supported for D102-TiO2 until the discrepancy is resolved, either by demonstrating a specific sTDA failure or by verifying the character of the FR-SGM state with an independent method.","section":"Section 4.4, Table 4"},{"comment":"The full-cage CT energies are fitted with the shifted expression E = a - b/(RDA - c), where c is set by hand to 3.4 Å without a reported justification or sensitivity analysis. The fitted IPD - EAA asymptote is then compared with the ligand-only fits to conclude that cage confinement lowers the CT energy and makes it accessible with a 400 nm pump. Because the asymptote and hence this physical conclusion depend directly on the arbitrary value of c, the analysis is not robust; the paper should either fit c, justify it, or demonstrate that the conclusions are insensitive to reasonable variations of c.","section":"Section 4.3, Eq. 13"},{"comment":"The large-system reliability of FR-SGM is inferred primarily from a dimer benchmark containing a single well-characterized low-lying ICT state, whereas the paper itself notes in Section 2 that the squared-gradient objective has undesired minima and cusps and that the quality of the initial guess determines the basin of attraction. The FR guess requires an a priori assignment of the hole and electron orbitals, and Fig. 3 shows that IMOM starting from related constrained guesses can converge to different, non-ICT states at short donor-acceptor distances. The cage and dye-TiO2 systems contain multiple close-lying CT and LMCT states, so the dimer evidence alone does not establish that FR-SGM reliably lands in the targeted basin for these larger systems. A characterization of the basin of attraction or tests with deliberately misassigned guesses would be needed to support the generalization.","section":"Section 4.2 and Section 4.3"},{"comment":"The large-system calculations use a 'rather loose' SGM convergence criterion of 10^-4, while the dimer benchmark uses 10^-5. Because SGM minimizes the squared gradient, a 10^-4 threshold on the squared gradient leaves a residual orbital gradient that is not negligible, and no estimate of the resulting energy error is provided. Since the large-system energies are used in quantitative fits and in the comparison between methods, the paper should either tighten the convergence or report how much the reported energies change when the threshold is reduced.","section":"Section 4.3, SGM convergence criterion"},{"comment":"The comparison of the GDD- and DCT-based tuning schemes excludes data points from the fits (e.g., the 56.3° point in Fig. 6A, the stacked conformer in Fig. 6B, and points marked with orange squares in Fig. 5) without a pre-defined, formal exclusion rule. Because the recommendation of LRC-omegaGDDPBE over DCT-based tuning rests on the smoothness and consistency of these fits, the post hoc exclusions could bias the comparison. The authors should either specify the outlier criterion in advance, report fits with and without the excluded points, or otherwise demonstrate that the conclusions are unchanged.","section":"Section 4.3, Figs. 5 and 6"}],"minor_comments":[{"comment":"The caption says 'the C matrix partitioning introduced in Fig. 2' but should refer to Fig. 1; the stray line 'Sincerely, Nicola Bogo' inside the Fock matrix display is a typographical artifact and should be removed.","section":"Figure 2 caption"},{"comment":"The term 'mean signed variance' is a misnomer because variance is non-negative; 'mean variance of the signed error' or simply 'variance' would be clearer.","section":"Table 1 and Section 4.2"},{"comment":"The text states that the 31.3° data point 'was found to be problematic also for other methods,' but it is not clear from the figure or caption whether this point is excluded from the fits for ALMO-SGM and FR-SGM; this should be stated explicitly.","section":"Section 4.3, Fig. 6 panels C and D"},{"comment":"The symbols a and b in Eq. 11 are defined as fitting parameters and later identified with IPD - EAA, but the identification is made only after the fits; please introduce the Mulliken identification before the fits are discussed in Table 2.","section":"Section 4.3, Eq. 11"},{"comment":"The text in Section 3 refers to 'the guess refinement method introduced in section 4.1,' but Section 4.1 is in the Results section; this cross-reference interrupts the narrative flow and could be rephrased.","section":"Section 3"},{"comment":"The statement 'we confirm the presence of low-lying D-A CT and LMCT excitations' is stronger than what the calculations support; 'predict' or 'find evidence for' would be more appropriate.","section":"Section 5"}],"recommendation":"major_revision","confidential_remarks":"The manuscript fits the scope of the journal. The conflict of interest is properly disclosed. The 'Sincerely, Nicola Bogo' line embedded in the Fock matrix display suggests the manuscript was assembled from a draft and needs careful proofreading. The core dimer benchmark is solid, but the large-system extension needs additional validation or at least a clear discussion of the D102-TiO2 discrepancy and the hand-set shift parameter in the cage fit."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The thing to know: this paper makes a good case that FR-SGM (freeze-and-release guess plus squared-gradient minimization) reliably finds targeted charge-transfer excited states in OO-DFT, but the evidence is much stronger for the small benchmark dimer than for the large systems it claims to handle. The D102-TiO2 result in Table 4 is a red flag that is never reconciled.\n\nWhat is actually new: the FR implementation in Q-Chem is a concrete realization of the freeze-and-release idea, and the paper provides a head-to-head comparison of FR and ALMO guesses combined with IMOM and SGM on a well-chosen benchmark, plus the first application to a Pd coordination cage and to dye-TiO2 complexes. The dimer benchmark is solid: mean error of 0.64 eV against EOM-CCSD(fT) with tiny variance, and the FR-SGM energy curve tracks the reference including the 3.5 Å point. That is a legitimate result and worth building on.\n\nThe soft spots are real but not fatal. First, the large-system claims rest only on density-difference plots and DCT values, with no independent reference or state-overlap check. The cage energies are internally consistent, but \"reliably converge\" is never quantified, and the SGM convergence criterion was loose (10^-4). Second, the D102-TiO2 FR-SGM energy is 5.45 eV versus 3.48 eV from sTDA—a 1.97 eV blue shift opposite in sign to every JK2 shift. The paper just reports it without comment. Maybe sTDA is wrong here, maybe FR-SGM landed on a different state; either way, that needs explanation before I would trust the large-system reliability claim. Third, some data points are excluded post hoc from the tuning fits (Figs. 5 and 6). They are clearly marked, which is honest, but it softens the tuning comparison.\n\nWho this is for: anyone using OO-DFT for CT states and people developing excited-state variational methods. It deserves a serious referee. The issues are addressable in revision—add a reference or overlap check for the large-system states, reconcile the D102-TiO2 discrepancy, report success rates—but they need to be addressed before the central claim is fully supported. I would send it to review.","headline":"Plausible recipe for reliable OO-DFT convergence to CT states, but the large-system validation is thinner than the dimer benchmark and the D102-TiO2 anomaly needs reconciling.","tokens_in":19921,"tokens_out":1986,"would_cite":true,"duration_ms":20639,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A constrained initial guess plus squared-gradient minimization lets variational DFT converge to charge-transfer excited states in large supramolecular systems.","keywords":["charge-transfer excitations","orbital-optimized DFT","squared-gradient minimization","frozen-orbital guess","ALMO guess","range-separated hybrid functionals","supramolecular photochemistry","dye-sensitized solar cells"],"falsifier":"Run FR-SGM on the tetrafluoroethylene-ethylene dimer at R_DA = 3.5-5 angstrom with deliberately corrupted guesses, for instance with the hole and electron orbitals swapped or rotated by a few degrees, and check whether any converge to a lower non-CT stationary point instead of the target ICT state; if any chemically relevant geometry does so, the claim that FR provides reliable convergence would need to be qualified.","tokens_in":18835,"feed_emoji":"🧪","tokens_out":7480,"duration_ms":74128,"temperature":0.7,"pith_summary":"This paper tackles a practical bottleneck in computing charge-transfer (CT) excited states of large molecular systems: variational orbital-optimized DFT (OO-DFT) has the accuracy but often fails to converge to the desired excited state because the energy surface is full of other stationary points. The authors propose that a good initial guess solves the problem. They construct two such guesses, one from a constrained optimization that freezes the electron and hole orbitals (FR), and one from absolutely-localized molecular orbitals on ionized fragments (ALMO), and then relax the constraint with squared-gradient minimization (SGM). On the tetrafluoroethylene-ethylene dimer benchmark, the resulting FR-SGM energies have a mean error of 0.64 eV and a per-point variance of 0.005 eV against a coupled-cluster reference, versus 1.27 eV for the standard TDA approach. The same recipe converges reliably on a large palladium coordination cage and on dye-TiO2 complexes, offering a low-scaling route to CT excitations in systems too large for wavefunction methods.","feed_headline":"A constrained initial guess makes charge-transfer DFT converge","feed_subtitle":"Freezing the hole and electron, then minimizing the squared gradient, beats TDA on a benchmark dimer: 0.64 vs 1.27 eV.","key_machinery":"The load-bearing object is the squared-gradient objective $$\\$\\Delta$=\\sum_{ja}\\left|\\frac{\\partial E}{\\partial \\theta_{ja}}\\right|^2,$$ where $\\theta_{ja}$ are the occupied-virtual orbital rotations. Feeding this objective to a direct-optimization algorithm turns the search for a saddle point on the electronic energy surface into a minimization problem, but it introduces undesired minima and cusps, so the starting point must lie in the correct quadratic well. The FR guess supplies that starting point: the MO coefficient matrix $C$ is partitioned and reordered so that the hole and electron orbitals form frozen blocks, and only the remaining occupied-virtual rotations enter the gradient; a second-order geometric direct minimization then produces a density close to the target CT state. The ALMO guess is an alternative that treats donor and acceptor as ionized fragments and works when the system has a clean fragment separation. SGM is the algorithm that carries the final optimization once the frozen blocks are released.","core_discovery":"On the paper's own terms, the discovery is that variational OO-DFT can be made to converge to target intermolecular charge-transfer excited states in large systems, provided the optimization starts from a guess that already has the correct electron-hole character. The FR guess is made by reordering the MO coefficient matrix so that the hole and electron orbitals are frozen and excluded from the gradient, then optimizing the remaining rotations with geometric direct minimization; the ALMO guess is made by computing the donor and acceptor as isolated ionic fragments. Once the constraint is lifted, squared-gradient minimization (SGM) converges to the targeted saddle point. The paper demonstrates this on the tetrafluoroethylene-ethylene dimer, where FR-SGM gives a mean signed error of -0.64 eV and a mean variance of 0.005 eV against EOM-CCSD(fT), compared with -1.27 eV for TDA, and then applies the method to the PTZ-ANQ donor-acceptor pair in a Pd coordination cage and to dye-TiO2 complexes. The paper also reports that the global density-dependent tuning of the range-separation parameter (omega_GDD) is smooth and reliable for TDA, while the DCT-based tuning gives irregular parameters.","pith_inferences":["Beyond the paper: if the near-constant FR-SGM shift holds for other donor-acceptor systems, then differences of CT energies across conformers, rather than absolute excitation energies, may be the most defensible quantity to compare with experiment.","Beyond the paper: the paper never characterizes the basin of attraction of SGM; a practical extension would be to perturb the FR hole and electron orbitals systematically, for example by small rotations, and map which perturbations still converge to the target CT state, giving a quantitative convergence radius.","Beyond the paper: the success of ALMO and FR guesses suggests that any cheap method that encodes the correct ion-pair electrostatics of the CT state, such as a constrained DFT density or a fragment charge assignment, could serve as a guess, not just the two tested here.","Beyond the paper: because the SGM objective squares the gradient, the method's condition number is intrinsically worse than ordinary minimization; an accelerated SGM with a better preconditioner could make the approach routine on even larger systems than those tested."],"forward_implications":["FR-SGM and ALMO-SGM give a low-scaling route to intermolecular CT excitation energies in large supramolecular systems, where wavefunction references are prohibitively expensive.","Because FR-SGM's errors are nearly a constant shift (variance 0.005 eV on the dimer scan), relative CT energies along a conformational coordinate are expected to be more reliable than the absolute numbers.","For TDA, the omega_GDD reparametrization of LRC-omegaPBE is recommended for CT excitations when conformational changes matter; DCT-based tuning should be used only through its asymptotic limit.","In the full Pd cage, FR-SGM predicts PTZ-ANQ CT excitations in the visible range for the stacked conformer, plus LMCT excitations in the same energy window, leaving open the possibility of relaxation to LMCT states after a 400 nm pump.","ALMO-SGM estimates an inter-cage CT excitation at 3.81 eV in a model of the interlocked cage, within the energy range of intra-cage CT states."],"supporting_citations":[{"why":"Supplies the benchmark tetrafluoroethylene-ethylene dimer system and documents the convergence failures that motivate the guess strategy.","marker":"[10]"},{"why":"Introduces the freeze-and-release constrained-optimization idea on which the FR guess is built and demonstrates it on the same benchmark.","marker":"[17]"},{"why":"Provide the squared-gradient minimization objective and the OO-DFT framework used for the final variational optimization.","marker":"[11,12]"},{"why":"Provides the IMOM method against which the refined and ALMO guesses are compared.","marker":"[16]"},{"why":"Supplies the geometric direct minimization algorithm used inside the constrained optimization for the FR guess.","marker":"[19]"},{"why":"Provide the ALMO equations used to construct the ionized-fragment guess.","marker":"[20,21]"},{"why":"Defines the DCT electron-hole distance descriptor used throughout to quantify charge-transfer character.","marker":"[46]"},{"why":"Provides the experimental coordination-cage system and its measured photophysics that the large-system calculations target.","marker":"[22]"},{"why":"Provides the dye-TiO2 complex structures used for the semiconductor-interface tests.","marker":"[23]"}],"fun_headline_variants":["Constrained guess tames charge-transfer DFT convergence","Freeze the hole and electron to nail charge-transfer states","Two-step guess cracks charge-transfer convergence in DFT","Variational DFT converges to CT states with smart start","Initial guess with right electron-hole character solves CT-DFT"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The result rests on the assumption that the initial guess is close enough to the target charge-transfer state that the squared-gradient optimization lands in the right one of its many valleys; the paper demonstrates this on three systems but never measures how close is close enough.","fun_headline_variants_meta":{"raw":{"variants":["Constrained guess tames charge-transfer DFT convergence","Freeze the hole and electron to nail charge-transfer states","Two-step guess cracks charge-transfer convergence in DFT","Variational DFT converges to CT states with smart start","Initial guess with right electron-hole character solves CT-DFT"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000167,"raw_usage":{"total_tokens":1279,"prompt_tokens":992,"completion_tokens":287,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":608,"completion_tokens_details":{"reasoning_tokens":209}},"tokens_in":608,"tokens_out":287,"duration_ms":3138,"temperature":1.0,"reasoning_tokens":209,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T20:30:00.862335+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run FR-SGM on the tetrafluoroethylene-ethylene dimer at R_DA = 3.5-5 angstrom with deliberately corrupted guesses, for instance with the hole and electron orbitals swapped or rotated by a few degrees, and check whether any converge to a lower non-CT stationary point instead of the target ICT state; if any chemically relevant geometry does so, the claim that FR provides reliable convergence would need to be qualified.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the benchmark tetrafluoroethylene-ethylene dimer system and documents the convergence failures that motivate the guess strategy."},{"cited_title":"L.; Selenius, E.; Levi, G","cited_arxiv_id":null,"evidence_quote":"Introduces the freeze-and-release constrained-optimization idea on which the FR guess is built and demonstrates it on the same benchmark."},{"cited_title":"M.; Gilbert, A","cited_arxiv_id":null,"evidence_quote":"Provides the IMOM method against which the refined and ALMO guesses are compared."},{"cited_title":"A geometric approach to direct minimization","cited_arxiv_id":null,"evidence_quote":"Supplies the geometric direct minimization algorithm used inside the constrained optimization for the FR guess."},{"cited_title":"A qualitative index of spatial extent in charge-transfer excitations","cited_arxiv_id":null,"evidence_quote":"Defines the DCT electron-hole distance descriptor used throughout to quantify charge-transfer character."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the experimental coordination-cage system and its measured photophysics that the large-system calculations target."},{"cited_title":"C.; Pastore, M.; Bahmann, H","cited_arxiv_id":null,"evidence_quote":"Provides the dye-TiO2 complex structures used for the semiconductor-interface tests."}],"review_version":1}