Image caption: An approach combining terahertz spectroscopy, X-ray diffraction, and solid-state density functional theory was utilized to accurately measure the elasticities of poly-l-proline helices by probing their spring-like vibrational motions. In their communication (DOI: 10.1002/anie.201602268), T. M. Korter and co-workers reveal that poly-l-proline is less rigid than commonly expected, and that the all-cis and all-trans helical forms exhibit significantly different Young’s moduli.
Accepted in the Journal of Physical Chemistry A, with my fingers crossed for pulling off the rare double-header in an upcoming print edition of the journal (having missed it by three intermediate articles with the Cs2B12H12 and HMX papers back in 2006 (you’d keep track, too). A fortuitous overlap of scheduled defense dates between P. Hakey, Ph.D. and M. Hudson, A.B.D.). A brief summary of interesting points from this study is provided below, including what I think is a useful point about how to most easily interpret AND represent solid-state vibrational spectra for publications.
1. AS USUAL, YOU CANNOT USE GAS-PHASE CALCULATIONS TO ASSIGN SOLID-STATE TERAHERTZ SPECTRA. It will take a phenomenal piece of data and one helluvan interpretation to convince me otherwise. As a more subtle point (for those attempting an even worse job of vibrational mode assignment), if the molecule exists in its protonated form in the solid-state, do not use the neutral form for your gas-phase calculation (this is a point that came up as part of an MDMA re-assignment published (and posted here) previously).
2. It is very difficult to find what I would consider to be “complete data sets” for molecules and solids being studied by spectroscopic and computational methods. For many molecular solids, the influences of thermal motion are not important to providing a proper vibrational analysis by solid-state density functional theory methods. Heating a crystal may make spectral lines broader, but phase changes and unusual spectral features do not often result when heating a sample from cryogenic (say, liquid nitrogen) to room temperature. Yes, there are thousands of cases where this is not true, but several fold more cases where it is. We are fortunate to live in a temperature regime where characterization is reasonably straightforward and yet we can modify a system to observe its subtle changes under standard laboratory conditions. The THz spectrum of S-(+)-Ketamine Hydrochloride gets a bit cleaner upon cooling, which makes the assignment easier. As the ultimate goal is to be able to characterize these systems in a person’s pocket instead of their liquid nitrogen thermos, the limited observed change to the spectrum upon cooling is important to note.
3. Crystal06 vs. DMol3 – This paper contains what is hoped to be a level, pragmatic discussion about the strengths and weaknesses of computational tools available to terahertz spectroscopists for use in their efforts to assign spectra. This type of discussion is, as a computational chemist using tools and not developing tools, a touchy subject to present on not because of the finger-pointing of limitations with software, but because the Crystal06 team and Accelrys (through Delley’s initial DMol3 code) clearly are doing things that the vast majority of their users (myself included) could in no way do by themselves. The analysis for the theory-minded terahertz spectroscopist is presented comparing two metrics – speed and functionality (specifically, infra-red intensity prediction). What is observed as the baseline is that both DMol3 and Crystal06 make available density functionals and basis sets that, when used at high levels of theory and rigorous convergence criteria, produce simulated terahertz spectra with vibrational mode energies that are in good (if not very good) agreement with each other. For the terahertz spectroscopist, Crystal06 provides as output (although this is system size- and basis set size-dependent) rigorous infrared intensity predictions for vibrational modes, inseparable from mode energy as “the most important” pieces of information for mode assignments. While DMol3 does not produce infrared intensities (the many previous terahertz papers I’ve worked on employed difference-dipole calculations that are, at best, a guesstimate), DMol3 produces very good mode energy predictions in 1/6th to (I’ve seen it happen) 1/10th the time of a comparable Crystal06 calculation. This is the reason DMol3 has been the go-to program for all of the neutron scattering spectroscopy papers cited on this blog (where intensity is determined by normal mode eigenvectors, which are provided by both (and any self-respecting quantum chemical code) programs).
Now, it should be noted that this difference in functionality has NOTHING to do with formalism. Both codes are excellent for what they are intended to do. To the general assignment-minded spectroscopist (the target audience of the Discussion in the paper), any major problem with Crystal06 likely originates with the time to run calculations (and, quite frankly, the time it takes to run a calculation is the worst possible reason for not running a calculation if you need that data. Don’t blame the theory, blame the deadline). In my past exchanges with George Fitzgerald of Accelrys, the issue of DMol3 infrared intensities came up as a feature request that would greatly improve the (this) user experience and Dr. Fitzgerald is very interested (of course) in making a great code that much better. Neither code will be disappearing from my toolbox anytime soon.
4. The Periodicity Of The Molecular Solid Doesn’t Care What The Space Group Is – One of the more significant problems facing the assignment-minded spectroscopist is the physical description of molecular motion in a vibrational mode. In the simplest motions involving the most weakly interacting molecules, translational and rotational motions are often quite easy to pick out and state as such. When the molecules are very weakly interacting, often the intramolecular vibrational modes are easy to identify as well, as they are largely unchanged from their gas-phase descriptions. In ionic solids or strongly hydrogen-bonded systems, it is often much harder to separate out individual molecular motions from “group modes” involving the in- and out-of-phase motions of multiple molecules. In the unit cells of molecular solids, it can be the case that these group modes appear, by inspection, to be extremely complicated, sometimes too involved to easily describe in the confines of a table in a journal article.
S-(+)-Ketamine Hydrochloride is one such example where a great simplification in vibrational mode description comes from thinking, well, “outside the box.” The image below shows two cells and the surrounding molecules of S-(+)-Ketamine Hydrochloride. As it is difficult to see why the mode descriptions are complex from just an image, assume that I am right in this statement of complexity. Part of this complexity comes from the fact that the two molecules in the unit cell are not strongly interacting, instead packed together by van der Waals and dispersion forces more than anything else. The key to a greatly simplified assignment comes from the realization that the most polar fragments of these molecules are aligned on the edges of the unit cell.
An alternate view of molecular vibrational motion comes from considering not the contents of the defined unit cell but the hydrogen-bonding and ionic bonding arrangement that exists between pairs of molecules between unit cells. The colorized image below shows two distinct chains (red and blue) that, when the predicted vibrational modes are animated, become trivial to characterize as the relative motions of a hydrogen/ionic-bonded chain. Rotational motions appear as spinning motions of the chains, translational motions as either chain sliding motions or chain breathing modes. It appears as a larger macromolecule undergoing very “molecular” vibrations. In optical vibrational spectroscopy, selection rules and the unit cell arrangement do not produce in- and out-of-phase motions of the red and blue chains, as only one “chain” exists in the periodicity of the unit cell. In neutron scattering spectroscopy, these relative motions between red and blue would appear in the phonon region. This same discussion was had, in part, in a previous post on the solid-state terahertz assignment of ephedrine (with a nicer picture).
So, look at the cell contents, then see if there’s more structure than crystal packing would indicate. It greatly simplifies the assignment (which, in turn. greatly simplifies the reader’s digestion of the vibrational motions).
Patrick M. Hakey, Damian G. Allis, Matthew R. Hudson, Wayne Ouellette, and Timothy M. Korter
Department of Chemistry, Syracuse University, Syracuse, New York 13244-4100
Abstract: The terahertz (THz) spectrum of (S)-(+)-ketamine hydrochloride has been investigated from 10 to 100 cm-1 (0.3-3.0 THz) at both liquid-nitrogen (78 K) and room (294 K) temperatures. Complete solid-state density functional theory structural analyses and normal-mode analyses are performed using a single hybrid density functional (B3LYP) and three generalized gradient approximation density functionals (BLYP, PBE, PW91). An assignment of the eight features present in the well-resolved cryogenic spectrum is provided based upon solid-state predictions at a PW91/6-31G(d,p) level of theory. The simulations predict that a total of 13 infrared- active vibrational modes contribute to the THz spectrum with 26.4% of the spectral intensity originating from external lattice vibrations.
“I’m high on life… and PCP.” – Mitch Hedberg
In press, in the Journal of Physical Chemistry A. If the current rosters of pending manuscripts and calculations are any indication, this PCP paper will mark the near end of my use of DMol3 for the prediction (and experimental assignment) of terahertz (THz) spectra (that said, it is still an excellent tool for neutron scattering spectroscopy and is part of several upcoming papers).
While the DMol3 vibrational energy (frequency) predictions are generally in good agreement with experiment (among several density functionals, including the BLYP, BOP,VWN-BP, and BP generalized gradient approximation density functionals), the use of the difference-dipole method for the calculation of infrared intensities has shown itself to be of questionable applicability when the systems being simulated are charged (either molecular salts (such as PCP.HCl) or zwitterions (such as the many amino acid crystal structures)). The previously posted ephedrine paper (in ChemPhysChem) is most interesting from a methodological perspective for the phenomenal agreement in both mode energies AND predicted intensities obtained using Crystal06, another solid-state density functional theory program (that has implemented hybrid density functionals, Gaussian-type basis sets, cell parameter optimization and, of course, a more theoretically sound prediction of infrared intensities by way of Born charges). The Crystal06 calculations take, on average, an order of magnitude longer to run than the comparable DMol3 calculations, but the slight additional gain in accuracy for good density functionals, the much greater uniformity of mode energy predictions across multiple density functionals (when multiple density functionals are tested), and the proper calculation of infrared intensities all lead to Crystal06 being the new standard for THz simulations.
After a discussion with a crystallographer about what theoreticians trust and what they don’t in a diffraction experiment, the topic of interatomic separation agreement between theory and experiment came up in the PCP.HCl analysis performed here (wasn’t Wayne). As the position of hydrogen atoms in an X-ray diffraction experiment are categorically one of those pieces of information solid-state theoreticians do NOT trust when presented with a cif file, I reproduce a snippet from the paper considering this difference below (and, generally, one will not find comparisons of crystallographically-determined hydrogen positions and calculated hydrogen positions in any of the THz or inelastic neutron scattering spectroscopy papers found on this blog).
The average calculated distance between the proton and the Cl– ion is 2.0148 Angstroms, an underestimation of nearly 0.13 Angstroms when compared to the experimental data. This deviation is likely strongly tied to the uncertainly in the proton position as determined by the X-ray diffraction experiment and is, therefore, not used as a proper metric of agreement between theory and experiment. The distance from the nitrogen atom to the Cl– ion has been determined to be an average of 3.0795 Angstroms, which is within 0.002 Angstroms of the experimentally determined bond length. This proper comparison of heavy atom positions between theory and experiment indicates that this interatomic separation has been very well predicted by the calculations.
Patrick M. Hakey, Matthew R. Hudson, Damian G. Allis, Wayne Ouellette, and Timothy M. Korter
The terahertz (THz) spectrum of phencyclidine hydrochloride from 7.0 – 100.0 cm-1 has been measured at cryogenic (78 K) temperature. The complete structural analysis and vibrational assignment of the compound have been performed employing solid-state density functional theory utilizing eight generalized gradient approximation density functionals and both solid-state and isolated-molecule methods. The structural results and the simulated spectra display the substantial improvement obtained by using solid-state simulations to accurately assign and interpret solid-state THz spectra. A complete assignment of the spectral features in the measured THz spectrum has been completed at a VWN-BP/DNP level of theory, with the VWN-BP density functional providing the best-fit solid-state simulation of the experimentally observed spectrum. The cryogenic THz spectrum contains eight spectral features that, at the VWN-BP/DNP level, consist of fifteen infrared-active vibrational modes. Of the calculated modes, external crystal vibrations are predicted to account for 42% of the total spectral intensity.
Accepted in Chemical Physics Letters. A solid-state density functional theory (DFT) follow-up to the solution-phase 2-pyridone (2PD) study published by Motley and Korter previously. Much of the work-up for this paper was straightforward, run-of-the-mill calculation and correlation (on the theory side, anyway). The most difficult part of the analysis was the identification of the easiest way to present the normal mode analysis of the 2PD crystal cell.
In terahertz (THz) spectroscopy, one observes the lowest-frequency vibrational motions of molecules (if the molecule has low-lying vibrational modes, of course). In the solid-state (such as molecular crystals), one observes both low-lying molecular vibrational motions (if they exist) and the relative motions between molecules in the unit cell. The boilerplate separation of internal (intramolecular) and external (between-molecule) modes is performed (and presented) as follows:
A crystal unit cell containing M molecules with N atoms contains 3N-6M internal modes (those modes associated with intramolecular motions), 6M-3 external modes (those modes associated with relative motions between the M molecules, such as rotations and translations), and three acoustic modes.
Some molecules simply do not absorb in the THz region, so all assignments are for external motions (and one simply identifies molecules sliding along axes or spinning around their centers of mass in their lattice site). Some molecules are very strongly bound to neighboring molecules in their lattice sites, which results in significant changes to the mode energies of low-lying vibrational modes (these are far more complicated systems to perform assignments of and a few of these cases are being prepared for future publications). Some molecules are strongly bound in very localized ways in their crystal cell to neighboring molecules and are very weakly bound to other neighbors in other ways. In 2PD, chains of molecules are strongly bound through hydrogen bonding along the crystal c-axis (see the figure below) and only weakly interacting between chains. In the figure below, the blue and red chains are strongly coupled in-chain (hydrogen-bonding) and only weakly coupled (dispersion and van der Waals forces) between chains.
The assignment of the 2PD solid is simplified by two important facts. First, the two chains (red and blue) are related by symmetry (the unit cell contains two anti-parallel 2PD chains). Second, the chains are very weakly interacting.
What point 1 means is that the chains, if in isolation, would undergo the same vibrational motions at the same energies (as if the chains were simply chiral molecules).
What point 2 means is that these chains are, because they interact very weakly, approaching a limit where there can, in fact, be considered isolated chains so that the unit cell will contain vibrational motions that involve the two chains undergoing the same motion in-phase with respect to reach other (in-phase here meaning that, for instance, both of your lungs are expanding at the same time) and out-of-phase with respect each other (the hypothetical case where the left and right lobes are out-of-sync with one another).
For instance, if both chains are sliding along the crystal c-axis in a vibrational mode, that makes the mode the in-phase acoustic translation in c (because the whole cell is sliding in one direction). If the two chains are sliding in opposite directions with respect to each other, this makes the mode the optical translation in c (the center of mass of the cell stays put while the chains undergo out-of-phase motions).
This simplification for the 2PD assignment (and other solid-state molecular chains) turned out to be the mode assignment based on the treatment of not the in-cell contents of atoms and molecular fragments (if we kept ourselves to only viewing what is happening in the cell, for instance), but instead the relative motions of the chains, which requires ever-so-slightly thinking outside of the box.
Tanieka L. Motley, Damian G. Allis, and Timothy M. Korter*
Department of Chemistry, Syracuse University, Syracuse, NY 13244-4100
Crystalline 2-pyridone has been investigated using terahertz vibrational spectroscopy in the range of 10 to 90 cm-1 (0.3 to 2.7 THz). Solid-state density functional theory (B3LYP, BP, and PW91 with the 6-311G(d,p) basis set) was used to simulate and assign both observed terahertz spectral features and a previously published far-infrared spectrum up to 400 cm-1. The PW91 functional was found to provide the best combination of crystal structure and vibrational frequency reproduction. Observed spectral features below 150 cm-1 are assigned to intermolecular movements of the 2-pyridone chains within the unit cell. The use of independent intramolecular and intermolecular frequency scalars is proposed.
Accepted in ChemPhysChem. Two important points. First, as shown in the crystal cell figure below, the low-frequency study of the ephedrine molecular solid is one that is best considered in the context of two infinite chains (red and blue) that are strongly interacting along the chain and very weakly interacting between chains. The key point is the realization that the ephedrine molecular solid is not best considered as four molecules packed into a crystal cell. The original round of mode assignments, based only in crystal cell contents, was a very complicated list of relative motions and nearly irreconcilable collisions of in- and out-of-phase motions. Thinking outside-the-unit-cell and realizing that the mode motions could be described far more easily (and logically) as chains instead of packed molecules made the final assignment and analysis of the terahertz spectrum very straightforward. The lesson is to take a good look at your molecular solid before attempting to describe the motions and consider divide-and-conquer approaches if you see correlations.
The second reason I am specifically pleased with this paper is that it is the first real terahertz study using Crystal06 that employs multiple generalized gradient approximation density functionals (BP, PBE, PW91) and basis sets (6-31G(d,p) and 6-311G(d,p)) and shows that these multiple levels of theory provide very similar results. That has, generally, NOT been the case with the many previous DMol3 studies that required difference-dipole intensity calculations instead of the use of more rigorous Wannier function-based intensities possible within the Crystal06 code.
Patrick M. Hakey, Damian G. Allis, Matthew R. Hudson, Wayne Ouellette, and Timothy M. Korter
Department of Chemistry, 1-014 Center for Science and Technology, Syracuse University, Syracuse, NY 13244-4100, USA
Abstract: The terahertz (THz) spectrum of (1R,2S)-(-)-ephedrine from 8.0 to 100.0 cm-1 has been investigated at liquid-nitrogen (78.4 K) temperature. A complete structural analysis has been performed in conjunction with a vibrational assignment of the experimental spectrum using solid-state density functional theory (DFT). In order to obtain the crystallographic lattice constants at a temperature relevant for the DFT simulations, the compound has also been characterized by cryogenic single-crystal X-ray diffraction. Theoretical modeling (solid-state and isolated-molecule) of the compound includes the use of three generalized gradient approximation density functionals (BP, PBE, PW91) and two Gaussian-type basis sets (6-31G(d,p) and 6-311G(d,p)). Assignment of the THz spectrum is performed at a PW91/6-311G(d,p) level of theory, which provides the best solid-state simulation agreement with experiment. The solid-state analysis indicates that the seven experimental spectral features observed at liquid-nitrogen temperature are comprised of 13 IR-active vibrational modes. Of these modes, nine are external crystal vibrations and provide approximately 57% of the predicted spectral intensity.
In press in the Journal of Physical Chemistry. This paper on the low-frequency vibrational properties of methamphetamine marks a transitional point in the simulation of terahertz (THz) spectra by density functional theory (DFT), as both Crystal06 and Abinit provide the means to calculating infrared intensities in the solid-state by a more rigorous method than the difference-dipole method that has been used in the many previous THz papers with DMol3 (performed externally from the DMol3 program proper). The original manuscript came back with two important comments from Reviewer 3 (that crazy Reviewer 3. Is there nothing they’ll think of to critique?).
The best-fit spectral assignment by visual inspection (BOP/DNP level of theory) and by statistical analysis (BP/DNP level of theory) are shown below (the paper, of course, contains significantly more on this point). With these two spectral simulations in mind, Reviewer 3 presented the following analysis that I think is certainly worth considering generally to anyone new to the computational chemistry game and even by general practitioners who might risk becoming complaisant in their favorite theoretical technique. There’s a reason we refer to the collection of computational quantum chemical tools as the “approximate methods.”
I have difficulty with what appears to be a generalization of the applicability of using density functional for modeling THz spectra… It is disturbing that the different functionals will generate different numbers of modes within the spectral region, and it is hard to imagine how we should move forward with density functional for calculating spectra of this type. In fact, it is true that one needs to include the “lattice” to get the spectra right in these regions, but it is not obvious that DFT will provide the level of rigor required to develop a predictive capability. Furthermore, given the “uncertainties regarding the number of modes”, is it possible that the mode assignments are invalid?
In my opinion, the authors point out the need for solid-state DFT, but should point out that in its current incarnation, that DFT is currently inadequate for quantitative comparison with experiment, and that more work needs to be done with the theory to make it quantitative.
The response to the reviewer about these points goes as such:
We agree completely with the reviewer’s criticism on these points of spectral reproduction, but we also believe that there should be a sharp separation between the capabilities of the DFT formalism and the capabilities of the many empirically-derived density functionals that currently make up the complement of “tools” within the DFT formalism. Unlike the selection of basis set, which we often presume will improve agreement because of the improvement to the description of the electronic wavefunction that comes from additional functions, it is the case (specifically among the survey studies in THz simulations performed by the authors in this and previous publications) among the currently available GGA density functionals that the reproduction of the physical property under consideration is determined by the functional. We also know that the reproduction of the lowest-energy solid-state vibrational features in molecular solids were NOT part of the initial complement of metrics used in gauging the accuracy of density functionals, so it is clear that we are performing survey calculations using available tools to determine which tools may be most reliably employed for performing THz assignments while not actively engaged in the development of new tools. In the simulation of vibrational spectra, it is clear that we can never entirely trust the simulations until it is known unambiguously by experimentalists exactly what the motion associated with each vibrational mode is, which brings up the need for polarization experiments, Raman experiments to complement the mode assignments, etc. Such rigorous detail for this region of the spectrum is very likely not known for a great many molecules of interest by the communities most interested in the benefits of THz spectroscopy.
In the meantime and in the absence of “complete datasets,” we agree with all of the reviewers (to a point either addressed directly or indirectly through questions along the same vein) that the best that a theoretical survey like the one presented here can do is aid in the generation of a functional consensus view, which is something that requires mode-by-mode analyses as mentioned by the reviewer.
Patrick M. Hakey, Damian G. Allis, Wayne Ouellette, and Timothy M. Korter
Abstract: The cryogenic terahertz spectrum of (+)-methamphetamine hydrochloride from 10.0 – 100.0 cm-1 is presented, as is the complete structural analysis and vibrational assignment of the compound using solid-state density functional theory. This cryogenic investigation reveals multiple spectral features not previously reported in room-temperature terahertz studies of the title compound. Modeling of the compound employed eight density functionals utilizing both solid-state and isolated-molecule methods. The results clearly indicate the necessity of solid-state simulations for the accurate assignment of solid-state THz spectra. Assignment of the observed spectral features to specific atomic motions is based upon the BP density functional, which provided the best-fit solid-state simulation of the experimental spectrum. The seven experimental spectral features are the result of thirteen infrared-active vibrational modes predicted at a BP/DNP level of theory, with more than 90% of the total spectral intensity associated with external crystal vibrations.
In press, in the journal Vibrational Spectroscopy. In a bit of a departure from the last several terahertz (THz) papers, this study involves the simulation of the solution-phase THz spectrum of the very, very thoroughly studied 2,2‘-bithiophene in solution (cis and trans geometries and lowest-frequency vibrational modes are provided in the figure below), a phase both easier and more difficult than the solid-state density functional theory (DFT) calculations that have been the mainstay of previous studies. Simplicity comes from the molecular symmetry and smaller size of the system under study, with no issues of the temperature dependence of the lattice constants or the intermolecular interaction predictions complicating the spectral assignment of the lowest frequency modes. The difficulty comes from the ability to employ multiple theoretical models to study the system and the need for far higher levels of theory in the gas phase to perform an analysis worthy of experimental comparison.
In this study, the DFT and MP2 quantum chemical calculations were used to consider molecular geometry, cis and trans conformational energy differences, rotational barrier heights, the prediction of normal mode energies, and relative peak intensities.
One topic addressed in solution that is not an issue in the crystals studied to date are the accessibility of relative conformational minima at ambient conditions (kT, room temperature). In the case of 2,2‘-bithiophene, the conformational flexibility is around the exocyclic thiophene-thiophene bond. With the description of the potential energy surface (PES) for rotation about the exocyclic bond determined by conformational calculations, the second step is the determination of relative populations of the cis and trans forms as a function of temperature. In this case, weighting of the PES by the Boltzmann distribution function yields the plot shown in the bottom of the figure below, from which the relative cis and trans populations can be determined by integration of the 0 to 90 (cis) and 90 to 180 (trans) regions.
The long-short of this particular study, which I save for the article itself, is that no single theory provides all the best answers, but sufficiently high levels of theory all do settle into the reasonable vicinity of accurate. At least, to the extent that all of the experimental data is in agreement.
Anna M. Fedor1, Damian G. Allis1,2, and Timothy M. Korter1
2. Nanorex, Inc. Bloomfield Hills, MI 48302-7188 USA
Abstract: The room temperature solution-phase terahertz (THz, 7 to 165 cm-1) spectrum of 2,2-bithiophene in cyclohexane is reported. Density functional theory (B3LYP) and ab initio (MP2) methods employing the 6-311++G(2d,2p) and aug-cc-pVDZ basis sets are used to assign the THz vibrational structure and determine the relative populations of the cis and trans conformations, as well as the trans-trans rotational barrier height and the effects of the cyclohexane solvent on the predicted molecular geometries and vibrational frequencies. Significant differences are seen in the performance of the different theoretical methods, with the best performing method dependent upon the molecular property of interest. The best fit model of the experimental THz spectrum is achieved using MP2/aug-cc-pVDZ, which places the relative trans and cis populations at 54% and 46%, respectively.
Christmas came one day late.
A special issue of the International Journal of High Speed Electronics and Systems dedicated to a previous symposium on terahertz spectroscopy and imaging applications has made its way into hardcover in the form of a Selected Topics in Electronics and Systems volume (Vol. 46). The original article was blogged on this site in post #50, including the abstract and a couple of figures from the article. The book reproduces all images in greyscale, although the cover image includes an HMX molecule in pixelated technicolor. In the interest of readability (had I known…), I’m posting the images here as a single pdf (3.0 MB).