CTNF 18/377,526 CTNF 98555 DETAILED ACTION 12-151 AIA 26-51 12-51 Status of Claims Claim(s) 1-20 are pending and are examined herein. Claim(s) 1-20 are rejected under 35 U.S.C. §§102 and 103. Notice of Pre-AIA or AIA Status 07-03-aia AIA 15-10-aia The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA. Information Disclosure Statement The information disclosure statement IDS(s) submitted on October 06, 2023 and December 01, 2023 are in compliance with the provisions of 37 CFR 1.97 and have been considered by the examiner. Claim Rejections - 35 USC § 112 07-30-02 AIA The following is a quotation of 35 U.S.C. 112(b): (b) CONCLUSION. —The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention. The following is a quotation of 35 U.S.C. 112 (pre-AIA), second paragraph: The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention. 07-34-01 Claim(s) 3-7 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA), as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor, for pre-AIA the applicant regards as the invention. Regarding Claim 3 , the claim recites the limitation “ wherein the one or more atomic values include total energy, atomic forces, atomic stresses, atomic charges, and/or polarization” without a clear antecedent for “the one or more atomic values .” The claimed recitation of “the one or more atomic values” was not introduced in the parent claim 1. Thus, it is unclear whether the term refers to the one or more values of the materials recited in claim 1 or introduces a new element. It is interpreted as referring to the values of the material. Regarding Claim 4 , the claim recites the limitation “ wherein the first set of structures are near-equilibrium structures .” However, the term “near” is a relative term that renders the claim indefinite. The claim nor the specification provides objective boundaries for determining/quantifying what constitute as “near.” Because the claim uses a term of degree and fails to clearly define it, a person of ordinary skill in the art cannot determine with reasonable certainty the scope of the near-equilibrium structures. Regarding Claim 5 , the claim recites the limitation “ wherein the near-equilibrium structures are substantially close to but not in a state of thermodynamic equilibrium .” However, the limitation uses terms and phrases such as “near” and “substantially close to but not” that define subjective terms and/or terms of degree which renders the scope of the claim indefinite. The claim does not define the terms and the specification does not provide a standard for determining or measuring the meaning of the terms of degree. Accordingly, a person of ordinary skill in the art cannot determine with reasonable certainty the scope of the claim. Regarding Claim 6 , the claim recites the limitation “ wherein the second set of samples are far-from-equilibrium structures .” The claim limitation recites “the second set of samples” without any prior recitation of “a second set of samples” in claims 4 or 1. Thus, the recitation of “the second set of samples” fails to provide proper antecedent basis in the claim, where it is unclear whether “second set of samples” refers to the same element as “second set of structures” in claim 1 or represents a different element. Additionally, the limitation recitation of “ far-from-equilibrium structures ” introduces a term of degree “far” without any defined measure, metric, or criteria for quantifying the degree point. The specification does not provide a standard for measuring the meaning of the term of degree and one of ordinary skill in the art wouldn’t be reasonably apprised of the metes and bounds of the claim scope. Regarding Claim 7 , the claim recites the limitation “ wherein the far-from-equilibrium structures are substantially far from a state of thermodynamic equilibrium .” Similarly, the claim limitation recites terms and phrases such as “far” and “substantially far” that define subjective terms and/or terms of degree which renders the scope of the claim indefinite. The claim does not define the terms and the specification does not provide a standard for determining or measuring the meaning of the terms of degree. Accordingly, one of ordinary skill in the art would not be reasonably apprised of the scope of the claim. In view of the above, the Examiner respectfully requests that Applicant thoroughly review the claims for compliance with the requirements set forth under 35 U.S.C. § 112. Appropriate correction is required. Claim Rejections - 35 USC § 102 07-07-aia AIA 07-07 The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action: A person shall be entitled to a patent unless – 07-08-aia AIA (a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention. 07-12-aia AIA (a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention. 07-15 AIA Claim (s) 1-3, and 10-13 are rejected under 35 U.S.C. 102( a)(1 ) as being anticipated by Meziere et al., (NPL: “Accelerating Training of MLIPs Through Small-Cell Training.” (April 2023)), hereinafter Meziere . Regarding Claim 1, Meziere discloses the following: An active learning machine learning interatomic potential (MLIP) training method comprising: ( Meziere , [Abstract] “Automated methods, such as active learning and on-the-fly learning, allow for the construction of reliable training sets, but these processes can be very resource-intensive when training a potential for use in large-scale simulations... Here, we demonstrate an easy-to-implement small-cell structures training protocol and use it to train a potential for zirconium and hydrides.” [p. 2, Section: I] “We demonstrate small-cell training with the moment tensor potential (MTP) as implemented in the MLIP-2 package, which includes active learning [3]. MTPs use an invariant polynomial basis to represent atomic densities... While we use the MTP/MLIP 2 framework, the methods we describe can be applied to other MLIPs that utilize active learning.”) receiving one or more datasets associated with a material ( Meziere , [P. 2, Section: II] “We demonstrate the effectiveness of small-cell training on the zirconium and Zr-H systems. Many of zirconium’s industrial uses are related to nuclear power applications because of its low neutron cross-section. Most current commercial reactor designs include fuel cladding made of zirconium alloys.” [P.3, Section: III] “To create a set of candidate structures for training, we enumerate supercells of the δ, γ, and phases while varying hydrogen concentration. Systematically enumerating possible structures is straightforward, but the number of these structures grows exponentially with cell size [51]. Small-cell training leads to a training set that efficiently represents all candidate structures and their relaxation trajectories.... Our approach begins by applying active learning with the δ, γ, and ϵ phases themselves... As shown in Fig. 2 (a), training with the initial set of structures required around 300 structures to be added to the training...”) ; and actively learning a dynamic trajectory in response to the one or more datasets associated with the material, the dynamic trajectory sampling a first set of structures and progressing to a second set of structures to create an actively learned MLIP to predict one or more values of the material. ( Meziere , [P. 1, Section: I] “Active learning, on the other hand, uses a predefined set of simulations to conduct training. Once the potential successfully finishes all simulations in the predefined set without triggering more DFT calculations, the potential is considered sufficient and no further training occurs.” [P.3, Section: III] “To create a set of candidate structures for training, we enumerate supercells of the δ, γ, and phases while varying hydrogen concentration. Systematically enumerating possible structures is straightforward, but the number of these structures grows exponentially with cell size [51]. Small-cell training leads to a training set that efficiently represents all candidate structures and their relaxation trajectories.... Our approach begins by applying active learning with the δ, γ, and ϵ phases themselves. Once the MLIP can relax these structures without the active learning proto col requesting additional DFT calculations, training be gins on structures with the next smallest supercells. The process is repeated for successively larger cells until the MLIP can successfully relax all of the enumerated structures.” [P. 4, Section: IV] “Active learning was then used with a two-atom heating simulation of α phase and a two-atom cooling simulation of β phase. After these simulations ran successfully without requiring additional DFT calculations, training was performed for four-atom simulations for both the heating and cooling simulations. The size of the supercells used during the simulations was increased until the potential could model the full-scale phase transition simulation, finding that the potential captured the phase transition after training on molecular dynamics temperature ramps containing only eight atoms (Fig. 4).” Further see Fig. 2(b).) [ Examiner’s Note : Meziere describes active learning applied to molecular dynamic trajectories of small-cell structures (e.g., two-atom) and progressing to large-scale simulation. The proposed small-cell training protocol creates a robust and effective MLIP for modeling complex material properties.]. Regarding Claim 2, Meziere teaches the elements of claim 1 as outlined above, and further teaches: predicting the one or more values of the material with the actively learned MLIP. ( Meziere , [P. 4, Section: III] “Training for stable structures also allowed the potential to accurately model related material properties. The equilibrium structures, elastic constants, and phonon band structures of the α phase and the hydride , δ, and γ phases were successfully modeled by the potential, as shown in the appendix. Thus, small-cell training is an effective tool to find stable structures of material systems and related material properties.” [P. 5, Section: V] “When compared with the MTP trained with large supercells, the phase transition of our MTP has good agreement.” [P. 8, Section: VIII] “Predictions of the elastic properties of the four phases were also reproduced with reasonable accuracy.”) Regarding Claim 3, Meziere teaches the elements of claim 1 as outlined above, and further teaches: wherein the one or more atomic values include total energy, atomic forces, atomic stresses, atomic charges, and/or polarization. ( Meziere , [P. 5, Section: V] “The initial state of the training set of the second potential included large cells with random atomic displacements. Then, the second potential was trained directly on the full-scale simulation with an active learning strategy.” [P. 1, Section: I] “MLIPs are created by fitting flexible, parameter-rich models to DFT data [3–6]. These potentials can repro duce energies, forces, and stresses for structures nearly as accurately as DFT, but are orders of magnitude faster [7]... If the level of confidence in a prediction is insufficient, energies and forces of that structure are calculated with DFT and added to the training set. In this way, active learning and on-the fly learning iteratively improve the training set until the resulting potential reproduces the material properties of interest [14, 19].” [P. 3, Section: II] “From an atomistic simulation perspective, no current classical interatomic interaction potential is able to describe the structure and energetics of the multiple Zr-H phases, which limits atom-scale studies of this system to the small length and time scales accessible to DFT,”) Regarding Claim 10, Meziere discloses the following: An active learning machine learning interatomic potential (MLIP) training method comprising: ( Meziere , [Abstract] “Automated methods, such as active learning and on-the-fly learning, allow for the construction of reliable training sets, but these processes can be very resource-intensive when training a potential for use in large-scale simulations... Here, we demonstrate an easy-to-implement small-cell structures training protocol and use it to train a potential for zirconium and hydrides.” [p. 2, Section: I] “We demonstrate small-cell training with the moment tensor potential (MTP) as implemented in the MLIP-2 package, which includes active learning [3]. MTPs use an invariant polynomial basis to represent atomic densities... While we use the MTP/MLIP 2 framework, the methods we describe can be applied to other MLIPs that utilize active learning.”) receiving one or more datasets associated with a material; ( Meziere , [P. 2, Section: II] “We demonstrate the effectiveness of small-cell training on the zirconium and Zr-H systems. Many of zirconium’s industrial uses are related to nuclear power applications because of its low neutron cross-section. Most current commercial reactor designs include fuel cladding made of zirconium alloys.” [P.3, Section: III] “To create a set of candidate structures for training, we enumerate supercells of the δ, γ, and phases while varying hydrogen concentration. Systematically enumerating possible structures is straightforward, but the number of these structures grows exponentially with cell size [51]. Small-cell training leads to a training set that efficiently represents all candidate structures and their relaxation trajectories.... Our approach begins by applying active learning with the δ, γ, and ϵ phases themselves... As shown in Fig. 2 (a), training with the initial set of structures required around 300 structures to be added to the training...”) ; and actively learning a dynamic trajectory in response to the one or more datasets associated with the material, the dynamic trajectory includes a temperature ramping to create an actively learned MLIP to predict one or more atomic values of the material. ( Meziere , [P. 1, Section: I] “Active learning, on the other hand, uses a predefined set of simulations to conduct training. Once the potential successfully finishes all simulations in the predefined set without triggering more DFT calculations, the potential is considered sufficient and no further training occurs.” [P.3, Section: III] “To create a set of candidate structures for training, we enumerate supercells of the δ, γ, and phases while varying hydrogen concentration. Systematically enumerating possible structures is straightforward, but the number of these structures grows exponentially with cell size [51]. Small-cell training leads to a training set that efficiently represents all candidate structures and their relaxation trajectories.... Our approach begins by applying active learning with the δ, γ, and ϵ phases themselves. Once the MLIP can relax these structures without the active learning proto col requesting additional DFT calculations, training be gins on structures with the next smallest supercells. The process is repeated for successively larger cells until the MLIP can successfully relax all of the enumerated structures.” [P. 4, Section: IV] “Active learning was then used with a two-atom heating simulation of α phase and a two-atom cooling simulation of β phase. After these simulations ran successfully without requiring additional DFT calculations, training was performed for four-atom simulations for both the heating and cooling simulations. The size of the supercells used during the simulations was increased until the potential could model the full-scale phase transition simulation, finding that the potential captured the phase transition after training on molecular dynamics temperature ramps containing only eight atoms (Fig. 4).” Further see Fig. 2(b) and Section V.) [ Examiner’s Note : Meziere describes active learning applied to molecular dynamic trajectories of small-cell structures (e.g., two-atom) and progressing to large-scale simulation (molecular dynamics temperature ramps). The proposed small-cell training protocol creates a robust and effective MLIP for modeling complex material properties.] Regarding Claim 11, Meziere teaches the elements of claim 10 as outlined above, and further teaches: predicting the one or more atomic values of the material with the actively learned MLIP. ( Meziere , [P. 4, Section: III] “Training for stable structures also allowed the potential to accurately model related material properties. The equilibrium structures, elastic constants, and phonon band structures of the α phase and the hydride , δ, and γ phases were successfully modeled by the potential, as shown in the appendix. Thus, small-cell training is an effective tool to find stable structures of material systems and related material properties.” [P. 5, Section: V] “When compared with the MTP trained with large supercells, the phase transition of our MTP has good agreement.” [P. 8, Section: VIII] “Predictions of the elastic properties of the four phases were also reproduced with reasonable accuracy.”) Regarding Claim 12, Meziere teaches the elements of claim 10 as outlined above, and further teaches: wherein the temperature ramping starts at a lower temperature and advances to higher temperatures. ( Meziere , [Pp. 4-5, Section: IV] “Active learning was then used with a two-atom heating simulation of α phase and a two-atom cooling simulation of β phase... The size of the supercells used during the simulations was increased until the potential could model the full-scale phase transition simulation, finding that the potential captured the phase transition after training on molecular dynamics temperature ramps containing only eight atoms (Fig. 4).” Fig. 4(b) (b) The MLIP trained on these small cells gives similar results to one trained using active learning on large-cell structures, but required 11.5 times less cpu time for training.) [ Examiner’s Note : Meziere defines the “heating simulation” which starts at a lower temperature and advances to higher temperature (i.e., fig. 4 depicts Volume vs. Temp plot showing the temperature progresses from 600k to 1800k), as part of the active learning molecular dynamics temperature ramps.] Regarding Claim 13, Meziere teaches the elements of claim 12 as outlined above, and further teaches: wherein the lower temperature is lower than the higher temperatures. ( Meziere , [Pp. 4-5, Section: IV] “Active learning was then used with a two-atom heating simulation of α phase and a two-atom cooling simulation of β phase... The size of the supercells used during the simulations was increased until the potential could model the full-scale phase transition simulation, finding that the potential captured the phase transition after training on molecular dynamics temperature ramps containing only eight atoms (Fig. 4).” Fig. 4(b) (b) The MLIP trained on these small cells gives similar results to one trained using active learning on large-cell structures, but required 11.5 times less cpu time for training.) [ Examiner’s Note : Meziere defines the “heating simulation” which starts at a lower temperature and advances to higher temperature (i.e., fig. 4 depicts Volume vs. Temp plot showing the temperature progresses from 600k to 1800k), as part of the active learning molecular dynamics temperature ramps.] 07-15 AIA Claim( s) 16-17, and 19 are r ejected under 35 U.S.C. 102(a )(1) as being a nticipated b y K ulichenko et al., (NPL: " Uncertainty-driven dynamics for active learning of interatomic potentials ." (January 2023)), hereinafter, Kulichenko. R egarding Claim 16, Kulichenko discloses the following: An active learning machine learning interatomic potential (MLIP) training method comprising: ( Kulichenko , [Abstract] “Machine learning (ML) models, if trained to data sets of high-fidelity quantum simulations, produce accurate and efficient interatomic potentials. Active learning (AL) is a powerful tool to iteratively generate diverse data sets. In this approach, the ML model provides an uncertainty estimate along with its prediction for each new atomic configuration... Here we develop a strategy to more rapidly discover configurations that meaningfully augment the training data set. The approach, uncertainty driven dynamics for active learning (UDD-AL), modifies the potential energy surface used in molecular dynamics simulations to favor regions of configuration space for which there is large model uncertainty.” [p. 2, Col. 1] “A well-established practical strategy for AL with NN potentials is ‘query by committee’47 (QBC), where the estimate of uncertainty is the disagreement between a collection of models within an ensemble.”) [ Examiner’s Note : The paper defines an active learning approach for training ML-based interatomic potentials (ANI NN potentials).] receiving one or more datasets associated with a material; ( Kulichenko , [P. 8, Col. 1, Section: Method] “The initial training set consisted of 125 glycine geometries that span the near-equilibrium structures of the glycine GM. These data were acquired from a separate 5-ps MD trajectory at 350 K with a 0.5-fs time step, initialized from the glycine GM.” [P. 4, Section: Glycine conformational space sampling] “An ensemble of NN potentials for the first AL iteration is trained on the initial data set of 125 conformers, spanning the near-equilibrium structures of the glycine GM. At each subsequent AL iteration, the MD simulation employs an ensemble of ANI-type NN potentials (Active learning section and ref. 57), trained on the initial data and data accumulated on all previous AL iterations. The starting geometries for the MD simulations and the initial training set contain only near- equilibrium geometries of a glycine GM (Fig. 2a).”) [ Examiner’s Note : The paper describes the initial dataset of structural conformers associated with materials (e.g., glycine, acetylacetone) obtained for the active learning.] and actively learning a dynamic trajectory in response to the one or more datasets associated with the material, the dynamic trajectory includes a biased sampling to create an actively learned MLIP to predict one or more atomic values of the material. ( Kulichenko , [P. 1, Col. 2] “Active learning (AL)33,34 attempts to expand the data set in areas where the ML model is most uncertain, which leads to more rapid model improvement. Another feature of AL is that it can employ physically meaningful dynamical trajectories for the sampling of configurations. In this Article we demonstrate how to keep these benefits of AL, while accelerating the rate of new data collection.” [P. 2, Col. 2] “In this Article, following the idea of QBC and ensemble uncertainty, we propose an AL sampling algorithm biased towards regions of high uncertainty—uncertainty-driven dynamics (UDD)... Thus, biasing MD in the direction of high ensemble uncertainty encourages the dynamics to visit new configurations, which are relevant for improving the diversity of the training set.” [p. 4, Col. 2] “At each subsequent AL iteration, the MD simulation employs an ensemble of ANI-type NN potentials (Active learning section and ref. 57), trained on the initial data and data accumulated on all previous AL iterations... Each MD simulation is terminated when the system meets the uncertainty selection criterion ρ of 0.35kcal × mol−1 ×NA −1/2 (Active learning section). If the MD simulation reaches the time limit, then the structure from the trajectory with the highest uncertainty is selected... After completing the entire AL procedure, the final models are trained on 1,280 glycine conformers collected during the procedure (+125 conformers in the initial training set). To access the accuracy of the four models, we use a test set of 50,000 glycine structures from a 50-ns MD simulation via an ANI-1ccx potential run at 400 K with a 0.5-fs time step19.” [P. 8, Section: Methods, Col. 1] “Each MD simulation was terminated when the system met the uncertainty selection criterion ρ... At each AL iteration, the MD was driven by an ensemble of ANI-type ML potentials, trained on initial data and data accumulated on previous AL iterations... The final data set had 1,280 data points sampled in the AL procedure and 125 data points from the initial data set.”) the dynamic trajectory includes a biased sampling ( Kulichenko , [P. 3, Section: Results] “The bias potential goes to zero in the absence of uncertainty, E b i a s ( 0 ) = 0 . Configurations with large uncertainty, ρ ≫ B, are favored by a bias energy of magnitude E b i a s ≈ - A .”) [ Examiner’s Note : The paper defines an active learning approach for training ML-based interatomic potentials (ANI NN potentials), it specifically learns dynamic MD trajectories at each iteration in response to the collected training dataset, the also introduces the uncertainty-based bias potential sampling into the trajectory. The actively trained model is then used to predict atomic energy and forces of materials.] Regarding Claim 17, Kulichenko teaches the elements of claim 16 as outlined above, and further teaches: wherein the biased sampling includes a subset of structures of the material prioritizing one or more characteristics of the one or more datasets associated with the material. ( Kulichenko , [P. 4, Section: Glycine conformational space sampling, Col. 2] “Each MD simulation is terminated when the system meets the uncertainty selection criterion ρ of 0.35kcal × mol−1 ×NA −1/2 (Active learning section). If the MD simulation reaches the time limit, then the structure from the trajectory with the highest uncertainty is selected.” [P. 8, Section: Methods, Col. 1] “Each MD simulation was terminated when the system met the uncertainty selection criterion ρ. The initial training set consisted of 125 glycine geometries that span the near-equilibrium structures of the glycine GM.”) [ Examiner’s Note : the UDD-AI biased sampling approach selects a specific set exceeding the uncertainty threshold, which prioritizing the characteristics of the one or more datasets with high model uncertainty.] Regarding Claim 19, Kulichenko teaches the elements of claim 16 as outlined above, and further teaches: wherein the biased sampling is a biased potential. ( Kulichenko , [P. 2, Section: Results] “Like metadynamics, the UDD-AL method modifies the physical energy by adding a bias potential, Ebias .” [P. 2, Section: Results] “The magnitude A and width B of the biasing should be selected empirically. The bias potential goes to zero in the absence of uncertainty, Ebias (0) = 0. Configurations with large uncertainty, ρ ≫ B, are favored by a bias energy of magnitude Ebias ≈ −A. Forces derived from the bias potential are strongest when the uncertainty ρ is of the same order as the parameter B.” [P. 4, Section: Glycine conformational space sampling, Col. 2] “Perhaps the most common way to accelerate sampling of high energy states is to run high-T MD. Thus, to illustrate the difference between the bias potential and a simple temperature increase, we also compared the low-T 350 K UDD-AL with the high-T MD-ALs at 600 K and 1,000 K simulation conditions (Fig. 1b). As in the case of UDD-AL sampling, the temperature is increased at the 15th iteration.”) Claim Rejections - 35 USC § 103 07-20-aia AIA The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action: A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made. 07-23-aia AIA The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows: 1. Determining the scope and contents of the prior art. 2. Ascertaining the differences between the prior art and the claims at issue. 3. Resolving the level of ordinary skill in the pertinent art. 4. Considering objective evidence present in the application indicating obviousness or nonobviousness. 07-21-aia AIA Claim (s) 4-7, 9, and 15 are rejected under 35 U.S.C. 103 as being unpatentable over Meziere in view of Kulichenko et al., (NPL: "Uncertainty-driven dynamics for active learning of interatomic potentials." (January 2023)) . Regarding Claim 4, Meziere teaches the elements of claim 1 as outlined above: While Meziere defines the equilibrium structures to accurately model related material properties, Meziere does not appear to explicitly teach: wherein the quantum computing function comprises a circuit cutting process . However, Meziere in view of Kulichenko teaches the limitation: wherein the first set of structures are near-equilibrium structures. ( Kulichenko , [P. 4, Section: Glycine conformational space sampling, Col. 2] “An ensemble of NN potentials for the first AL iteration is trained on the initial data set of 125 conformers, spanning the near- equilibrium structures of the glycine GM. At each subsequent AL iteration, the MD simulation employs an ensemble of ANI-type NN potentials (Active learning section and ref. 57), trained on the initial data and data accumulated on all previous AL iterations. The starting geometries for the MD simulations and the initial training set contain only near-equilibrium geometries of a glycine GM (Fig. 2a).” [P. 8, Section: Methods] “Each MD simulation was terminated when the system met the uncertainty selection criterion ρ. The initial training set consisted of 125 glycine geometries that span the near-equilibrium structures of the glycine GM. These data were acquired from a separate 5-ps MD trajectory at 350 K with a 0.5-fs time step, initialized from the glycine GM. Every 80th MD step was included in the initial data training set.”) Meziere and Kulichenko are from the same field of endeavor and their disclosure generally relates to (learning interatomic potentials). Accordingly, at the effective filing date, it would have been prima facie obvious to one ordinarily skilled in the art to modify the combination of Meziere and Kulichenko to incorporate the UDD-AL sampling based active learning method as taught by Kulichenko. One would have been motivated to make such a combination in order to produce accurate and efficient interatomic potentials to efficiently explore the chemically relevant configuration space, which may be inaccessible using regular dynamical sampling at target temperature conditions (Kulichenko [Abstract]). Regarding Claim 5, Meziere in view of Kulichenko teaches the elements of claim 4 as outlined above: wherein the near-equilibrium structures are substantially close to but not in a state of thermodynamic equilibrium. ( Kulichenko , [P. 2, Col. 1, Para. 2] “MD is susceptible to trapping in near-minimum conformations and only rarely enters chemically important regions such as transition states, which are key data for reactive simulations of chemical processes.” [P. 4, Section: Glycine conformational space sampling] “Depending on the AL iteration, near-GM structures have a value of ρ of 0.024 ± 0.005kcal ×mol−1 ×NA −1/2 (σE = 0.15 ± 0.03 kcal mol−1). We thus selected A equal to 15.4 kcal mol−1, which corresponds to a bias potential of ~15.0 kcal mol−1 at near-GM values of ρ (with B = 0.12 kcal mol−1). We further increased A by 15% at the 140-ps time step if the uncertainty criterion was not met at this simulation stage.... The starting geometries for the MD simulations and the initial training set contain only near-equilibrium geometries of a glycine GM (Fig. 2a). Stated differently, NNs have no initial information about higher-energy conformers, and the MD simulations have to reach them from the bot tom of the potential energy surface. Each MD simulation is terminated when the system meets the uncertainty selection criterion ρ of 0.35kcal × mol−1 ×NA −1/2 (Active learning section). If the MD simulation reaches the time limit, then the structure from the trajectory with the highest uncertainty is selected. DFT reference data (MD simulations section) are then computed for the final conformations and added to the training set for the next iteration of the AL process.”)[ Examiner’ Note : The paper defines the near-equilibrium are initialized at the global energy minimum (i.e., thermodynamic equilibrium).] Regarding Claim 6, Meziere in view of Kulichenko teaches the elements of claim 4 as outlined above: wherein the second set of samples are far-from-equilibrium structures. ( [P. 2, Col. 2] “Most importantly, the proposed approach enables efficient conformational and configurational sampling at low-T conditions, making this approach essential for temperature-sensitive molecules. UDD assists in sampling the chemically relevant subspace of high-energy space, which contains important data such as transition states.” [P. 5, Col. 2] “First, a high energy configurational space (points inside the green oval in Fig. 2d) is more densely sampled in the UDD-AL data set. There are 289 points in this non-equilibrium region in the UDD-AL sampling, compared to 105 points in the 350 K MD-AL. Second, UDD-AL encountered a new conformational path in the top right corner of Fig. 2d that was not accessed by 350 K MD-AL. This region corresponds to rotation of the –OH group around the C–O bond, which is a distinct conformational transition, and a high-energy profile with a barrier of 15 kcal mol−1.”) [ Examiner’s Note : the UDD-AL bias potential identifies and collects high-energy structures (i.e., non-equilibrium region configuration).] Regarding Claim 7, Meziere in view of Kulichenko teaches the elements of claim 6 as outlined above: wherein the far-from-equilibrium structures are substantially far from a state of thermodynamic equilibrium. ( Kulichenko , [P. 2, Col. 2] “First, UDD-AL is used for conformational sampling of a glycine molecule. We find that the bias potential technique generates a diverse data set covering both low- and high-energy regions. As we show in the Results, this contrasts with high-T MD-AL, which tends to skip over low-energy regions. Second, in tests with acetylacetone at low-T conditions, the bias potential is observed to encourage the sampling of the phase space relevant to a proton transfer. Here we find that, in contrast with regular high-T MD, the bias potential technique encourages the reactive transition with very little distortion to the distribution of other degrees of freedom in the system.” [P. 5, Col. 2] “First, a high energy configurational space (points inside the green oval in Fig. 2d) is more densely sampled in the UDD-AL data set. There are 289 points in this non-equilibrium region in the UDD-AL sampling, compared to 105 points in the 350 K MD-AL. Second, UDD-AL encountered a new conformational path in the top right corner of Fig. 2d that was not accessed by 350 K MD-AL. This region corresponds to rotation of the –OH group around the C–O bond, which is a distinct conformational transition, and a high-energy profile with a barrier of 15 kcal mol−1.” [P. 7, Section: Discussion, Col. 2] “The key advantage of UDD-AL over regular high-T sampling is that UDD AL facilitates the sampling of important under-represented chemical data,... This feature can be used for efficient sampling of the con formational and/or configurational space of temperature-sensitive or metastable systems. Our tests also indicate that the bias potential can facilitate sampling of high-energy chemical space, without sacrificing the sampling of low-energy configurations. This means that UDD will produce robust data sets that are applicable to both lower-energy, near-GM data and high-energy chemical space, which usually corresponds to important reactive structural data such as transition states and intermediates. One topic of future research could be the interface of the ML potential trained on UDD-AL data with weighted ensemble methods for obtaining the pathways and rates of chemical reactions.”) [ Examiner’s Note : The paper discusses the structures corresponds to transition states, high-energy conformers, and reactive intermediates. The high-energy state transition and non-equilibrium structure broadly corresponds to far from a state of thermodynamic equilibrium.] Regarding Claim 9, Meziere teaches the elements of claim 1 as outlined above: While Meziere teaches the active learning machine learning interatomic potential (MLIP) training, Meziere does not appear to define the MLIP as a deep learning. However, it would have been obvious in view of Kulichenko. Hereinafter, Meziere in view of Kulichenko teaches the limitation: wherein the actively learned MLIP is a deep learning based MLIP. ( Kulichenko , [P. 8, Section: Methods] “Active learning For the glycine simulations, we used the ANI deep learning model57 to generate ensembles of NN potentials prepared using an eightfold cross-validation split of the data set.” [P .8, Section: NN architecture] “The first atom-centered function was shifted to 0.8 Å from the atomic center. The ANI potential used in this work contained three hidden layers and had the architecture 768:32:16:8:1, where each number describes the number of neurons at each subsequent layer in the network. The ANI potential used in this work contained three hidden layers and had the following architecture: 768:32:16:8:1. Gaussian activation functions were used in the hidden layers and linear activation in the final layer.”) The same motivation that was utilized for combining Meziere in view of Kulichenko as set forth in claim 4 is equally applicable to claim 9. Regarding Claim 15, Meziere teaches the elements of claim 10 as outlined above: While Meziere teaches the active learning machine learning interatomic potential (MLIP) training, Meziere does not appear to define the MLIP as a deep learning. However, it would have been obvious in view of Kulichenko. Hereinafter, Meziere in view of Kulichenko teaches the limitation: wherein the actively learned MLIP is a deep learning based MLIP. ( Kulichenko , [P. 8, Section: Methods] “Active learning For the glycine simulations, we used the ANI deep learning model57 to generate ensembles of NN potentials prepared using an eightfold cross-validation split of the data set.” [P .8, Section: NN architecture] “The first atom-centered function was shifted to 0.8 Å from the atomic center. The ANI potential used in this work contained three hidden layers and had the architecture 768:32:16:8:1, where each number describes the number of neurons at each subsequent layer in the network. The ANI potential used in this work contained three hidden layers and had the following architecture: 768:32:16:8:1. Gaussian activation functions were used in the hidden layers and linear activation in the final layer.”) The same motivation that was utilized for combining Meziere in view of Kulichenko as set forth in claim 4 is equally applicable to claim 15 . 07-21-aia AIA Claim (s) 8 and 14 are rejected under 35 U.S.C. 103 as being unpatentable over Meziere in view of Vandermause et al., (NPL: "On-the-fly active learning of interpretable Bayesian force fields for atomistic rare events." (2020)) . Regarding Claim 8, Meziere teaches the elements of claim 1 as outlined above: While Meziere teaches the active learning machine learning interatomic potential (MLIP) training, Meziere does not appear to explicitly teach: wherein the actively learned MLIP is a Gaussian Process (GP) based MLIP. However, Meziere in view of Vandermause teaches the limitation. Hereinafter, Vandermause, in combination with Meziere, teaches: wherein the actively learned MLIP is a Gaussian Process (GP) based MLIP. ( Vandermause , [Abstract] “We present an adaptive Bayesian inference method for automating the training of interpretable, low-dimensional, and multi-element interatomic force fields using structures drawn on the fly from molecular dynamics simulations. Within an active learning framework, the internal uncertainty of a Gaussian process regression model is used to decide whether to accept the model prediction or to perform a first principles calculation to augment the training set of the model. The method is applied to a range of single- and multi-element systems and shown to achieve a favorable balance of accuracy and computational efficiency, while requiring a minimal amount of ab initio training data.” Further see P. 8, Section: Methods.) Accordingly, it would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, having the combination of Meziere and Vandermause, to incorporate the on-the-fly learning method using GP as taught by Vandermause. One would have been motivated to make such a combination in order to achieve a favorable balance of accuracy and computational efficiency, while requiring a minimal amount of ab initio training data (Vandermause [Abstract]). Regarding Claim 14, Meziere teaches the elements of claim 1 as outlined above: The claim recites substantially similar limitations as corresponding claim 8 and is rejected for similar reasons as claim 8 using similar teachings and rationale . 07-21-aia AIA Claim (s) 18 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Kulichenko in view of Yoo et al., (NPL: "Metadynamics sampling in atomic environment space for collecting training data for machine learning potentials." (2021)) . Regarding Claim 18, Kulichenko teaches the elements of claim 16 as outlined above: While Kulichenko discusses the use of collective variable (CVs) of materials as traditional in Metadynamics potential-energy-surface exploration, where the sampling trajectories are pushed towards less-visited configurational regions. Kulichenko does not appear to explicitly suggest: wherein the biased sampling drives a collective variable of the material. However, it would have been obvious in view of Yoo. Hereinafter, Yoo, in combination with Kulichenko, teaches: wherein the biased sampling drives a collective variable of the material. ( Yoo , [P. 1, Section: Introduction] “MD simulations are conditioned by the Boltzmann statistics, which over-represents low-energy regions and can sample only a few distinct configurations separated by low thermal barriers... We herein propose one such approach based on metadynamics25. The metadynamics defies the Boltzmann distribution by accumulating bias potentials along the collective variables (CVs). Instead of usual implementations that formulate CVs from a set of atomic positions in the real space26, we employ as CVs the coordinates in the abstract atomic-environment space, which is spanned by the atom-centered symmetry-function vector (G)27... By accumulating bias potentials in the G space, the present metadynamics (abbreviated as G-metaD hereafter) drives each atom to evolve towards unvisited points in the G space.” [P. 2, Section: Metadynamics simulation] “The present G-metaD employs the G vector as the CV. The local bias potential (ub) is defined as a function of G, and the summation of atomic local biases constitutes the total bias potential (Ub) applied on the system...”) Accordingly, it would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, having the combination of Kulichenko and Yoo, to incorporate the proposed metadynamics sampling as taught by Yoo. One would have been motivated to make such a combination in order to obtain a small number of metadynamics trajectories that can provide reference structures necessary for training high-fidelity MLPs (Yoo [Abstract]). Regarding Claim 20, Kulichenko teaches the elements of claim 16 as outlined above: While Kulichenko discusses the use of collective variable (CVs) of materials as traditional in Metadynamics potential-energy-surface exploration, where the sampling trajectories are pushed towards less-visited configurational regions. Kulichenko does not appear to explicitly suggest: wherein the biased potential drives a collective variable of the material . However, it would have been obvious in view of Yoo. Hereinafter, Yoo, in combination with Kulichenko, teaches: wherein the biased potential drives a collective variable of the material. ( Yoo , [P. 1, Section: Introduction] “MD simulations are conditioned by the Boltzmann statistics, which over-represents low-energy regions and can sample only a few distinct configurations separated by low thermal barriers... We herein propose one such approach based on metadynamics25. The metadynamics defies the Boltzmann distribution by accumulating bias potentials along the collective variables (CVs). Instead of usual implementations that formulate CVs from a set of atomic positions in the real space26, we employ as CVs the coordinates in the abstract atomic-environment space, which is spanned by the atom-centered symmetry-function vector (G)27... By accumulating bias potentials in the G space, the present metadynamics (abbreviated as G-metaD hereafter) drives each atom to evolve towards unvisited points in the G space.” [P. 2, Section: Metadynamics simulation] “The present G-metaD employs the G vector as the CV. The local bias potential (ub) is defined as a function of G, and the summation of atomic local biases constitutes the total bias potential (Ub) applied on the system...”) Accordingly, it would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, having the combination of Kulichenko and Yoo, to incorporate the proposed metadynamics sampling as taught by Yoo. One would have been motivated to make such a combination in order to obtain a small number of metadynamics trajectories that can provide reference structures necessary for training high-fidelity MLPs (Yoo [Abstract]) . Conclusion 07-96 AIA The prior art made of record and not relied upon is considered pertinent to applicant's disclosure : (Pub. No.: US 20240321401 A1) – “Makoto TAKAMOTO” relates to “Accelerating particle simulations using machine learning and molecular dynamic simulations.” (Pub. No.: US 20240232576 A1) – “Teresa Head-Gordon” relates to “Methods and systems for determining physical properties via machine learning.” (Pub. No.: US 20240153595 A1) – “Nawaf Alampara” relates to “System and method for performing accelerated molecular dynamics computer simulations with uncertainty-aware neural network.” NPL: Jinnouchi, Ryosuke, et al. "On-the-fly active learning of interatomic potentials for large-scale atomistic simulations." (2020). NPL: Merchant, Amil, et al. "Scaling deep learning for materials discovery." (2023). NPL: Smith, Justin S., et al. "Automated discovery of a robust interatomic potential for aluminum." (2021). NPL: Chen, Ming. "Collective variable-based enhanced sampling and machine learning." (2021). Any inquiry concerning this communication or earlier communications from the examiner should be directed to SADIK ALSHAHARI whose telephone number is (703)756-4749. The examiner can normally be reached Monday - Friday, 9 a.m. 6 p.m. ET. Examiner interviews are available via telephone, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice . If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Li Zhen can be reached on (571) 272-3768. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300. Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov . Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000. /S.A.A./Examiner, Art Unit 2121 /Li B. Zhen/Supervisory Patent Examiner, Art Unit 2121 Application/Control Number: 18/377,526 Page 2 Art Unit: 2121 Application/Control Number: 18/377,526 Page 3 Art Unit: 2121 Application/Control Number: 18/377,526 Page 4 Art Unit: 2121 Application/Control Number: 18/377,526 Page 5 Art Unit: 2121 Application/Control Number: 18/377,526 Page 6 Art Unit: 2121 Application/Control Number: 18/377,526 Page 7 Art Unit: 2121 Application/Control Number: 18/377,526 Page 8 Art Unit: 2121 Application/Control Number: 18/377,526 Page 9 Art Unit: 2121 Application/Control Number: 18/377,526 Page 10 Art Unit: 2121 Application/Control Number: 18/377,526 Page 11 Art Unit: 2121 Application/Control Number: 18/377,526 Page 12 Art Unit: 2121 Application/Control Number: 18/377,526 Page 13 Art Unit: 2121 Application/Control Number: 18/377,526 Page 14 Art Unit: 2121 Application/Control Number: 18/377,526 Page 15 Art Unit: 2121 Application/Control Number: 18/377,526 Page 16 Art Unit: 2121 Application/Control Number: 18/377,526 Page 17 Art Unit: 2121 Application/Control Number: 18/377,526 Page 18 Art Unit: 2121 Application/Control Number: 18/377,526 Page 19 Art Unit: 2121 Application/Control Number: 18/377,526 Page 20 Art Unit: 2121 Application/Control Number: 18/377,526 Page 21 Art Unit: 2121 Application/Control Number: 18/377,526 Page 22 Art Unit: 2121
Prosecution Insights