Substitution Structures of Multiple Silicon-Containing Species by Chirped Pulse FTMW Spectroscopy Nathan A. Seifert, Simon Lobsiger, Brooks H. Pate University of Virginia Gamil A. Guirgis, Jason S. Overby College of Charleston James. R. Durig University of Missouri, Kansas City Introduction Goal: Using CP-FTMW spectroscopy to assign a standard molecular structure with minimal sample usage and ~1 day of experimental effort Motivating question: How can we make the process of spectral fitting in MW spectroscopy efficient and easy, so the technique is available to even those not trained in rotational spectroscopy? To solve this, we have developed: Autofit – Automated triples fitter (Front end for SPCAT/SPFIT1) (see additionally Steber et al., TC10 2010; Shipman, et al., RH01 2011) (Steve Shipman of New College of Florida and Ian Finneran of Caltech did most of the programming work, so our eternal gratitude to them!) Anecdote: On a good day, a Pate lab group member can check a given triple for goodness of fit once every two to three minutes using JB95. Every day, Autofit can fit approximately >250 triples in a single second. Fitting a rigid rotor spectra is a purely mechanical task, so why not automate it? The beauty in rotational spectroscopy, after all, is in the analysis. 1. H. M. Pickett, J. Mol. Spectrosc., 148 (1991), 371. Autofit Program Flow Enter guess rotational constants & dipole moments (ab initio predictions) SPCAT Choose three transitions to fit INPUT STAGE No Are transitions sufficiently linearly independent to fit A, B & C? CHOICE STAGE Input window size Δν: For checking triples with expt. freqs within ±Δν Input two column peakpick of target broadband spectrum M cores Split full triples lists into M subarrays Yes Choose N additional transitions for checking fit correctness PROCESS STAGE M SPFIT instances Fit each A/B/C triple with N + 3 transitions Sort triples fits by OMC, collate & output Technical features: • • Written in Python; works on any platform with a Python interpreter • Only requires compatible SPCAT/SPFIT binaries Easily scalable to multiple CPU cores Autofit Benchmarking CPU used: Intel Core i7-3770S (3.1 GHz / 3.9 GHz turbo) • 4 physical cores (8 logical w/ hyperthreading) • 8MB L3 cache, 4 x 256KB L2 cache • $300 on newegg.com (as of May 2013) Some performance observations: • Invariant to RAM availability • Strong dependence on CPU cache & cache bandwidth Optimal performance would likely be found using a high performance CPU with a good cache, like the Intel Xeon or AMD Opteron series Motivating Examples 5 molecules chosen – in collaboration with Gamil Guirgis of College of Charleston: (results also available for 1-isocyanatosilyl-cyclopropane) CH3SiHFNCO 1-cyanosilyl-cyclopropane 1,1,3,3-tetra(fluoro/hydro)-1,3-disilacyclopentane 1-isocyanato-1-silacyclohexane In all cases, a single Autofit run detected the parent species as well as all isotopologues detectable in natural abundance Autofit philosophy: • Initial guess rotational constants/dipoles from MP2/6-311++g(d,p) structures • Chose the 3 fit transitions from typical strong features that can fit A, B & C well • 5-7 additional strong transitions for checking fit, typically between 7-12 GHz • Peakpick cutoff made to include all lines ca. >4:1 signal to noise ratio • ~500 MHz window for all autofit runs • CP-FTMW spectra for all systems taken with the new upgrades at 6.5-18 GHz (Similar to 2-8 GHz improvements seen in the slides of Cristobal’s talk, TH10). • Spectra range from 120k to 500k averages (one afternoon max. of averaging) CH3SiHFNCO – Autofit Results 120,000 avg. spectrum (1 hr) Top 16 fits from Autofit output: A (MHz) B (MHz) C (MHz) OMC (kHz) Assignment 6427.20 1585.675 1265.95 27.7 No match 6150.49 1516.959 1291.71 28.2 13CH 3 6298.88 1518.946 1299.04 31.6 N13CO 6277.93 1527.439 1305.77 33.8 30Si 6217.09 1534.432 1311.16 35.2 NS 6288.59 1536.618 1330.514 36.3 No match 6294.44 1554.153 1274.29 37.7 No match 6257.27 1534.867 1310.90 37.8 NS 6300.71 1534.827 1310.89 38.8 NS 6295.23 1529.44 1308.519 38.8 29Si 6427.03 1585.684 1265.95 42.7 No match 6248.73 1517.939 1300.05 44.5 N13CO 6294.93 1535.399 1310.27 45.0 NS 6441.8 1600.815 1241.58 45.2 No match 6294.33 1533.594 1308.77 45.5 15N 6294.77 1533.987 1308.88 45.9 15N (rough) • 6.14 million triples (7 hrs, 8 cores) Fit transitions: • One a-type Ka = 0 (B+C) • Two b-type transitions (A; asymmetry) Note: • OMCs unusually high due to A/E splitting and hyperfine CH3SiHFNCO – Final Results • A/E states + quad fit simultaneously with XIAM • Structural trends consistent with CH3SiF2NCO results (2012 MH08) • V3 decreases with additional halogenation; r[H3C-Si] decreases by 0.3 Å • Difluoro V3: 446(5) cm-1; CH3SiF2Cl V3: 463(3) cm-1 • Similar bend in –NCO: <[Si-N-C] = 159.8(9)° (F2); 152.6(18)° (HF) Parent Species Results MP26- Expt. 311++g(d,p) MP2/6-311++g(d,p) structure overlaying rs coordinates Isotopologue Results (Other parameters held fixed) 29Si 30Si 13CH 3 N13CO 15N A (MHz) 6290.281(46) 6279.457(60) 6159.135(51) 6299.287(50) 6295.336(28) B (MHz) 1530.6263(68) 1526.2869(11) 1516.9197(43) 1518.6229(42) 1532.7715(44) C (MHz) 1307.7311(61) 1305.0436(10) 1291.7792(37) 1298.4160(35) 1308.5262(38) Nlines 64 57 22 24 20 σfit (kHz) 20.3 20.8 17.9 21.4 19.6 A (MHz) 6086.35 6301.415(45) B (MHz) 1524.42 1535.078(39) C (MHz) 1304.04 1310.485(39) ΔJ (kHz) 0.606 0.742(33) ΔJK (kHz) 101.01 41.50(14) ΔK (kHz) -115.14 -25.39(10) δJ (kHz) 0.034 0.067(13) δK (kHz) 69.19 25.6(19) 3/2 χaa (MHz) 3.27 2.655(11) ¼(χbb-χcc) (MHz) 46.36 23(18) V3 (cm-1) 436 480(19) F (GHz) [158] 155.7(66) θ (deg) 57.6 58.6(10) Nlines -- 150 σfit (kHz) -- 18.9 1-isocyanato-1-silacyclohexane – Autofit Results 320,000 avg. spectrum (2.8 hrs) Axial-, selected Autofit results (with sorted rankings) Rank A (MHz) B (MHz) C (MHz) OMC (kHz) Assignment 1 1942.32 951.832 923.025 15.2 15N 2 1953.55 954.585 926.465 20.8 NS 4 1936.75 954.847 924.089 21.2 29Si 6 1952.15 945.318 917.625 21.4 13C-2 8 1935.21 949.289 921.503 22.2 13C-4 9 1921.58 955.012 921.342 23.1 30Si 12 1934.06 948.689 922.151 24.8 13C-3 15 1936.25 946.925 916.908 26.5 13CN Axial ΔE = 0 cm-1 • 6.13 million triples (10 hrs, 4 cores) Fit transitions: • Three a-types; Ka = 0, Ka = 1, Ka = 2 Little A dependence, so error on A is high Equa-, selected Autofit results (with sorted rankings) Rank A (MHz) B (MHz) C (MHz) OMC (kHz) 1 3441.4 690.101 608.989 7.6 NS Assignment 2 3438.2 689.609 608.981 11.9 30Si 4 3432.5 689.845 608.972 17.6 29Si 5 3450.2 682.162 602.817 18.3 13CN 7 3427.6 682.189 603.121 19.0 13C-4 10 3370.2 689.485 606.833 20.9 13C-2 12 3412.9 685.212 603.925 21.6 13C-3 13 3410.3 687.168 606.641 23.4 15N • 5.03 million triples (~9 hrs, 4 cores) Fit transitions: • Three a-types; Ka = 0, Ka = 1, Ka = 2 Equatorial ΔE = 414 cm-1 1-isocyanato-1-silacyclohexane – Results Open issue: MP2 (as well as DFT) Severely underestimates NCO bending angle (Xaa discrepencies are a real tell!) MP2 6-311++g(d,p) Expt. MP2 6-311++g(d,p) Expt. A (MHz) 2261.9804 1949.06(44) A (MHz) 3530.4152 3440.37(26) B (MHz) 808.24933 954.8421(20) B (MHz) 686.04800 690.1025(18) C (MHz) 779.80522 926.7693(17) C (MHz) 606.12607 608.9931(18) ΔJ (kHz) -- 1.6593(94) ΔJ (kHz) -- 0.0910(61) ΔJK (kHz) -- -2.767(45) ΔJK (kHz) -- 8.076(52) ΔK (kHz) -- 1597(155) ΔK (kHz) -- [0] δJ (kHz) -- -0.328(82) δJ (kHz) -- [0] δK (kHz) -- [0] δK (kHz) -- 2.92(80) 3/2 χaa (MHz) 2.10 0.941(33) 3/2 χaa (MHz) 3.03 2.461(80) ¼(χbb-χcc) (MHz) -0.153 0.276(13) ¼(χbb-χcc) (MHz) [0] [0] Nlines -- 66 Nlines -- 123 σfit (kHz) -- 14.6 σfit (kHz) -- 16.4 Conclusions Autofit is a fast and efficient way to quickly assign broadband spectra for both parent species and isotopologues for the purpose of molecular identification or structural determination. Orders of magnitude faster than routine, manual spectra fitting Helps enable structure determination via CP-FTMW become a routine activity With new CP-FTMW spectra approaching line densities of over >1 MHz-1, visual pattern recognition for weakly abundant species is nearly impossible • • • The Future of Autofit: • • Automated optimization for choosing a frequency window size with respect to typical ab initio error on A/B/C Automated choosing of 3 fit transitions based on optimal linear independence to fit A/B/C • A graphical interface where autofit results are integrated into a JB95/PGOPHER/AABS-like interactive fitting program On the technical side: CPU scaling via the cloud is cheap and perhaps even trivial! • A quick back of the envelope calculation for large scaling on Amazon EC2 clusters: • Assume 50 Hz/core; 20 EC2 logical cores/instance for $0.580/hr • Maximum of 20 instances at once, so 400 logical cores for $11.60/hr • Effective compute speed of 20 kHz, so 500 seconds to fit 107 triples $1.60 per autofit run! Acknowledgements Shameless plug – Autofit is freely available at http://tinyurl.com/autofitcpftmw • git repository: git clone git://github.com/pategroup/bband_scripts.git • Works in Windows, but even easier to setup in x86/x86-64 Linux! Pate Group Brooks Pate Cristobal Perez Simon Lobsiger Luca Evangelisti Brent Harris Amanda Steber Nathan Seifert Thanks for your time! Thanks to the NSF for funding: MRI-R2, Award CHE-0960074 CH3SiHFNCO – Final Results How does it compare to other species in the series? Species V3 (cm-1) R[H3C-Si] (Å) Citation (CH3)2SiH2 576.4 1.867(2) L. Pierce, J. Chem. Phys. 34 (1961) 498. CH3SiH2Cl 644 1.849(5) R. H. Schwendeman, G.D. Jacobs, J. Chem. Phys. 36 (1962) 1251. CH3SiHCl2 731 -- J. R. Durig, C. W. Hawley, J. Chem. Phys. 59 (1973) 1. CH3SiCl3 875 1.848 M. A. Qtaitat, et al., Spectrochim. Acta A, 50 (1994) 621. CH3SiH2F 545(1) 1.849(5) L.C. Krisher, L. Pierce, J. Chem. Phys. 32 (1960) 1619. CH3SiHF2 439(10) 1.840(1) L.C. Krisher, L. Pierce, J. Chem. Phys. 32 (1960) 1619. CH3SiF3 413.994(9) 1.812(14) J.R. Durig, Y.S. Li, C.C. Tong, J. Mol. Struct. 14 (1972) 255. CH3SiF2NCO 446(5) 1.814(5) G.A. Guirgis, et al. J. Phys. Chem. A. 116 (2012) 7822. CH3SiF2Cl 468(3) 1.814(1) N. A. Seifert, et al. J. Mol. Struct. 1023 (2012) 222. CH3SiHFNCO 480(19) 1.843(7) This work