Broadband Microwave Spectroscopy and Automated Analysis of 12 Conformers of 1-Hexanal Nathan A. Seifert, Cristobal Perez, Daniel P. Zaleski, Justin L. Neill, Amanda L. Steber, Richard D. Suenram, Brooks H. Pate University of Virginia Steven Shipman, Ian Finneran New College of Florida Alberto Lesarri Universidad de Valladolid Motivation Broadband rotational spectroscopy enables detection of all polar emitters in a given sample at once, within sensitivity limits But what good is all this data if its too dense to interpret traditionally? How we can enable molecular discovery in an efficient fashion? The answer: computer-aided assignment The AUTOFIT approach Instead of introducing complexity, lets fit spectra with our computers as simply and rationally as we can: Given an input prediction for a target rotational spectrum (e.g. from ab initio geometry and dipole moments) Choose 3 strong, linearly independent transitions (the triple) to minimize our uncertainties on A, B and C Search for these transitions in our experiment with search windows sized correspondingly to the expected ab initio uncertainty Forward predict a candidate set of other strong transitions to corroborate / invalidate the triple-fit set of rotational constants and iterate through this approach through all possible triplets of experimental lines in our search windows Introduction to Hexanal Stark effect results for 6 conformers previously reported See: Alverez-Valtierra et al., 2008 WF12 [OSU ISMS] Multiple previous studies conformationally flexible, long-chain aliphatics: 1-heptanal (2006): 13 conformers detected [Balle-Flygare] 1 2-hexanol (2008): 14 conformers [B-F + CP-FTMW] 2 1-pentene through 1-octene (2000/2001): 7-15 conformers [B-F] 3 1-hexanal is an ideal case study for AUTOFIT : 1.Fisher, J. M.; Xu, L.-H.; Suenram, R..; Pate, B.; Douglass, K. J. Mol. Struct. 2006, 795, 143 Tubergen, M. J.; Conrad, A. R.; Chavez III, R. E.; Hwang, I.; Suenram, R. D.; Pajski, J. J.; Pate, B. H. J. Mol. Spectrosc. 2008, 251, 330 (a) Fraser, G. T.; Xu, L.-H.; Suenram, R. D.; Lugez, C. L. J. Chem. Phys. 2000, 112, 62096217. (b) Fraser, G. T.; Suenram, R. D.; Lugez, C. L. J. Phys. Chem. A 2000, 104, 11411146. (c) Fraser, G. T.; Suenram, R. D.; Lugez, C. L. J. Phys. Chem. A 2001, 105, 98599864. Rich, intense broadband spectrum Geometric constraints on conformations limit candidate structure search On a side note: 1-Hexanal 123-O123-O = a, g Minima when 1 / 2 / 3 are anti ( a ) or gauche ( g ); 0 is either gauche ( g, +/- 120 ) or eclipsed ( e, 0 , lowest energy) With 0 = e, there are 14 possible conformers (considering symmetry) 8 with energies < 500 cm -1 of global minimum Some 0 = g rotamers are also < 500 cm -1 Spectra previously taken at: 6-18 GHz (~ avg) GHz (~ avg) New spectrum with averages Experimental 0.2% hexanal in Neon, using ca. 1 atm backing pressure Using CP-FTMW spectroscopy, 34 GHz of bandwidth achieved over three measurements GHz spectrometer Dynamic range comparable to 6-18 GHz measurement GHz more favorable for Boltzmann weighting More directional transmission compared to GHz, allowing 3 nozzles instead of 2. Before CutAfter Cut GHz: 1669 lines 3:1, 518 after cut (69% removed) 6-18 GHz: 1432 lines 3:1, 655 after cut (54%) GHz: 2220 lines 3:1, 823 after cut (63%) 3205 total hexanal lines from 12 conformers + 26 isotopologues assigned from 6-40 GHz, plus ~100 spurious and known contaminant lines At the very least: Only 130 lines at >10:1 in cut! Detected conformers of 1-hexanal B3LYP-D3/aug-cc-pVTZ optimized structures All assignments made with AUTOFIT, with unambiguous results compared to publication-quality fits in terms of agreement. B i (i = A, B, C)mean B i / kHz (B i ) / KHz A B330 C-1683 mean fit RMS (no distortion) 35 kHz mean fit RMS (with distortion) 10 kHz Using just AUTOFIT : Heavy atom isotopologues detected in natural abundance for top 4 conformers Autofit vs Expt Constant Residuals: So AUTOFIT works. But how can we optimize our search speed? Always the go-to answer: Lets appeal to higher levels of theory for better guesses! Thermochemistry: Can we accurately determine our energy cutoff for minima on a PES and compare it to the spectral sensitivity limits? Two approaches: Rotational constants: Can our guesses for rotational constants improve beyond the typical MP2 / M06-2X standard? 3:1 S/N limit at ln(I/I SPCAT ) ~ -7.0 (or E ~ 650 cm -1 ) Slope - (-1/k B T) T = 135(11) K (D3/pVTZ) (using Neon) Cooling in the jet biasing against detection of more energetic rotamers? Thermochemistry (Note: y-axis is least-squares fit logarithm of population difference, using B3LYP-D3 determined dipole moments) Observed Ordering E (cm -1 ) ConformerM06-2XRI-MP2D3D3/pVTZ 1 (aaa-e) (aag-e) (aga-e) (gaa-e) (g + ag - -e) (g + ag + -e) (aag + -g - ) (ag + g + -e) (aaa-g) (gga-e) (ag + g + -g - ) (g - ag - -g + ) Potential isomers not observed: E (cm -1 ) E (cm -1 ) / (est max 135 K) ConformerM06-2XD3/pVTZ ggg-e (4:1) g + g + g + -g (5:1) ag + g - -e (5:1) g + aa-g + (gauche rotamer of #4) (4.5:1) g + aa-g - (gauche rotamer of #4) (4:1) aga-g (gauche rotamer of #3) (4.5:1) Remaining unobserved isomers are weak at best (predicted S:Ns correspond to S:N of strongest transition) No positive matches using Autofit (or by eye) Thermochemistry An aside about DFT Last year in PCCP: [ Phys. Chem. Chem. Phys., 15, (2013)] Significant improvements in rotational constants by moving to -D3 corrections Take home point: Better than MP2 accuracy in B3LYP computational timescales! How does this fare for AUTOFIT ? How about hexanal? mean [] A, %B, %C, %combined -D3/aVTZ-0.49 [0.87]0.81 [0.77]0.58 [0.38]0.30 [0.90] D [1.17]1.11 [0.94]0.91 [0.55] 0.66 [1.05] M06-2X0.45 [2.24]-2.06 [2.25]-1.82 [1.71] [2.37] MP21.78 [2.12]-1.69 [1.76]-1.46 [1.45] [2.40] B3LYP0.19 [2.68]1.85 [2.68]1.66 [1.48] 1.23 [2.47] (Basis set is g(d,p) unless otherwise noted) How does this improve our run times? Estimated AUTOFIT run time (min) [rel speedup wrt. B3LYP/ g(d,p)] Method A-type fitB-type fitA/B-hybrid fit B3LYP M06-2X 220 [-17%]3190 [-21%]714 [-8%] MP2101 [82%]560 [352%]420 [56%] B3LYP-D37 [2516%]83 [2953%]75 [772%] B3LYP-D3/aVTZ5 [3850%]49 [5009%]58 [1029%] Modeled typical search windows for A/B type fits with 3 certainty Runtime (window size) 3 2x better constant prediction -> 8x faster Autofit runtime Using D3/aVTZ predictions, all 12 hexanal conformers + 26 isotopologues searches were run (all with successful detections) in < 7 hours. Conclusion Using CP-FTMW spectroscopy in conjunction with desktop-level Ab initio calculations and AUTOFIT, 12 conformers of hexanal were detected 4 heavy atom ( 13 C O) Kraitchman structures found automatically Sensitivity limits for additional conformers set using modern DFT calculations for accurate thermochemistry Grimmes D3 corrections give significantly improved results without additional computational time w.r.t. traditional DFT AUTOFIT, even in its early stage, make assignments in broadband spectra fast and simple AUTOFIT 2.0 will be more flexible, faster and more powerful Distortion, simple hyperfine problems should be easy to implement Automated high-barrier limit C 3v implementation with XIAM? Questions? Thank you for your time! This work was funded by the National Science Foundations Major Research Instrumentation program, award # CHE AUTOFIT Finds spectra by matching a candidate set of predicted transitions to every possible set of experimentally-observed transitions (the triple) within a given search window 3 fit transitions to establish initial fit of A/B/C (or B/C for typical near-prolate a -type spectra) Arbitrary set of check transitions found using forward predictions set by the initial fit of 3. Can automatically refine selected results with users choice of distortion constants Typical triple sets range from ~ in a broadband spectrum (e.g. ~ 10 GHz) Interfaces with CALPGM (SPCAT/SPFIT) for predictions/fitting With input ab initio geometry, can automatically scale and search for isotopologues Search speed 400 Hz on a typical modern desktop (8 threads) Brute force, but more stable than a genetic algorithm and is embarrassingly parallel (easy to expand to arbitrary number of threads) (More discussion available in 2013 RC12, TC , RH ) AUTOFIT timing model Simulate AUTOFIT error windows (and consequently runtime) using model a, b, and a/b-type near-prolate spectra a -type fit: two K a = 0, one K a = 1 transition Little dependence on A; B+C / B-C dominant b -type fit: two K a = 0, one K a = 1 transition (e.g., 2x J 0J -> J 1J, 1x J 1J -> J 0J ) Strong dependence on A, B and C 3 error windows determined by rotational constant prediction errors for 1-hexanal data set for a variety of ab initio methodologies. Kraitchman results (B3LYP-D3/aug-cc-pVTZ structures (and energies) overlaying Kraitchman r s coordinates [green spheres]) Sufficient intensity to resolve all 13 C isotopologues for four lowest energy conformers, and 18 O on the two lowest AUTOFIT Accuracy B i (i = A, B, C)mean B i / kHz (B i ) / KHz A B330 C-1683 Large deviations in A data due to conformers with only a -type transitions in search C deviations due to b -type searches mean RMS error of fits (without distortion) 35 kHz mean RMS error of fits (automatically fitting distortion) 10 kHz (Data set consists of all 12 hexanal conformers, plus detectable isotopologues. Initial rotational constant guesses derived from M06-2X/ g(d,p) calculations)