<?xml-model href='http://www.tei-c.org/release/xml/tei/custom/schema/relaxng/tei_all.rng' schematypens='http://relaxng.org/ns/structure/1.0'?><TEI xmlns="http://www.tei-c.org/ns/1.0">
	<teiHeader>
		<fileDesc>
			<titleStmt><title level='a'>Excluded-volume descriptors for dipole-bound anions: Amine &lt;i&gt;N&lt;/i&gt; -oxides as a test case</title></titleStmt>
			<publicationStmt>
				<publisher></publisher>
				<date>02/07/2020</date>
			</publicationStmt>
			<sourceDesc>
				<bibl> 
					<idno type="par_id">10359421</idno>
					<idno type="doi">10.1063/1.5142624</idno>
					<title level='j'>The Journal of Chemical Physics</title>
<idno>0021-9606</idno>
<biblScope unit="volume">152</biblScope>
<biblScope unit="issue">5</biblScope>					

					<author>Thomas Sommerfeld</author><author>Megan C. Davis</author>
				</bibl>
			</sourceDesc>
		</fileDesc>
		<profileDesc>
			<abstract><ab><![CDATA[Dipole-bound anions can be theoretically characterized at three fundamentally different levels. The highest are ab initio calculations, which themselves range from fairly approximate, say, Koopmans's Theorem (KT) or MP2, to highly sophisticated, say, electron affinity EOM-CCSDT, which rivals experiments in reliability. The next level down is represented by one-electron model Hamiltonians. Again, one-electron model Hamiltonians can be fairly approximate, especially if the molecular system is modeled by a simple point-dipole and point-polarizable site; however, very reliable models have been developed for specific systems, for example, water clusters. At the lowest level, one can qualitatively explain trends in classes of dipole-bound anions in terms of the dipole moment, µ, the polarizability, α, and the so-called excluded volume, V x .This project aims at the qualitative level. While the dipole moment and the polarizability possess clear-cut definitions, the excluded volume must-similar to all molecular volumes-remain a rather vaguely defined term, and so far we are unaware of any quantitative definition in the literature. Here we introduce and investigate three descriptors for V x . To this end, we first establish a data set with consistent ab initio results for 25 amine N-oxides structures. Then we demonstrate that the descriptors are indeed able to explain trends for sets of isomers and conformers and investigate to what extent the descriptors are able to predict electron binding energy of dipole-bound states using simple quantitative structure-property relationship-like models. It turns out that µ and V x provide a reasonably accurate prediction of the electrostatic part of the electron bind energy (the KT value) and that the polarizability α provides an acceptable prediction of the electron correlation contribution.]]></ab></abstract>
		</profileDesc>
	</teiHeader>
	<text><body xmlns="http://www.tei-c.org/ns/1.0" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xmlns:xlink="http://www.w3.org/1999/xlink">
<div xmlns="http://www.tei-c.org/ns/1.0"><head>I. INTRODUCTION</head><p>It is useful to divide anions into two major classes: valence and non-valence anions. In a valence anion, the excess electron occupies a compact valence orbital showing a spatial extent comparable to the nuclear framework. In contrast, in a non-valence anion, the excess electron occupies a spatially extended orbital largely localized outside the valence region, that is, outside the van der Waals surface of the neutral molecule.</p><p>Examples for valence anions are stable closed-shell anions such as OH -or NO - 3 known from solution and atmospheric chemistry or radical anions such as the quinone anion <ref type="bibr">1,</ref><ref type="bibr">2</ref> .</p><p>Examples for non-valence anions are dipole-bound anions (an excess electron attached to the molecular dipole) <ref type="bibr">3,</ref><ref type="bibr">4</ref> , correlation-bound anions (an excess electron unbound in Hartree-Fock but bound in correlated calculations) <ref type="bibr">5,</ref><ref type="bibr">6</ref> , doubly-Rydberg anions (two excess electrons bound to a positive core) <ref type="bibr">7,</ref><ref type="bibr">8</ref> , as well as solvated-electron states in the widest sense (see e.g. the recent review <ref type="bibr">9</ref> ). We emphasize that these two classes correspond to extremes on a continuous scale, because, on the one hand, valence and non-valence states can mix and, on the other hand, a valence state can evolve into a non-valence state, and in these more general cases it is more appropriate to speak of the valence or non-valence "character" of an anionic species.</p><p>Dipole-bound anions have been extensively studied for many decades <ref type="bibr">1,</ref><ref type="bibr">3,</ref><ref type="bibr">4,</ref><ref type="bibr">9</ref> and probably represent the best-characterized subclass of non-valence states. The name "dipole-bound" itself originates from the leading electrostatic term in an "intermolecular interaction-style" analysis, yet, instead of two molecules, the interaction between a neutral and an electron is analyzed.</p><p>A detailed quantitative analysis in this spirit can be based on the comparison of computed binding energies with different ab initio methods <ref type="bibr">10,</ref><ref type="bibr">11</ref> . In this scheme, the most useful comparison is probably that of the Koopmans's Theorem (KT) value for the electron binding energy E b with a reliable value from a coupled-cluster with single, double, and perturbative triple substitutions (CCSD(T)), or equation-of-motion couple-cluster with single and double substitutions (EOM-CCSD) calculation: KT treats exchange-repulsion and electrostatic interactions between a neutral described by Hartree-Fock theory and an excess electron. It defines therefore an electrostatic picture of a dipole-bound anion.</p><p>If electron correlation is then "switched on" at the CCSD(T) or EOM-CCSD level or tend to be substantially higher than the associated KT values, however, the percent-change is strongly molecule dependent. In fact, correlation-bound states are a limiting case of this trend showing a 100% increase. In any event, based on two ab initio electron binding energies-the approximate KT and one reliable value-the total electron binding energy can be split into an electrostatic and a correlation contribution.</p><p>An alternative theoretical description of dipole-bound states is provided by one-electron model potentials and the associated one-electron model Hamiltonians. Generic one-electron models typically consist of a point-dipole, a point-polarizable site, and a repulsive potential (see, e.g., <ref type="bibr">[12]</ref><ref type="bibr">[13]</ref><ref type="bibr">[14]</ref> ), yet finite dipoles and polarizability models based on Drude oscillators (charge-on-a spring models) represent other choices. More sophisticated one-electron models have been developed for specific systems, including water <ref type="bibr">15,</ref><ref type="bibr">16</ref> , methanol <ref type="bibr">17</ref> , and ammonia clusters <ref type="bibr">18</ref> , and these models can be highly competitive especially for large systems. Since it is possible to study the impact of switching on or off any part of the electron-molecule interaction, one-electron model potentials may also be used for interpretative purposes. Regarding the current context, however, we make use of an approximate closed-form solution of a particular simple model <ref type="bibr">14</ref> ; we will come to this point in section III.</p><p>Last, much insight has been gained from a qualitative conceptual framework developed by comparing different molecules in related families. In retroperspective, it is hardly surprising that apart from the dipole moment, electron binding energies of non-valence states are impacted by the molecular polarizability and the excluded volume <ref type="bibr">3,</ref><ref type="bibr">19</ref> . Loosely speaking, the dipole moment may be the leading electrostatic term in a long-range expansion, however, first, the binding energy is equally determined by intermediate and short-range interactions, and second, the dipole moment may be small, while other terms may be large.</p><p>At intermediate distances, the molecular polarizability may contribute substantially to the attractive interaction, and the short-range repulsion is characterized by the excluded volume: Consider, for example, a molecule, where the molecular dipole is associated with a certain functional group, say a cyano or nitro group. Depending on the molecular structure, the excess electron may then be able to easily access the positive end of this substructure, or it may be crowded out by other fairly non-polar groups. The volume where Pauli-repulsion associated with this crowding effect is noticeable, is referred to as excluded volume.</p><p>Examples clearly demonstrating the excluded volume effect are (</p><p>where Ar atoms crowd the positive end of the (H 2 O) 2 dipole <ref type="bibr">19</ref> , electrons dipole-bound to nitriles, where increasing subsitution at the &#945; carbon leads to higher dipoles but lower binding energies <ref type="bibr">3</ref> , and the 2s-like electron in Li(NH 3 ) N (N = 0 -4) clusters, which is increasingly pushed outside of the first solvation shell of a Li + cation 20 . On the other hand, a recent study of electron detachment from para-substituted alkyl-phenolate ions (the dipole-bound state is an excited state of the anion and gets populated during the photodetachment experiment) shows effectively no influence of the alkyl chain length on the spectroscopic signature of the dipole-bound state <ref type="bibr">21</ref> . While we have no definite explanation for this observation, we will briefly comment on it in the discussion.</p><p>Out of the three descriptors required for the qualitative level, the dipole moment and the polarizability possess clear-cut definitions, whereas the excluded volume represents a more vague term. Here, the goal is to develop excluded volume descriptors useful for analyzing trends both qualitatively and quantitatively, and that may also point the way to reducing the number of free parameters in one-electron model Hamiltonians for non-valence-bound states.</p><p>To this end, first the new class of dipole-bound electrons to amine N-oxides is investigated so as to establish a consistent data set of electron binding energies and properties. Amine N-oxides were chosen for three reasons: First, owing to their very polar coordinate covalent NO bond, they possess high dipole moments. Second, the three substituents on the N atom directly control the excluded volumes. Third, dipole-bound states of amine N-oxides have not been systematically studied, and thus the data set alone represents a significant finding.</p><p>Then, various excluded volume descriptors are defined, evaluated, and accessed regarding their use in the qualitative comparison of certain isomer groups as well as their use as quantitative descriptors in qualitative structure-property relationships (QSPR) for the whole data set.</p></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head>II. COMPUTATIONAL METHODS</head></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head>A. Excluded Volume Descriptors</head><p>While giving a qualitative description of an excluded volume is fairly straightforward (see the Introduction), giving a quantitative definition is less straightforward. Let us at this point emphasize that the excluded-volume as understood in the context of dipole-bound anions is quite different from the excluded-volume used in the theory of liquids and gases. Both are "crowding effects", however, while in solution and gases the whole molecular volume of molecule A is unavailable to molecule B, in the context of dipole-bound states the question pertains to groups or molecules excluding the excess electron from approaching the positive center of the dipole.</p><p>To compute the excluded-volume in dipole-bound states, it is thus insufficient to know the volume of the crowing groups or cluster members. Instead an integral is needed, which weighs volume closer to the positive center and volume in the direction of the dipole vector more heavily, and the type of atoms in the volume to be integrated over make little difference.</p><p>All descriptors defined below rely on the identification of a "positive center", R + , and are therefore most easily applicable to the special case of a molecule possessing a single strongly polar group from which the molecular dipole derives. As chemical intuition suggests, and</p><p>L&#246;wdin or Mullikan populations confirm, for amine N-oxides the choice</p><p>Other obvious examples with a fairly obvious choice of R + are organic molecules that feature only a single nitrile or nitro group, although for these two compound classes the precise location of R + becomes less clear if the polar group is attached to a conjugated &#960;-system. Further generalizations are obviously needed for molecules with various binding sites, say, water clusters with multiple dangling OH bonds.</p><p>While a more general definition of our excluded volume descriptors is in principle possible, the definitions for a single positive center are attractive due to their simplicity, their small number of parameters, and their suitability for the present context. Therefore, for this pilot application, we restrict ourselves to simple models.</p><p>A first descriptor for the excluded volume is based on a density for the excess electron &#961;( r) and measures directly how close the excess electron can approach the positive center:</p><p>Possible choices for &#961;( r) are the lowest unoccupied orbital (LUMO) of a Hartree-Fock calculation for the neutral, the singly occupied orbital of a Hartree-Fock calculation for the anion, or natural or Dyson orbitals from direct methods to compute the electron attachment energy such as EA-EOM-CCSD or ADC <ref type="bibr">22</ref> .</p><p>Yet, the descriptor r + possesses-by construction-two great disadvantages. First, any descriptor based on &#961;( r) is, of course, unsatisfactory in that it implies finding the density of the anion first. Indeed, r + tastes and feels more like an analysis approach for the electronic structure of anions than a qualitative descriptor of the neutral. Second, while r + reflects without any doubt the effects of any substituent that prevents the excess electrons from approaching the positive center, it also reflects any other influence on the electron distribution, say, that of the dipole moment. Therefore r + can only serve as an indirect descriptor for the excluded volume; it may, however, serve as a general descriptor of a dipole-bound state and aid in interpreting electron distributions of non-valence states (c.f. Ref. <ref type="bibr">23</ref> ).</p><p>More direct descriptors for the excluded volume can be defined by integrating the van der Waals volume of the neutral molecule from the positive center in the direction of the dipole. To compute these descriptors, V</p><p>x , one needs four ingredients: the position of the nuclei to define the van der Waals volume, the position of the positive center R + (here, the position of the N atom), the direction of the dipole &#181; (here, the direction of the N-O bond), and, last, a weight function, w i that picks out the part of the molecule hindering the excess electron to approach R +</p><p>Here the integral is evaluated over the van der Waals volume of the neutral (see ref. <ref type="bibr">24</ref> for radii), and w i (r + , cos &#952;) is one of three weight functions defined below, which depend on the distance from the positive center r + and the angle &#952; between the dipole vector and the vector from the positive center to the position of the electron r:</p><p>The first weight function w 1 (cos &#952;) simply integrates all volume in the direction of the dipole vector and goes to zero for &#952; = 90 &#8226; , but does not discriminate for distance:</p><p>Alternative definitions are motivated by ab initio and model electron densities of dipole bound states, which show maxima close to the positive center and then fall off suggesting that the weight function should also fall off with r + . Two models were considered:</p><p>and</p><p>Both weight functions w 2 and w 3 rise quadratically from the origin, show a maximum, and fall off to zero, and for both the parameter r 0 is chosen such that the maximum occurs at the van der Waals radius of an N atom (1.55 &#197;24 ).</p><p>The absolute excluded volumes computed by these means are more or less arbitrary;</p><p>what matters are the relative values used to compare different N-amine oxides. We therefore express all excluded volumes relative to a reference value, and choose</p><p>x of H 3 NO for this purpose. Thus, regardless of weight function, the excluded volume of H 3 NO is defined to be 1, and trends obtained with different weights can be compared in a meaningful way.</p></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head>B. Ab initio Methods</head><p>The following computational protocol has been used to establish the data set comprising electron binding energies and properties: First, the geometry of the neutral amine-N-oxide was optimized using second-order M&#248;ller-Plesset perturbation theory (MP2) and the jul-cc-pVTZ basis set <ref type="bibr">25,</ref><ref type="bibr">26</ref> . The same basis set was used to compute self-consistent-field (SCF) and MP2 dipole moments as well as SCF and MP2 polarizabilities.</p><p>Dipole-bound states of molecules normally show only a very small geometry relaxation upon electron attachment, and the vertical electron affinity of the respective neutral, the vertical electron detachment energy of the anion, and the adiabatic electron affinity typically vary by less than 5%. [In contrast, dipole-bound states of clusters, show often substantial geometry changes, the water dimer is a drastic example <ref type="bibr">4</ref> .] Therefore, only the vertical electron affinity of the neutral was computed, and we refer to it as binding energy, E b .</p><p>In all calculations of the binding energy, the aug-cc-pVDZ basis <ref type="bibr">25</ref> was augmented with several further sets of diffuse s, p, and d functions centered on the amine N atom, which represents the positive center of the molecular dipole (see ref. <ref type="bibr">9,</ref><ref type="bibr">27</ref> for basis set requirements of dipole-bound states). The exponents of these functions were chosen as an even-tempered series starting from the respective smallest exponent in the nitrogen aug-cc-pVDZ set with a scaling factor of 3.5. The number of additional spd sets needed to converge the computed binding energies depends on the specific amine N-oxide and the theoretical method employed.</p><p>In most cases a (5s5p5d) set yields well-converged results, and all results reported here were obtained with at least this set. However, some low-binders require up to (7s7p7d) to converge the uncorrelated Koopmans's Theorem (KT) estimate of E b .</p><p>Apart from KT, E b was computed with so called &#8710;-methods, where E b is obtained as the difference between the total energies of the neutral and the anion as well as with so-called direct methods, which avoid computing an explicit anion wavefunction and yield E b directly.</p><p>Initially many &#8710;-methods and direct methods were compared for the two smallest systems and the less costly methods were further compared for two typical intermediate-size amine N-oxides.</p><p>The particuar &#8710;-methods used in this study were &#8710;SCF, &#8710;MP2, &#8710;OO-MP2 (orbital optimized MP2), and the &#8710; coupled-cluster variants &#8710;CCSD, &#8710;CCSD(T), &#8710;CCSDT, &#8710;CCSDT(Q), that is, up to &#8710;-coupled cluster with iterative single, double, and triple substitutions and non-iterative quadruple substitutions. For all &#8710;-methods spin-unrestricted SCF calculations were used to compute the energy of the anion, and, as expected for dipole-bound states, spin contamination is negligible.</p><p>All direct methods beyond KT were variants of the equation-of-motion (EOM) coupledcluster family <ref type="bibr">28,</ref><ref type="bibr">29</ref> : EOM-CCSD, its second-order approximation EOM-CCSD(2), which is sometimes referred to as EOM-MP2, EOM-CCSDT, as well as three different corrections of EOM-CCSD for triple substitutions: EOM-CCSD*, EOM-CCSD(T)(a)*, and EOM-CCSD(T)(a) (see the discussion in Ref. <ref type="bibr">29</ref> for details).</p><p>Four ab initio codes were used. The Orca package <ref type="bibr">30,</ref><ref type="bibr">31</ref> (version 4.1) was used for optimizations, to compute dipoles, polarizabilities as well as OO-MP2 and many EOM-CCSD calculations. However, geometry optimizations with confined dihedral angles were carried out using the Psi4 package, version 1.2 <ref type="bibr">32</ref> . Further EOM-CCSD calculations, in particular for natural orbitals, as well as all other EOM calculations were carried out with version 2.1 of the CFOUR package <ref type="bibr">33</ref> . Last, CCSDT, CCSDT[Q], and CCSDT(Q) calculations were carried out through the MRCC <ref type="bibr">34</ref> interface provided by CFOUR.</p></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head>III. RESULTS AND DISCUSSION</head><p>In this section we will first report and discuss ab initio results and excluded volume descriptors for all amine N-oxides studied. Then, we will compare particular sets of isomers, conformers, or conformations that show excluded volume influences most clearly, before discussing how the descriptors can be used in QSRP models of the binding energy.</p><p>Sixteen different amine N-oxides were investigated (see Tab. I); eleven of these have just one relevant conformer, yet, ethylamine N-oxide, isopropylamine N-oxide, and piperidine N-oxide possess two, ethylmethylamine N-oxide possesses three, and propylamine N-oxide possesses four. Moreover, we include the synperiplanar (syn) conformation of ethylamine N-oxide (torsion angle of 0.0 &#8226; between the NO and the CC bonds) making a total of 25 dipole-bound states. Cartesian coordinates of all considered structures are available as supplementary information.</p></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head>A. Correlation Effects on the Binding Energy of Dipole-bound States</head><p>Predicting accurate electron binding energies represents one of the more challenging tasks of quantum chemistry. Well known issues include the more diffuse nature of anions, lack of error compensation due to the neutral and anion possessing a different number of electrons (methods must be size extensive), and, in case of density functional calculations, the need for the correct behavior of the self-interaction energy <ref type="bibr">2,</ref><ref type="bibr">35</ref> . Our molecule set includes some fairly large amine N-oxides, and it was clear that we could not afford to compute highly accurate E b values for the larger species. Nevertheless, in order to establish a consistent data set, we needed an affordable yet sufficiently accurate method. Hence, the two smallest amine N-oxides were studied with very accurate methods, and two intermediate-size systems were studied with some of the more costly methods.</p><p>Our results are collected in Tab. II. The rough trends emerging for the N-oxides follow the typical trends that were initially established in Refs. <ref type="bibr">10,</ref><ref type="bibr">11</ref> and summarized in Refs 2,4,9 :</p><p>Relative to a purely electrostatic picture (no electron correlation) as defined by the KT binding energy, electron correlation increases E b dramatically. In contrast, average orbital relaxation as described by a &#8710;SCF calculation has virtually no impact on E b .</p><p>In view of being the most primitive correlated level, &#8710;MP2 represents a big improvement over KT, but to capture a substantial share of the correlation contribution to E b , at the very least an &#8710;OO-MP2 or an EOM-CCSD(2) calculation is needed. For many other dipolebound or other non-valence anions, even that may be insufficient for acceptable accuracy <ref type="bibr">6</ref> , but for the large-dipole low-polarizability amine N-oxides considered here, these two secondorder methods are very competitive. As discussed in Ref. <ref type="bibr">6</ref> , the critical ingredient for a reliable method is its ability to account for orbital relaxation in the presence of electron correlation, and both orbital-optimized MP2 as well as equation-of-motion methods share this feature.</p><p>Correlation effects beyond EOM-CCSD can be investigated in two ways. First, the EOM-CCSD energy can be corrected for triple substitutions. The four molecules considered in Tab. II demonstrate that the absolute effects are only a few meV, but the relative changes are increases in E b of about 5%.</p><p>Second, the EOM-CCSD value can be compared with &#8710;-methods, but owing to the totally different theoretical approaches, it is not so clear which &#8710;-method compares most directly to EOM-CCSD itself. What is clear, however, is that &#8710;CCSD captures very roughly 90% of the binding energy, and that &#8710;CCSDT seems converged as judged by additional changes of perturbative quadruple substitutions, but slightly too small in comparison to EOM-CCSDT.</p><p>The so called golden standard, &#8710;CCSD(T), yields somewhat smaller binding energies than &#8710;CCSDT, and in contrast to many "closed-shell-only properties," CCSD(T) seems inferior to the full triples method in the present context-presumably owing to the lack of error compensation typically associated with calculating electron binding energies.</p><p>Based on the results in Tab. II, we decided on EOM-CCSD as a workhorse method, because we were able to do EOM-CCSD calculations for all but one amine N-oxide, and because the accuracy of EOM-CCSD(2) can be expected to deteriorate for the high polarizability, low E b cases 6 . Moreover, E b was computed for all molecules using KT to provide an electrostatic reference and using &#8710;OO-MP2 as the best tradeoff between accuracy and affordability.</p></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head>B. Ab Initio Results and Excluded Volume Descriptors</head><p>For every amine N-oxide considered, Tab. III reports the electron binding energy computed with KT, EOM-CCSD, and &#8710;OO-MP2, the dipole moment &#181;, the polarizability &#945;, and the excluded volume descriptors defined in eq.( <ref type="formula">1</ref>) and eq.( <ref type="formula">2</ref>).</p><p>Before trying to identify any specific relationship, let us discuss a few general trends in the established data set:</p><p>1. Similar to the four amine N-oxides discussed in depth in the last subsection, the electrostatic value for E b obtained using KT is always substantially lower than the many-body result obtained using EOM-CCSD or &#8710;OO-MP2. 3. The MP2 dipole moments range from 4 to 6 D, and the SCF values are only slightly larger than the MP2 dipoles (differences of 0.18 to 0.08 D, which is less than 5%).</p><p>Clearly, the relaxation of the electron density due to many-body effects is small, and SCF dipoles are a good approximation for this molecule class.</p><p>4. Similar to the dipole moments, the SCF and MP2 polarizabilities agree well (to about 3%) with each other. However, in contrast to the dipole predictions, the two values correlate perfectly with each other (coefficient of determination of 1.0) and almost perfectly with the molar mass (coefficient of determination of 0.999).</p></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head n="5.">The excluded volume values, V</head><p>x , from our models (eq.( <ref type="formula">2</ref>)) display surprisingly little "group additivity," even when considering very simple series such as, say, H 3 NO, All four plots show some degree of correlation-the dipole moment a positive one, and the three other parameters a negative one-but even by-eye it is obvious that only r + shows a semi-acceptable correlation with E b . As discussed above, any E b vs. r + correlation is not much of a surprise, since both quantities originate from closely related ab initio calculations.</p><p>On the contrary, we find it rather astounding that this correlation is not significantly better.</p><p>More astounding, in a class of closely related dipole-bound anions with dipoles varying over a wide range of 2 D, one should expect a clear correlation of the binding energy with the dipole moment, but even ignoring H 3 NO, the E b vs. &#181; correlation is obviously poor. We therefore conclude that the trends in the binding energy cannot be explained by a single parameter, but that several parameters have to be combined.</p><p>One way to identify at least qualitative trends consists of comparing the isomers and groups and conformers having the same molar mass. By this means, polarizability can be effectively eliminated as a variable, as the involved polarizabilities are almost equal, and the trends for this specific subgroup can be studied in terms of the dipoles and the excluded volume alone. The first group of amine N-oxides that can be compared in this way is the C 2 H 7 NO group, which consists of Me 2 HNO and EtH 2 NO. We first consider the different EtH 2 NO conformations in some detail before comparing them with their Me 2 HNO isomer.</p><p>Then we discuss the C 3 H 10 NO group.</p><p>EtH 2 NO possesses two conformers, anti and gauche, and its conformational space can be studied by varying a single variable: the O-N-C-C torsion angle &#952;. We studied the behavior along this coordinate by keeping this angle frozen at 13 different values, optimizing the geometry of all other structural parameters, and computing E b , &#181;, and</p><p>for each optimized structure (Fig. <ref type="figure">2</ref>).</p><p>The binding energy is essentially flat (46 -49 meV), showing only a small oscillation with the 120 &#8226; frequency that is typically associated with the changes between eclipsed and staggered conformations. In contrast, both the dipole and the excluded volume descriptors display sigmoid character. The dipole increases from about 5 D to 5.7 D, and the excluded volume descriptors increase almost synchronously with it. However, while V</p><p>(1) x and V (2) x approximately double, V (3) x</p><p>increases "only" by 25%. Be that as it may, the major trend of a binding energy, which is essentially independent of torsion angle, can only be explained by considering both, &#181; and V x , and the impacts of the two must practically cancel each other out.</p><p>A visual impression of the excluded volume effect can be gained from cuts through the "singly occupied" EOM-CCSD natural orbitals (occupation numbers of 0.998), which de-scribe the dipole-bound electron (Fig. <ref type="figure">3</ref>). For both the syn and the anti conformation, the electron distribution resembles a large s-type orbital off the dipole created by the NO group.</p><p>Loosely speaking, both electron distributions look like apples, with two stems: the N-O bond, and the N-ethyl chain. The NO bond creates the dipole and points more of less towards the apple core. In contrast, for the syn conformation, the ethyl group bends back from the electron density (the apple) and may influence it only via long range forces. while, for the anti conformer, the ethyl group points directly into the electron density; in the apple picture, the ethyl-stem bents back and punches a hole into the core.</p><p>While the natural orbitals offer an intuitive understanding of the excluded volume, they unfortunately offer little understanding of why the smooth synchronous increases of both dipole moment and excluded volume lead to such small variations in binding energy. (3)</p><p>x , on the other hand, yields a somewhat different pattern; however, the conclusion reached is identical.</p><p>In the C 3 H 10 NO group, the highest binding energies are associated with the lowest excluded volumes and fairly small dipole moments (lower left corner of Fig. ( <ref type="formula">4</ref>)). If one then keeps the dipole fixed at about 5 D and increases V x , the binding energy drops accordingly, emphasizing again the excluded volume effect and that the definition of our descriptor makes at least qualitative sense (left edge of Fig. ( <ref type="formula">4</ref>)). However, if both &#181; and V x are changed at the same time, no clear pattern emerges. At best, one can say that along the diagonal of Fig. ( <ref type="formula">4</ref>), the binding energy shows no clear trend, but that the variations are also smaller than in the lower left corner of Fig. (4). Therefore, similar to the EtH 2 NO conformation, the increasing dipole moment and excluded volume must largely cancel each other out. While a straightforward quantitative trend may not be obvious (at least not to us), it seems obvious that dipole moment and excluded volume both represent significant but opposing influences on the binding energy.</p></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head>C. Empirical Models for the Binding Energy</head><p>Having discussed trends in the binding energies of isomer groups in the light of our excluded volume descriptors, this section explores the utility of the descriptors as quantitative parameters in QSPR of the binding energy. The mathematical form of these relationships separates the effects of the dipole moment &#181;, the excluded volume V x , and the polarizability &#945; and is motivated partly by the careful analysis of one-electron model Hamiltonians 14 , partly by our own empirical observations, and partly by the separation of the binding energy into two contributions, one due to electrostatic one due to electron correlation effects <ref type="bibr">4,</ref><ref type="bibr">10</ref> .</p><p>One-particle model Hamiltonians describing a light charged particle bound to a dipole have been developed by many groups (see e.g. <ref type="bibr">12,</ref><ref type="bibr">14,</ref><ref type="bibr">36,</ref><ref type="bibr">37</ref> ). As a rule, the model potentials display cylinder symmetry, but nevertheless separate into an angular and a radial equation in spherical coordinates. While the exact solution of these equations requires numerical steps, for certain simple models approximate closed-form solutions can be obtained. One such model potential consists of a point-dipole and a hard sphere repulsive shell <ref type="bibr">14</ref> . In this model the binding energy is approximately</p><p>where &#181; 0 is the critical dipole moment of 1.625 D, r s is the radius of the hard-sphere repulsive core, and A, B and C are three parameters: A is a proportionality constant, the B term is the leading term describing the threshold behavior, and the C term is the first correction.</p><p>Ref. <ref type="bibr">14</ref> treat them as fitting parameters and found that eq.( <ref type="formula">7</ref>) almost perfectly reproduces the exact (numerical) solutions of their model.</p><p>We note that the analysis in Ref. <ref type="bibr">14</ref> initially meanders towards a primitive representation of an excluded volume in terms of the parameter r s , which can be computed for any molecule provided &#181; and E b are known: r s (E b , &#181;). However, the r s values computed by these means were deemed "nonphysical," and the model Hamiltonian was then extended by a polarization potential featuring an additional polarizability dependent cutoff radius, chosen so as to render r s a system-independent constant.</p><p>Be this as it may, the analytical expression for E b (eq.( <ref type="formula">7</ref>)) immediately suggests replacing</p><p>x , and we consider the following two simple QSPRs:</p><p>and</p><p>where the exponent m of V (i)</p><p>x is an additional fit parameter that would be 2/3 if the excluded volume was spherical and the analytical expression from Ref. <ref type="bibr">14</ref> was exactly valid. Since both QSPRs do not include the polarizability, it is natural to use eqs.( <ref type="formula">8</ref>) and ( <ref type="formula">9</ref>) to fit the KT values for E b , and to include electron correlation effects at a second stage.</p><p>A third QSPR considered is based on our empirical observation that in a log-log plot electron binding energies of dipole-bound states show a roughly linear correlation with the dipole moment (needless to say, this correlation shows a lot of noise and its application is limited to "lines that guide the eye"). Still, a power law for the dipole dependence of the binding energy may be a reasonable model, and as a second three-parameter model we study</p><p>Again, this relationship lacks any dependency on the polarizability, and is therefore fitted to KT binding energies.</p><p>Results from fitting the three QSPRs defined in eqs.( <ref type="formula">8</ref>), ( <ref type="formula">9</ref>) and ( <ref type="formula">10</ref>) are collected in Tab. IV. Each of the three relationships is fitted in combination with each of the three excluded volume descriptors</p><p>yielding a total of nine parameter sets. Regarding opti-mization of parameters, we minimized the mean square error based on relative errors</p><p>Relative errors are needed for this purpose because the electron binding energy of H 3 NO represents a brutal outlier that would dominate &#967; 2 otherwise. However, for the purpose of comparing different model/excluded-volume-descriptor combinations, we do not use &#967; 2 , but rather the mean of the absolute value of the relative errors (MARE)</p><p>because it has the same unit as the relative error and is the more intuitive quantity.</p><p>For all nine QSPR/excluded-volume-descriptor combinations, the MARE fall in the range between 7% and 15%. While these errors may seem large, they should be seen in the context of the simplicity of the three QSPRs and the obvious lack of any direct correlation of E b with either &#181; or V x .</p><p>The main MARE differences are due to the excluded volume descriptors with V</p><p>(2) x performing best, V (1) x being a very close runner-up, and V (3) x performing significantly worse.</p><p>However, we note that both V</p><p>(2) x and V</p><p>(3) x depend on a length-scale parameter that has been arbitrarily chosen to reflect the size of an N atom. An optimization of this parameter will, of course, result in a smaller MARE, the question is only whether the gain is worth introducing yet another free parameter. At this point our data set seems too small to explore this question.</p><p>In contrast to the different excluded volume descriptors, the three QSPRs perform at a very similar level. While the second model (eq.( <ref type="formula">9</ref>)) must by definition outperform the first (eq.( <ref type="formula">8</ref>)), introducing the C parameter does not significantly improve the description of the data. Similarly, only slight differences between the performance of the first and the third model can be observed for a given excluded volume descriptor.</p><p>In addition to the similar performance of the three QSPRs, note the following consistencies in the optimized parameters: (1) The exponent m of the excluded volume depends only on the excluded volume descriptor, but is virtually independent of the QSPR. (2) The exponents associated with V (1) x and V</p><p>(2) x are roughly 0.85 &#177; 0.02 and 0.77 &#177; 0.02, so V (1) x and V</p><p>(2) x indeed behave similar to a hard-sphere volume (m = 2/3). (3) The exponent of the dipole moment, n, in the empirical model (eq.( <ref type="formula">10</ref>)) is fairly close to 4. Thus, at least a posteriori it seems possible to reduce the number of free parameters by replacing a subset (n and m are obvious candidates) with fixed integers or fractions, e.g., for V (2)   x : n = 4 and m = 3/4. Last, we turn to the correlation contribution to electron binding energy</p><p>In order to create a consistent data set for all amine N-oxides, E b is estimated using OO-MP2. Naively, one might expect that E abs corr should correlate with polarizability, perhaps even in a linear fashion. Yet, this is clearly not so: the lightest molecule, H 3 NO, possesses the smallest polarizability, but it shows by far the largest E abs corr . Indeed, as surprising as this may seem, a plot of E abs corr vs. &#945; suggests that E abs corr and &#945; are in fact uncorrelated. On the other hand, one-electron model Hamiltonians demonstrate that a wide range of data can be reproduced by point-dipole and point-polarizable site models (see e.g. <ref type="bibr">14,</ref><ref type="bibr">37,</ref><ref type="bibr">38</ref> ).</p><p>Hence, one is led to the conclusion that electrostatics and polarizability do not act independently but rather cooperatively. In an ab initio context, which separates E b into KT and rest, we unfortunately cannot directly account for cooperation, however, we can switch to a definition of the correlation contribution in relative terms:</p><p>At least for the present data set the relative correlation energy E corr generally grows with molecular polarizability and for the highest polarizabilities-by construction-shows a saturation close to 100%.</p><p>The definition of E corr as a relative quantity has two further advantages. First, the overall shape of an E corr (&#945;) model can be adumbrated from its asymptotics: For vanishing &#945;, E corr must approach zero, and for large &#945;, E corr must approach 100%, and thus E corr (&#945;) must behave like a switching or sigmoid function. Second, eq.( <ref type="formula">14</ref>) naturally includes correlationbound states, which do not bind at KT, and normally show only relatively small binding energies, but large polarizabilities.</p><p>One strategy is then to fit the data with one of the well known sigmoid functions from the literature, say,</p><p>where b is a free scaling parameter. This straightforward one-parameter function fulfills the conditions for &#945; = 0 and &#945; &#8594; &#8734; and yields a very satisfactory fit with a MARE of 4.4%. The fit can be improved by three means: First, a sigmoid function with more free parameters, such as the general logistic function</p><p>may be chosen, where Q, b and n are free parameters. Second, the condition E corr (0) = 0 may be relaxed so as to fit only in the given domain of polarizabilities, in particular, because the threshold law of E corr (&#945;) is unknown. Third, after making the first two improvements, the fit can be further enhanced by replacing the isotropic polarizability, which has so far been tacitly used, by &#945; &#8869; , the average polarizability perpendicular to the direction of the dipole vector. [It goes without saying that replacing &#945; with the parallel component, &#945; , yields a correspondingly worse fit.] Still, implementing all three improvements brings the MARE down by only 1.5% and the best fits that can be obtained in this way show MAREs in the order of 2.9%. The two optimized switching functions (eq.( <ref type="formula">15</ref>) and eq.( <ref type="formula">16</ref>)) are compared in Fig. <ref type="figure">5</ref>.</p></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head>IV. SUMMARY AND CONCLUSIONS</head><p>Trends in electron binding energies of dipole-bound states can be understood in terms of three properties of the neutral molecule: Its dipole moment, &#181;, its polarizability, &#945; and its excluded volume, V x , which is any volume close to the positive end of the dipole, a dipolebound electron is crowded out of by unpolar substituents or other molecules <ref type="bibr">3,</ref><ref type="bibr">19</ref> . All three, &#181;, &#945;, and V x are also used in one-electron models for dipole-bound states, however, while the dipole moment and the polarizability are well-defined quantities, V x is not.</p><p>Here, we aim at developing descriptors for the excluded volume that are useful for analyzing trends qualitatively and quantitatively and that may also point the way toward a reduction of the number of free parameters in general one-electron model Hamiltonians for non-valence-bound states.</p><p>To this end, first a data set comprising electron binding energies and properties of 25 amine N-oxides is established. Amine N-oxides were chosen because they possess high moments on account of their very polar NO bond, and their excluded volume can be easily controlled by varying the three substituents on the N atom. Moreover, dipole-bound states of amine N-oxides have not been systematically studied, and thus the data set alone represents a significant finding. Specifically, the electron binding energy, E b , the dipole moment and the polarizability are computed at the one-particle and the many-particle level.</p><p>Next, various descriptors for V x are defined (see eqs.( <ref type="formula">1</ref>) and ( <ref type="formula">2</ref>)). Two descriptors depend on the density of the excess electron. These are-by construction-less useful, because, on the one hand, a reliable, and therefore computationally expensive density must be computed first, and, on the other hand, the electron density itself is influenced by &#181;, &#945;, and V x , and isolating the V x influence is anything but straightforward. In contrast, the three definitions in eq.( <ref type="formula">2</ref>) are functions of the molecular structure and the direction of the dipole moment only, any prior computation of the anion state is unnecessary, and computing V x is computationally inexpensive.</p><p>The benefit of the excluded volume descriptors V</p><p>(n) x</p><p>(n = 1 -3) defined in eq.( <ref type="formula">2</ref>) was demonstrated in a systematic comparison of sets of isomers, conformers and conformations.</p><p>These subsets of amine N-oxides possess identical molar masses and therefore almost identical polarizabilities, and regarding the present context, their differences in E b should thus be due to their differences in dipole and excluded volume. For all subsets considered in this way, even including sets that just contained conformations of a single isomer, it turned out that the trends in E b could only be explained by a combination of &#181; and V x , but not by either factor on its own.</p><p>Our findings at the qualitative level are clearly reflected in the simple QSPRs developed to predict the electrostatic part of the binding energy as function of &#181; and V x (eqs.( <ref type="formula">8</ref>) -( <ref type="formula">10</ref>)). The most direct comparison of the impacts of &#181; and V x is perhaps possible for the empirical model (eq.( <ref type="formula">10</ref>)) and V</p><p>x : In this QSPR/V (i)</p><p>x combination the exponent of (&#181; -&#181; 0 ) is close to 4 and the exponent of V x is close to -3/4, and therefore one might think that variations in dipole moment have a larger impact on the binding energy. While this is true from a strict mathematical point of view, the typical ranges invert the picture: Dipoles vary only from 4 to 6 D (150%), whereas V</p><p>(2) x varies from 1 to 17, and thus the smaller exponent of V</p><p>(2) x is easily over-compensated by its range of values, giving the excluded volume-at least for this class of molecules-a somewhat larger impact on the binding energy.</p><p>Further, let us note that combinations of any of the three QSPRs with either V (1) x or V (2)  x fit the ab initio data reasonably well (MAREs in the order of magnitude of 7 -8%). This is quite remarkable because the electron-molecule interactions in dipole-bound states are weak and subject to subtle effects, and while the dipole and excluded volume may dominate the long and short-range interactions, respectively, there will be various higher-order and cross terms that become important at intermediate ranges.</p><p>Fitting in a second step E corr suggests the following interpretation: Starting from the electrostatic contribution to the electron-molecule interaction, which is determined by the dipole moment and the excluded volume, the correlation contribution is determined by the molecular polarizability. However, electrostatics and polarizability act cooperatively, and hence only the relative correlation contribution to the binding energy (eq.( <ref type="formula">14</ref>)) can be related to polarizability, where the functional form E corr (&#945;) must follow a sigmoid function.</p><p>A fit of E corr can be slightly improved by using the polarizability component perpendicular to the dipole vector, &#945; &#8869; , as a dependent variable, suggesting that while the polarizability can be thought of as contracting the electron density in all directions, the direction orthogonal to the dipole axis is somewhat more important than the direction of the dipole itself.</p><p>Only a slight improvement, but nonetheless one that demonstrates that even simple fitting procedures can sometimes offer a glimpse of physical insight.</p><p>Our best QSPR of E corr features a MARE of less than 3%, which is highly competitive in view of the high computational cost of obtaining a reliable E corr in ab initio calculations. Thus, our results suggest that-at least for a family of closely related compounds-an accurate electron binding energy can be predicted from the dipole moment, the polarizability, one of our excluded volume descriptors, and the KT value of the binding energy.</p><p>Last, let us briefly comment on the recent measurements by Castellani et. al. <ref type="bibr">21</ref> . In summary, Liu et. al. <ref type="bibr">39</ref> measured the binding energy of the non-valence anion state of phenoxide (Ph-O), which is 12 meV. In Ref. <ref type="bibr">21</ref> , spectra of various para-substituted phenoxides were compared with Ph-O itself, and it was demonstrated that this non-valence state is "not destroyed" by alkyl substitution; however, the binding energy itself was not reported.</p><p>We emphasize that the following is conjecture, because, in the first place, the particular values of the binding energies with increrasing alkyl group length are unknown. In the second place, while the dipole-moments of the phenoxides are predicted to be about 4 D <ref type="bibr">21,</ref><ref type="bibr">39</ref> and therefore comparable to the smallest dipoles in the N-amine oxide class, the observed binding energy of 12 meV is considerably smaller than that of any N-amine oxide energies.</p><p>The most direct comparison can perhaps be made between Ph-O and Ph 3 NO, because both possess similar dipole moments and Ph 3 NO possesses the smallest E b in our data set. Another option may be PhMe 2 NO, which may provide a closer match in excluded volume (the excluded volume of Ph-O cannot be defined in as simple terms as the one for N-amine oxides, since the positive center is at least to some extent delocalized over the ring).</p><p>Nonetheless, for both Ph 3 NO and PhMe 2 NO the lion's share of the binding energy stems from electron correlation, which together with the small binding energy of 12 meV suggests that the non-valence state of Ph-O has significant polarization-bound character.</p><p>Based on this scant information, we put forward two possible explanations for the experimental observations: First, provided the electrostatic contribution to the binding energy of a non-valence state is small, the excluded volume will have only a minor impact on the total binding energy, because it affects only the former (the KT part for the N-amine oxides).</p><p>Second, the impact of alkyl substitution on the electrostatic part of the binding energy may be compensated by an increase in polarizability, in particular, if the electrostatic contribution is small. In-depth experimental or theoretical studies are needed to shed light on this fundamental issue.</p></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head>SUPPLEMENTARY MATERIAL</head><p>See supplementary material for the optimized structures of all N-amine oxides considered.</p><p>TABLE I. Systematic names of amine N-oxides studied and compact formulas used in the rest of the paper. Note that syn ethylamine N-oxide does not represent a conformer, but a conformation, and is therefore labeled with its torsion angle 0 &#8226; . Name Conformer Formula amine N-oxide H 3 NO methylamine N-oxide MeH 2 NO dimethylamine N-oxide Me 2 HNO ethylamine N-oxide anti EtH 2 NO-a gauche EtH 2 NO-g syn EtH 2 NO-0 trimethylamine N-oxide Me 3 NO ethylmethylamine N-oxide anti EtMeHNO-a gauche-1 EtMeHNO-g1 gauche-2 EtMeHNO-g2 isopropylamine N-oxide anti-gauche iPrH 2 NO-ag gauche-gauche iPrH 2 NO-gg propylamine N-oxide anti-anti nPrH 2 NO-aa anti-gauche nPrH 2 NO-ag gauche-gauche-1 nPrH 2 NO-gg1 gauche-gauche-2 nPrH 2 NO-gg2 tert-butylamine N-oxide tBuH 2 NO piperidine N-oxide axial aPNO equatorial ePNO phenylamine N-oxide PhH 2 NO methylphenylamine N-oxide PhMeHNO dimethylphenylamine N-oxide PhMe 2 NO diphenylamine N-oxide Ph 2 HNO methyldiphenylamine N-oxide Ph 2 MeNO triphenylamine N-oxide Ph 3 NO different ab initio methods. H 3 NO MeH 2 NO aPNO PhH 2 NO KT 53.5 22.6 6.66 7.33 HF 59.3 24.6 6.72 8.16 MP2 98.5 36.8 13.0 18.0 OO-MP2 123.9 54.7 29.0 47.6 CCSD 116.1 53.6 25.9 36.6 CCSD(T) 123.1 57.0 28.1 40.0 CCSDT 125.0 58.5 --CCSDT[Q] 124.4 58.1 --CCSDT(Q) 125.6 58.9 --EOM-CCSD(2) 121.1 60.3 29.5 43.1 EOM-CCSD 121.4 58.3 28.0 42.0 EOM-CCSD* 127.9 60.1 29.3 45.6 EOM-CCSD(T)(a)* 125.9 59.0 30.1 43.3 EOM-CCSD(T)(a) 128.1 60.9 --EOM-CCSDT 126.9 59.9 --TABLE III. Ab initio results and excluded volume descriptors. Table entries are ordered by molar mass and isomer. The polarizability is the isotropic value, and the excluded volume descriptors</p><p>x are expressed in terms of the respective "reference" volume for H 3 NO obtained with the same model.</p><p>x r + r + KT EOM-CCSD OO-MP2 MP2 MP2 model model model SCF EOM meV meV meV D &#197;3 V ref x V ref x V ref x &#197; &#197; H 3 NO 52.6 121.4 123.9 5.502 2.85 1.00 1.00 1.00 9.5 6.4 MeH 2 NO 23.0 58.2 54.7 5.370 4.67 1.92 2.42 1.59 14.6 9.1 Me 2 HNO 13.9 36.6 33.3 5.209 6.41 2.76 3.71 2.12 18.3 11.2 EtH 2 NO-a 16.7 46.9 45.4 5.708 6.38 4.06 5.82 2.15 17.8 10.6 EtH 2 NO-g 15.6 47.3 44.3 5.123 6.37 2.17 2.81 1.69 17.2 9.8 EtH 2 NO-0 15.4 48.9 45.4 5.031 6.30 2.30 3.02 1.70 17.0 9.6 Me 3 NO 9.3 29.1 28.1 5.058 8.08 3.51 4.88 2.59 21.1 12.2 EtMeH-a 11.4 32.1 29.9 5.436 8.10 4.75 6.89 2.58 17.8 10.6 EtMeH-g1 9.2 28.2 24.8 4.918 8.14 3.09 4.23 2.21 17.2 9.8 EtMeH-g2 9.9 31.3 28.8 4.964 8.05 3.11 4.27 2.22 17.0 9.6 iPrH 2 NO-ag 11.5 37.1 35.5 5.374 8.09 4.46 6.46 2.28 20.3 11.5 iPrH 2 NO-gg 11.6 46.6 45.4 4.918 8.06 2.69 3.64 1.81 19.2 9.8 nPrH 2 NO-aa 13.6 40.7 39.3 5.860 8.16 5.68 8.03 2.18 19.8 11.5 nPrH 2 NO-ag 14.2 42.1 42.1 5.935 8.09 6.10 8.95 2.44 19.7 11.5 nPrH 2 NO-gg1 13.9 48.3 45.8 4.977 8.09 2.38 3.14 1.72 17.9 9.7 nPrH 2 NO-gg2 11.4 40.4 38.3 4.971 8.05 2.76 3.74 1.79 19.5 10.5 tBuH 2 NO 8.6 34.6 34.2 5.092 9.78 4.87 7.11 2.41 22.1 11.6 aPNO 5.3 29.4 29.0 4.842 10.68 3.88 5.48 2.33 25.3 12.3 ePNO 11.5 36.2 36.8 5.855 10.79 7.56 11.28 2.98 21.4 12.4 PhH 2 NO 7.3 42.0 47.6 4.994 12.41 5.68 7.87 2.08 24.5 10.7 PhMeHNO 6.7 29.0 32.6 4.754 14.20 6.32 8.87 2.57 27.9 12.8 PhMe 2 NO 3.7 23.9 27.1 4.526 15.89 6.86 9.74 2.96 32.9 13.7 Ph 2 HNO 2.6 22.1 30.3 4.732 22.26 9.28 13.15 2.87 39.8 -Ph 2 MeNO 1.5 18.3 25.6 4.284 23.91 9.56 13.64 3.21 47.1 -Ph 3 NO 0.7 -20.9 3.971 32.15 12.07 17.27 3.39 69.1 -TABLE IV. Results from fitting QSPRs defined in eqs.( <ref type="formula">8</ref>), ( <ref type="formula">9</ref>) and (10) to the KT electron binding energies listed in Tab. III. The first entry in every table cell is the mean absolute errror (see eq.( <ref type="formula">12</ref>)).</p><p>The parameters are in units so that the dipole unit is D, the excluded volumes are measured on a scale relative to H 3 NO, and E b is in meV.  for &#181;, and our models V (i)</p><p>x (see eq.( <ref type="formula">2</ref>)) (1, 2, 3 shown in blue, orange, and green). As in Tab. III,</p><p>x is expressed relative to the respective values for H 3 NO.</p><p>5.0 5.5 6.0 &#956; [D]   <ref type="formula">15</ref>) and ( <ref type="formula">16</ref>)).</p></div></body>
		</text>
</TEI>
