Download as pdf or txt
Download as pdf or txt
You are on page 1of 23

ISSN 1990-7931, Russian Journal of Physical Chemistry B, 2007, Vol. 1, No. 4, pp. 307–329. © Pleiades Publishing, Ltd., 2007.

Chemisorption Energetics and Surface Reactivity:


UBI–QEP versus DFT Projections
E. Shustorovich
American Scientific Materials Technologies, Inc., 485 Madison Avenue, 24th Floor, New York, NY 10022, USA
E-mail: eshusto1@rochester.rr.com
Received November 12, 2006; in final form, December 21, 2006

Abstract—The wealth of information accumulated recently by ab initio/DFT calculations of adsorption ener-


getics and reactivity on metal surfaces makes it possible to systematically compare projections of the DFT and
UBI–QEP approaches. We make such comparisons covering both qualitative regularities in and the quantitative
accuracy of binding energies and activation barriers. We focus on the following areas: (1) DFT verification of
the UBI–QEP assumptions and rigorous projections concerning atomic and molecular binding energies at low
coverage; (2) coverage effects revealing in the M–A bond energy competition (metal sharing A–M–A) under
atomic co-adsorption; (3) reactivity patterns including both qualitative (periodic dependence of the activation
barriers, barrier vs. reaction enthalpy relationships, geometry of the transition state, etc.) and quantitative
aspects. We demonstrate the broad agreement between the DFT and UBI–QEP projections and point out the
areas where further testing by the ab initio/DFT techniques might be necessary and illuminating. In particu-
lar, while discussing the ab initio/DFT absolute values of the atomic binding energies QA and their periodic
trends, we revised our previous assumptions about monotonic changes of the QA values with the changing
atom A valence and along the VIII and IB group metals, where the periodic trends appear to be complicated.
Accordingly, we make corrections to the original UBI–QEP parameters, summarizing the current recom-
mended values of QA.
DOI: 10.1134/S1990793107040045

1. INTRODUCTION these projections and evaluations are falsifiable (in the


The recent years have seen an explosion of ab ini- Popperian sense) as well.
tio/DFT (Density Functional Theory) calculations of For the last two decades the BOC–MP/UBI–QEP
chemisorbed systems exploring adsorption energetics model has provided rather accurate estimates of chemi-
and metal surface reactivity (see recent reviews [1–4]). sorption energetics and surface reactivity, in broad
More complex systems have been treated and more agreement with experiment [9, 11–13]. The very fact
detailed and accurate information has been obtained. that a wide variety of phenomena could be interrelated,
But an interesting recent trend is that the calculated explicitly and quantitatively, within the same analytical
quantitative (numerical) results more and more often framework provides the major justification of the UBI–
are used to discern qualitative (periodic and other) reg- QEP model, certainly for its practical use. Still, until
ularities and project simple relationships between phys- the UBI–QEP basic assumptions are deduced from the
ical entities assumed to be the major factors [5–8]. first principles some intellectual dissatisfaction and
The phenomenological analytical UBI–QEP model practical uncertainty will remain. This is the major area
[9–11] (and its previous version as the BOC–MP model where accurate ab initio/DFT approaches might be
[12, 13]) allows one to calculate chemisorption energet- uniquely useful. In general, the phenomenological
ics (atomic and molecular binding energies and reac- UBI–QEP model and quantum mechanical ab ini-
tion activation barriers) for a wide variety of systems tio/DFT approaches are complementary and therefore
adsorbed on transition metal surfaces. The analytical synergetic in their applications to chemisorption phe-
UBI–QEP model is based on a few well-defined nomena. By and large, a good correlation and even
assumptions and provides explicit interrelations numerical closeness between ab initio/DFT and UBI–
between various entities, energetic and structural, by QEP projections has been found [9, 11]. Because the
minimization of total potential energy. Because these recent years have provided a wealth of DFT data it
entities are observable their accuracy can be directly would be informative to make further systematic com-
tested by experiment and/or by accurate ab initio/DFT parisons of the UBI–QEP versus DFT results to get a
calculations. Most importantly, the UBI–QEP model better understanding of the theoretical validity and
provides quantitative relationships, which allows one to practical usefulness of the UBI–QEP model.
directly project periodic and other regularities and eval- Before doing this, however, some clarifications
uate the relative importance of various factors. And might be necessary. The first principles ab initio calcu-

307
308 SHUSTOROVICH

lations, either DFT or wave function (HF/CI, etc.), are 2. THE UBI–QEP MODEL ASSUMPTIONS:
still approximate and as such they can be more or less A REMINDER
accurate [1–4]. These computational frameworks con-
tain various assumptions, approximations, and con- Let us recall the basis of the UBI–QEP model. The
straints, so their accuracy, depending on many factors, two-center M–A bond energy Q is assumed to depend
can hardly be foreseen a priori. In the present article, only on the M–A distance r, or on some function x(r) of
the ab initio calculated entities, if not stated otherwise, this distance, expressed in a polynomial form. The gen-
will be the DFT ones. The reader interested in the DFT eral potential E(x) has a two-term quadratic form [9]
technicalities is referred to the original articles. Here 2
we will mention only three most important factors: E ( x ) = – Q ( x ) = – Q 0 [ 2x 0 x – x ] (1)
(1) choice of the exchange-correlation functionals, 2
(2) (finite) cluster or (infinite) periodic slab modeling, with the only one energy minimum of Emin = – Q 0 x 0 at
and (3) frozen (fixed) or relaxed (variable) geometry of the equilibrium M–A distance r0 with the equilibrium
the system. Depending on the individual variations value of x(r0) = x0, and Q0 > 0 is a constant. The x(r)
within the computational frameworks used, results of function, called the bond index, BI, is assumed to be
calculations of the same entities may differ, often sig- exponential
nificantly (see below). Of course, if DFT values of the
molecular binding energies and the activation barriers x ( r ) = exp [ – ( r – r 0 )/a ], (2)
are very accurate, comparison with the relevant UBI–
QEP values would be a legitimate test of the UBI–QEP where a is some constant. The equilibrium value of
model accuracy/validity. However, if the DFT values x(r0) is x0 = 1, so Eq. (1) becomes
are not accurate enough (judged by reliable experimen- 2
tal data or by the spread of calculated values) such com- E ( x ) = – Q ( x ) = – Q 0 [ 2x – x ] (3)
parisons might be not informative. Anyway, once the with Emin = –Q0, where Q0 is the maximal (equilibrium)
UBI–QEP versus DFT data set is chosen it is always
prudent to compare those theoretical results with exper- M–A bond energy.
iment, particularly if there is disagreement between the For atomic A adsorption, there are three model
data. assumptions.
With those clarifications in mind, below we will 2.1. The two-center M–A potential is described by
provide representative UBI–QEP versus DFT compari- the quadratic exponential potential (QEP), Eq. (3).
sons, both conceptual/qualitative and numerical/quanti- 2.2. In the many-center Mn–A potential the total BI
tative. We will focus on the following areas: xA of all two-center contributions of the M(i)–A bonds is
1.1. DFT verification of the UBI–QEP assumptions conserved at (normalized to) unity
(a) rigorous quantitative UBI–QEP projections:
atomic binding energies at low coverage xA = ∑x iA = 1, i = 1, 2, …, n. (4)
(atop/bridge/hollow ratios, site preference, diffusion
barriers); The constraint expressed by Eq. (4) is called the unity
(b) the M–A bond lengths at different coordination bond index (UBI).
sites; 2.3. The number n is limited to the size of the regular
(c) the constants r0 and a in QEP. unit mesh Mn (the nearest neighbor approximation), say
n = 3 (equilateral triangle) for fcc(111) and hcp(0001),
1.2. Molecular adsorption at low coverage n = 4 (square) for fcc(100) or n = 5 (square pyramid) for
(a) rigorous qualitative UBI–QEP projections: con- bcc(100). Then, all M(i)–A bonds in Mn–A are equiva-
tact atoms, site insensitivity; lent and, therefore, the values of xi are fully determined
(b) molecular binding energies: quantitative aspects; by symmetry, namely xi = 1/n, regardless of the values
(c) UBI–QEP possible deficiency: methylidine CH. of two constants, r0 and a, which otherwise should
1.3. Atomic coadsorption and coverage effects somehow be determined (see below).
(a) M–A bond energy competition in metal atom The rest is straightforward algebra of minimizing
sharing A–M–A: coverage dependence; the total potential energy of the Mn–A interactions.
For atomic A adsorption within a symmetric Mn–A
(b) quantitative aspects. unit mesh, the UBI–QEP formalism rigorously
1.4. Reactivity projects that the Mn–A bond energy monotonically
(a) barrier vs. reaction enthalpy: “universal” linear increases with n as
relationships;
(b) geometry of the transition state: preferred sites Q A ( n ) = Q 0A ( 2 – 1/n ), (5)
for weakly and strongly bound coadsorbates; when the equilibrium M–A distance is
(c) UBI–QEP vs. DFT projections: quantitative
comparisons. r n = r 0 + a ln n. (6)

RUSSIAN JOURNAL OF PHYSICAL CHEMISTRY B Vol. 1 No. 4 2007


CHEMISORPTION ENERGETICS AND SURFACE REACTIVITY 309

Table 1. Binding energies Q of atoms and diffusion barriers Ediff on some fcc(111) surfaces: DFT and UBI–QEP data
Preferred site Q Ediff Ediff/Q
Metal surface Atom
calc. exp. PW91 RPBE PW91 RPBEP PW91 RPBE UBI–QEP
Rh(111) [18] O fcc fcc 4.88 4.31 0.45 0.41 0.09 0.10 0.10
N hcp … 5.31 4.87 0.68 0.64 0.13 0.13 0.10
S fcc fcc 5.42 4.95 0.46 0.46 0.08 0.09 0.10
C hcp … 7.11 6.65 0.60 0.56 0.08 0.08 0.10
Ir(111) [17] O fcc fcc 4.57 4.00 0.55 0.49 0.12 0.12 0.10
N fcc … 4.80 4.40 0.55 0.50 0.11 0.11 0.10
hcp … 4.80 4.40 0.55 0.50 0.11 0.11 0.10
S fcc fcc 5.11 4.67 0.58 0.57 0.11 0.12 0.10
C hcp … 6.71 6.29 0.49 0.47 0.07 0.07 0.10
Pt(111) [15] O fcc fcc 3.87 3.27 0.52 0.47 0.13 0.14 0.10
N fcc … 4.35 3.91 0.85 0.81 0.20 0.21 0.10
S fcc fcc 4.94 4.42 0.56 0.53 0.11 0.12 0.10
C fcc … 6.75 6.27 0.75 0.72 0.11 0.11 0.10
Note: DFT data for two exchange-correlation functionals, PW91 (self-consistent) and RPBEP (non-self-consistent). All energies are in
electronvolts.

3. AB INITIO/DFT VERIFICATION appears to be the hydrogen atom whose energy profiles


OF THE UBI–QEP ASSUMPTIONS are rather flat. Typically, hydrogen still prefers the hol-
low sites although the bridge and atop sites are close in
From Eq. (5), it follows that:
energy [14]. In particular, for the H/Pt(111) system,
(1) Within the Mn–A mesh, the hollow site (n = 3, 4, several DFT studies (e.g., [15, 16]) found practically no
or 5) is the only one global energy minimum. At zero preference for a coordination site, and in the DFT study
(low enough) atomic coverage, this highest coordina- of the adsorption system H/Ir(111) [17] the slightly pre-
tion site is always preferred. ferred hydrogen coordination was found to be atop (in
(2) The ratio of binding energies Q in the symmetric apparent agreement with the cited experimental data).
sites, atop (T), bridge (B), and hollow (H), is deter- Within the UBI–QEP model, deviation from the
mined only by the value of n, namely Q(T) = Q0, Q(B) = highest coordination might occur only if the UBI
(3/2)Q0, and Q(H) = (5/3), (7/4), and (9/5) Q0 for n = 3, (Eq. (4)) becomes incompatible with the QEP (Eq. (2)),
4, or 5, respectively, the Q(H)/Q(B) ratio being 1.11, which in turn violates the interrelations between the
1.17, and 1.20. hollow vs. bridge vs. atop binding energies, Eq. (5).
(3) Within the Mn–A mesh, the atop site (n = 1) is a This incompatibility can happen if the term (r – r0)/a is
maximum and the bridge site (n = 2) a saddle point so large that the equivalent M(i)–A bonds in the Mn–A
between two adjacent hollow site minima. The lowest- unit mesh would have the bond indices xi < 1/n. Intu-
energy diffusion of A to the adjacent Mn–A unit mesh itively, this might be expected if the A atom is too small
corresponds to the hollow-bridge-hollow path with the and the M–M bond lengths are too large (so that the dif-
barrier ference ri – r0 becomes too big). The smallest atom is
hydrogen, and the 5d metals have the largest M–M
∆E dif = γ n Q nA = Q nA ( n – 2 )/ ( 4n – 2 ). (7) bond lengths, which makes such H/M systems most
Thus, the atomic diffusion barrier ∆Edif is propor- probable candidates. This reasoning is consistent with
tional to the binding energy QnA with the proportional- the above DFT results revealing the rather flat energy
ity coefficient (corrugation ratio) of γn = (n – 2)/(4n – 2), profiles of the adsorbed hydrogen (particularly on Ir
which is γ3 = 0.10 for fcc(111) and hcp(0001), n = 3, and Pt surfaces), and it would be interesting to see if its
γ4 = 0.14 for fcc(100), n = 4, and γ5 = 0.17 for bcc(100), validity could be explicitly corroborated by DFT calcu-
n = 5. lations of the numerical values of xi , that is of the con-
stants r0 and a (see below).
The predictions concerning the preferred coordina-
tion sites are in full agreement with experiment [9, 12] For other atoms, typical examples and numerical
and DFT calculations. Specifically, at low coverage all comparisons of the DFT and UBI–QEP energetics are
adatoms occupy the hollow sites (of the highest coordi- given in Tables 1 and 2. Table 1 provides results of the
nation), which are distinctly more favorable than the systematic, state-of-the-art DFT calculations of binding
bridge and especially atop sites. The only exception energies QA and diffusion barriers Ediff for adsorption of

RUSSIAN JOURNAL OF PHYSICAL CHEMISTRY B Vol. 1 No. 4 2007


310 SHUSTOROVICH

Table 2. Atomic binding energies for the hollow H vs. bridge B coordination sites on some fcc(111), fcc(100), and bcc(100)
surfaces: DFT vs. UBI–QEP
DFT energiesa H/B ratio fcc(100)/(111) ratio
Adatom fcc bcc
H siteb B site DFT UBI–QEPc DFT UBI–QEPc
O Cu(111)d 96.9 86.5 1.12 1.11
Cu(100)k 114.0 95.5 1.19 1.17 1.18l 1.05
Ag(111)e 86.3
Ag(100)e 95.3 77.5 1.23 1.17 1.10 1.05
Ni(111)f 114.3 102.0 1.12 1.11
Ni(111)g 114.7
Ni(100)g 129.1 1.12 1.05
Ni(111)q 112.3
Ni(100)q 124.3 1.11q 1.05
Pt(111)i 75.7 63.0 1.20 1.11
Pt(111)h 98.3 83.7 1.17 1.11
Pt(111)m 89.3 77.3 1.16 1.11
75.4 64.6 1.17 1.11
Rh(111)o 112.6 102.2 1.10 1.11
99.4 90.0 1.10 1.11
Ir(111)n 105.4 92.7 1.14 1.11
92.3 81.0 1.14 1.11
N Pt(111)p 107.3 84.4 1.27 1.11
Pt(111)m 100.4 80.8 1.24 1.11
90.2 71.5 1.26 1.11
Rh(111)o 122.5 106.8 1.15 1.11
112.4 97.8 1.15 1.11
Ir(111)n 110.7 98.0 1.13 1.11
101.5 90.0 1.13 1.11
W(100)j 170 147 1.16 1.20
C Pt(111)m 155.7 138.4 1.12 1.11
144.6 128.0 1.13 1.11
Rh(111)o 164.0 150.2 1.09 1.11
153.4 140.5 1.09 1.11
Ir(111)n 154.8 143.3 1.08 1.11
145.1 134.3 1.08 1.11
S Pt(111)m 114.0 101.1 1.13 1.11
102.0 89.8 1.14 1.11
Rh(111)o 125.0 114.4 1.09 1.11
114.2 103.6 1.10 1.11
Ir(111)n 117.9 104.5 1.13 1.11
107.8 94.6 1.14 1.11
Notes: a Calculated at low coverages θ ≤ 1/n (ZCL values). b For all atoms, the energy minimum was found to be in the hollow sites (for
details, see Table 1 ). Double sets for O, N, C, and S atoms on Pt(111), Rh(111), and Ir(111) are DFT data for two exchange-cor-
relation functionals, PW91 and RPBE (see Table 1). All energies are in kcal/mol. c Equation (5). d [19]. e [20]. f [21]. gAb initio
HF/CI [22]. h [23]. i [24]. j [25]. k [26]. l This fcc(100)/(111) ratio corresponds to the values obtained in different DFT studies [19, 26]
and therefore is less informative. m [15]. n [17]. o [18]. p PW91 [27]. q Experimental calorimetric data [28].

RUSSIAN JOURNAL OF PHYSICAL CHEMISTRY B Vol. 1 No. 4 2007


CHEMISORPTION ENERGETICS AND SURFACE REACTIVITY 311

A = O, N, S, and C on fcc(111) surfaces of Rh [18], Ir can be used to estimate the “experimental” value of a
[17], and Pt [15]. [12]. The value of a can also be determined from fre-
The calculations were made at low coverage θ ≤ 1/n quencies ωn of vibrations normal to the surface for the
(which is the “zero coverage limit” (ZCL) in the UBI– Mn–A bonds, namely
QEP model) for two most popular DFT exchange-cor- 1/2
relation functionals, PW91 (self-consistent) and RPBE ω n = [ cos ψ n /a ] [ 2Q nA /µ ] , (9)
(non-self-consistent). For all atoms, the energy mini- where ψn is the angle between the M–A vector and the
mum was found in hollow sites, typically in the fcc site surface normal, and µ is the mass of an adatom A. Some
but sometimes in the hcp site (for carbon and nitrogen estimates of a from the relevant experimental data have
on Rh(111) and Ir(111)). From Table 1 one can see that been made (see [12, Table 8]). Those estimates are
the PW91 binding energies are systematically larger qualitatively consistent with the above UBI–QEP rela-
than the RPBE ones by ~0.5 eV (12–15 kcal/mol) but tionships but far from providing quantitative corrobora-
the diffusion barriers (differences between the binding tion. The major reason is that even small errors in the
energies in the hollow H and bridge B sites) are very experimental bond lengths translate into large uncer-
similar. This reflects a rather general situation that the tainties of a. For example, for the O–Ni bond lengths of
relative DFT values are more reliable than the absolute 1.9–2.0 Å, the typical error bounds ±0.05 Å (<3%)
ones. Accordingly, Table 2 provides results of various result in the uncertainty of ±0.2 Å (35–50% of the esti-
state-of-the-art DFT calculations of atomic binding mated value of a) [12].
energies Q for the H versus B coordination sites on
some fcc(111), fcc(100), and bcc(100) surfaces. The In principle, the UBI–QEP constants, r0 and a, can
relevant UBI–QEP values are given for comparison. be determined by ab initio/DFT calculations along the
lines outlined above. In particular, Eq. (6) can be
One should mention that the UBI–QEP model (in its explicitly tested by calculating bond lengths rn in all
nearest neighbor approximation) does not distinguish symmetric sites, atop T, bridge B, and hollow H, which
in the fcc(111) surface two adjacent hollow sites, fcc would give both r0 and a. However, the calculations
and hcp (differing by the lack or the presence of a metal should have the accuracy adequate for the testing task.
atom in the second layer below the M3–A hollow site). At present this testing appears to be unrealistic but it
The ab initio/DFT calculations do distinguish the fcc might be possible in the future.
and hcp hollow sites, usually finding the fcc site pre-
ferred although the hcp site is comparable and some-
times preferred (for atoms of the highest valency such 3.2. Atomic Binding Energies
as N and C, as seen from Table 1). Because the fcc-hcp General comments. The atomic binding energies QA
energy difference is small being typically of the order constitute the basic UBI–QEP parameters which are
of few per cent of QnA (see, for example, [15, 17, 18]), taken from external sources—experiment (if available)
it practically does not affect the UBI–QEP projections or ab initio/DFT calculations. Reliable experimental
for fcc(111) surfaces. It is worth mentioning that values of QA exist for H and O on many surfaces and for
because the hollow sites are often the only true energy N on some surfaces. For C, the experimental values of
minima the DFT energies of the top and bridge sites are QC were reported only for two Ni surfaces, fcc(111) and
obtained by introducing some additional constraints (100) [29, 30], but there are serious doubts about their
(see, for example [27]). With all these in mind, the accuracy (see, e.g., [31]). Thus, most of QN and all QC
agreement between the UBI–QEP and DFT values is are obtained from theoretical calculations and/or some
excellent. extra/interpolations.
Some clarifications are worthy. For practical rea-
3.1. Determination of the Constants r0 and a sons, the major source of adsorbed atoms A is A2
(and/or AB) molecules which should be available in the
Within the UBI–QEP model, Eqs. (3)–(7) are exact, gas phase. This is the case for H, O, and N but not for
so their accuracy is the most informative test of the C. Furthermore, direct and accurate measurements of
model’s validity. One should stress that the rigorous QA can be made mainly for metal surfaces which disso-
consequences of the energy relation Eq. (5) are deter- ciate A2 molecules without or with slight activation.
mined by symmetry and therefore independent of the These conditions are fulfilled for H2 and O2 on most
constants r0 and a which define the A–M bond index, metals and for N2 on active metals of the V and VI
Eq. (2), and determine changes of the A–M bond length groups, so one can at least extrapolate the experimental
with n, Eq. (6). Because only the hollow sites are exper- values of QN to the VIII group metals. At practical con-
imentally observed the constant r0 cannot be experi- ditions, carbon forms not C2 but graphite. The reported
mentally verified but the constant a can. In particular, experimental values of QC on Ni(111) and (100),
the relation r3 versus r4 for the hollow sites in the namely QC = 171 kcal/mol [29, 30], were suspiciously
fcc(111) M3–A and fcc(100) M4–A unit meshes close to the bond energy of graphite, and in the same
work a smaller estimate of QC ≤ 160 kcal/mol was sug-
r 4 – r 3 = a ln ( 4/3 ) = 0.287a (8) gested [30].

RUSSIAN JOURNAL OF PHYSICAL CHEMISTRY B Vol. 1 No. 4 2007


312 SHUSTOROVICH

Table 3. Selected ab initio/DFT values QO: mutual compar- are best defined for the fcc(111) and hcp(0001) sur-
isons and experimenta faces. Based on the experimental and theoretical data
accepted at the time (see [13, Table 1]), the UBI–QEP
Ab initio/DFT
Experi- values of QA reflected two major periodic regularities:
Surface UBI–QEPa
individual
range menta (1) For a given metal surface, the absolute values of
values QA increase with an increase of the atomic valence
Rh(111) 99.4–112.6b 100–127 102 102 along the period, in particular (mono-H<) di-O < tri-N <
i ii tetra-C.
111.9 113.0 88–102dd
d (2) The values of QA decrease monotonically from
120.4 left to right along the period/row and from top to bot-
116–127bb tom along the column.
Ir(111) 92.3–105.4c 92–105 93 93 These regularities appear to be general enough for
f 112h
Ni(111) 114.3 110–137 115 the VIII group metals whereas for the IB metals the
110.0–124.1 ff 115 periodic trends might be more complicated (see below).
114.7 g Some of the recommended UBI–QEP numerical values
of QA have been changed when the more accurate esti-
125.0i mates became available. The last time these changes
137.0y were made in 2003 (see [32, Table 4]). Now we will dis-
Pd(111) 80.5–94.1ff 80–102 86hh 87 cuss whether the current state of ab initio/DFT calcula-
95.6xx 95.7ii 87
tions makes it necessary and possible to make further
changes/corrections.
102.4i
There are many ab initio/DFT techniques, and they
Pt(111) 75.7k 75–102 85 85 vary, often significantly, in their accuracy, so there is
75.4–89.3m 96z always a range of ab initio/DFT values of QA. Cluster-
83.3yy 87z type calculations of the chemisorption energetics usu-
98.3 l ally are less reliable because they are very dependent on
the cluster size/structure (cf. the accompanying article
101.3x by Gomes et al. [4a]). This size dependence produces
102.2j artifacts due to boundary effects and makes it particu-
Cu(111) 96.9n 97–116 103 103 larly difficult to evaluate surface relaxation effects
99.9 i (often, large). In principle, periodic slab calculations
are preferred but their accuracy depends on how well
116.0x they treat many influencing factors. Not surprisingly,
Ag(111) 72.2i 73.8jj 72–86 80 80 more accurate calculations, being more comprehensive,
80.7zz are more expensive, and practitioners should strike
86.3e affordable balance. In particular, calculations vary in
considering the following major factors:
Notes: a All energies are in kcal/mol. The experimental values and (1) The number of metal layers and surface relax-
references, if not shown otherwise, are from [13, Table 1].
b The RPBE–PW91 range [18]. bb Other DFT values cited ation.
in [18]. c The RPBE–PW91 range [17]. d [38]. dd Other These factors are interrelated. The minimal number
experimental values cited in [38]. e [20]. f [21]. ff The of metal layers is two but slab calculations usually
RPBE–PBE–PW91 range [41]. g Ab initio HF/CI: [22]. employ from three to five (and more) layers. The more
h [28]. hh [42]. i [43]. ii [59]. j [44]. k [24]. l [23]. m The metal layers are considered the more accurate is treat-
RPBE–PW91 range [15]. n [19]. x [45]. y [46]. z The initial ment of surface relaxation, which might be significant,
(θ ~ 0) and average (for θ = 0–0.33) values, respectively especially for more open surfaces and strongly bound
[36]. See text. xx [47]. yy From Q * adsorbates, affecting the binding energies (see, e.g.,
O = 23.8 kcal/mol [48] for
the calculated gas-phase value D(O2) = 130.4 kcal/mol [49]. [27] and references therein).
zz Average value for three coverages in the range of θ = (2) The size and structure of a periodic unit cell
0.11–0.33 (θ ≤ 1/n) [35]. jj [50]. (adsorbate coverage).
Smaller unit cells require smaller computing time
and cost but, by definition, they correspond to higher
In this article we will mainly concentrate on the coverage (the adatom/metal atom ratio). For example,
chemisorption energetics on fcc(111) and hcp(0001) the smallest unit cell A-p(1 × 1), having the size of the
surfaces (having the same M3 unit mesh). They are surface unit mesh Mn, corresponds to θ = 1 ML but cov-
close packed and the least prone to lattice changes after erages of practical interest are usually much lower, typ-
chemisorption. For this reason the experimental bind- ically smaller than θ = 0.3–0.5 (higher coverages often
ing energies and, therefore, the UBI–QEP values of QA cannot be realized and/or adatoms form some under-

RUSSIAN JOURNAL OF PHYSICAL CHEMISTRY B Vol. 1 No. 4 2007


CHEMISORPTION ENERGETICS AND SURFACE REACTIVITY 313

surface structures (see, e.g., [33]). In particular, on Table 4. Selected ab initio/DFT values (kcal/mol) of QN
fcc(111) surfaces the coverages θ = 0.50, 0.33, 0.25, and QC: mutual comparisons and experiment
0.17, and 0.11 are realized within the rapidly increasing
Ab initio/DFT
unit cells (2 × 1) < ( 3 × 3 )R30° < (2 × 2) < (3 × 2) < Surface UBI–QEPa
(3 × 3), respectively. In general, a given coverage can be individual values range
realized by different atomic A configurations when Atom N
some atoms may be non-equivalent. Such coverage
Rh(111) 112.4–122.5b 112–122 116h
variations are accompanied by changes in the binding q
energies QA, so the choice of coverage and the unit cell 106.6
is of much importance for comparisons of various DFT Ir(111) 101.5–110.7b 102–111 127
calculations among themselves and experiment. Ni(111) 135
(3) The choice of the DFT functionals. Pd(111) 90.7q 91 130
Pt(111) 90.2–100.4 b 90–109 102–116h
There are many of them (see [34] and references
therein), and calculations were often made with only 107.3c 108.9d
one (this or that) functional. Two most popular DFT Ag(111) 7.6p
exchange-correlation functionals are PW91 (self-con- Atom C
sistent) and RPBEP (non-self-consistent). The DFT- o
Ru(0001) 165.1 165–175
PW91 calculations usually overestimate the binding
energies but the DFT-RPBE calculations underestimate 174.6o
them, as seen from Tables 1–3. Co(0001) 152.7e 153–163
(4) An account of the vibrational zero-point energy. 158.7f
162.6f
Most DFT calculations neglect this energy. When
calculated, sometimes it was indeed small but some- Rh(111) 153.4–614.0b 153–168
165.0 m
times it was found to contribute strongly to binding
energies of both atoms and molecules and therefore to 168.5m
the reaction activation barriers. A recent example of the Ir(111) 145.1–154.8b 145–155
zero-point energy effects is DFT calculations of
adsorption of N, O, H atoms and molecules formed by Ni(111) 154.1f 135–155 171
146.5 i
those atoms on Pt(111) [27].
134.0j
It might be useful to clarify the interrelation
between the DFT and UBI–QEP approaches to the QA 156.0l
coverage dependence. In general, the values of QA Pd(111) 151.8g 148–152 160
decreases with coverage, and DFT calculations univer- 147.5 n
sally reveal this decrease, which is caused by the M–A
Pt(111) 144.6–155.7b 145–157 150
bond competition (A–M–A metal sharing). On a sur- k
face with a unit mesh Mn–A for the coverage range of 157
0 < θ ≤ 1/n, there is no co-adsorption in the adjacent Ag(111) 36.9p
Mn–A meshes and therefore, within the nearest neigh-
bors approximation, there is no metal sharing. Accord- Notes: a All the UBI–QEP values are from [13, Table 1], if not
ingly, in the UBI–QEP model, where binding is limited stated otherwise. b From Table 2. c [27]. d [45]. e [51]. f [31,
to nearest neighbors, the atomic binding energy QA for 52]. g [47]. h [53]. i [54]. j [55]. k Evaluated from Fig. 4 in
zero coverage remains constant for coverages up to [16]. l [46]. m The two values for θ = 0.25 and 0.11, respec-
θ ≤ 1/n. However, the DFT calculations include all tively [57]. n [58]. o The two values are for θ = 0.25
neighbors and therefore show the decrease in QA and 0.11, respectively [57, 75]. p [50]. q [60].
already within the range 0 < θ ≤ 1/n. For example, when
the coverage of carbon increased from θ = 0.11 to 0.25,
a decrease in QC was found to be 4 kcal/mol on As said, the zero-coverage UBI–QEP parameter
fccRh(111) but 10 kcal/mol on hcpRu(0001) (see QA(θ = 0) is assumed to be constant within the range
Table 6 below). Another example is the system 0 < θ ≤ 1/n. If QA actually does decrease, the UBI–QEP
O/fccAg(111) where periodic slab DFT calculations QA(θ = 0) should be taken as an average of QA(θ) for
[35] found a monotonic decrease of QO in the range of the range of 0 < θ ≤ 1/n. The same proposition holds
θ = 0.11, 0.25, 0.33, namely QO = 83.3, 81.2, and true for the experimental binding energies if they are
78.0 kcal/mol, respectively. Most periodic DFT calcu- taken as the UBI–QEP parameters QA. However, unlike
lations were made for only one coverage, usually θ = theoretical calculations for an ideal surface, the experi-
0.25 but often for θ = 0.50 or 1.00, which should be mental situation may be more complicated. At very low
taken into account in comparisons. (zero) coverage the measured value of QA may be

RUSSIAN JOURNAL OF PHYSICAL CHEMISTRY B Vol. 1 No. 4 2007


314 SHUSTOROVICH

Table 5. The UBI–QEP parameters QA for the VIII and IB metalsa


Adatom Co Rh Ir Ni Pd Pt Cu Ag Au
Hb ≥63 61 58 63 62 61 56 52 ≤52
Ob ≥115 102 93 115 87 85 103 80 ≤75
Nc >135 116 ≥110 (127) 135 ≥110 (130) 110 – – –
Cd 160 160 155 155 155 155 – – –
Notes: a The values QA (in kcal/mol) are for the metal surfaces (hcp(0001) for Co and fcc(111) for the rest) having the three-fold unit
meshes M3–A. The standard (two-center) UBI–QEP parameter Q0A can be obtained as Q0A = 0.6 QA (Eq. (5)).
b All numbers are the experimental values (taken from [13, Table 1]) with the exception of Co, which are extrapolations. The aver-
aged DFT values (see [14] for QH and Table 3 for QO) are in good agreement with these experimental data.
c The DFT values from Table 4 averaged to be in reasonable agreement with the experimental data [13, Table 1] and to show the
conventional periodic trends. The only exception is Ir and Pd, where the differences are large and therefore we suggest two sets
of values, DFT and experimental (in parentheses). See text.
d The averaged DFT values from Table 4 showing unconventional periodic trends. See text.

affected by inevitable surface defects having a more was chosen as the UBI–QEP value of QO on Pt(111)
open surface and therefore the higher QA. For this rea- (see Table 3). For most systems, however, such
son the initial value of QA may be (artificially) overes- (detailed) coverage data are lacking, which leads to
timated, and the following decline in QA may be (arti- uncertainties in choosing both theoretical and experi-
ficially) much steeper. A probable example is a micro- mental values of QA.
calorimetric study of oxygen adsorption on Pt(111) For these reasons it is not always clear which DFT
[36]. The initial binding energy was measured to be values can be compared with one another and with the
QO ~ 100 kcal/mol but it dropped precipitously to UBI–QEP values and experiment. The systematic anal-
85 kcal/mol at θ = 0.11 and slower to 75 kcal/mol at ysis of the ab initio/DFT calculations of QA is beyond
θ = 0.33. The initial (zero-coverage) value QO(θ = 0) ~ the scope of the present work. Here we will provide
100 kcal/mol, if taken for Pt(111), appears to be only some examples, hopefully representative of the
strongly overestimated but QO(θ = 0.11) ~ 85 kcal/mol state of affairs.
or the average value QO(θ = 0–0.33) ~ 87 kcal/mol are
in good agreement with the generally accepted TDS Numerical values. The experimental range of the
experimental value of QO ~ 85 kcal/mol [37], which hydrogen binding energies for the VIII group metals,
particularly Co–Ir and Ni–Pt, is narrow being QH = 60–
65 kcal/mol [13, Table 1]. The majority of ab ini-
Table 6. Selected DFT versus UBI–QEP binding energies tio/DFT calculations reproduce the absolute values of
a
Q CH3 O QH rather well. However, because the differences in QH
are very small the periodic regularities are not clear
Metal DFT Q CH3 O UBI–QEPb [14]. Also, the accuracy of calculated values of QA for
surface individual range Q CH3 O QO hydrogen does not guarantee the accuracy of QA for
other atoms. Indeed, as seen below, for a given metal
Rh(111) 52c 52 54 102 surface, the calculated values of QA for A = O, N, and
Ni(111) 62c 62 65 115 C usually show a wide spread.
Pd(111) 38h 39–47 43 87 Let us start with oxygen. Experimental low-cover-
39d age values of the oxygen binding energies QO for the
41e
47c VIII group metals cover a wide range of QO = 80–
Pt(111) 34c 34–39 41 85
120 kcal/mol (see [13, Table 1]). Although DFT calcu-
36f lations of oxygen chemisorption are plentiful, only a
39e part of them is suitable for our purpose. First, we are
Cu(111) 48g 48–56 55 103
comparing only the O binding energies at low coverage,
51c θ ≤ 1/n, so we exclude DFT calculations which use unit
56d cells corresponding to high coverage, θ > 1/n. Second,
Ag(111) 44c 44 38 80
in DFT calculations of O/M systems the O binding
energies are presented either as the absolute value QO or
Notes: a All energies are in kcal/mol. b The values of QCH O were
3 as the relative value Q O* = QO – (1/2)D(O2) referred to
calculated from Eq. (10) for k = 1, DAB = 90 kcal/mol (see
text). The atomic binding energies QO are taken from exper- the gas-phase bond energy D(O2) ( Q O* is the heat of
iment (see Table 3). c [67]. d [68]. e [69]. f [70]. g [71]. h [72]. dissociative adsorption O2, gas 2Oads per atom).

RUSSIAN JOURNAL OF PHYSICAL CHEMISTRY B Vol. 1 No. 4 2007


CHEMISORPTION ENERGETICS AND SURFACE REACTIVITY 315

These values are uniquely interrelated only if the calcu- Table 7. CH3 binding energies: UBI–QEP vs. DFTa
lated value of D(O2) is known. Many DFT calculations,
Metal surface DFTb DFTc UBI–QEPd
however, provide only the values of Q O* without D(O2),
which prevents determination of QO. The fact is that the Ni(111) 47 25–50 48 (40)
DFT values of D(O2) vary strongly, often being in the Pd(111) 41 36–39 42 (40)
range of 130–145 kcal/mol (see, for example [19, 35, Pt(111) 30 40 38 (40)
38, 39]). Therefore the substitution of the DFT D(O2) Rh(111) 41 40–45 ≥42
by the experimental value of D(O2) = 119 kcal/mol [40] Notes: a All energies are in kcal/mol. b Periodic DFT slab calcula-
might be a poor approximation. tions [63]. c Range of other ab initio/DFT values (periodic
and cluster-type) cited in [63]. d From [13, Table 8]. The val-
For further certainty we will limit our comparisons ues in parentheses are for the modified QC = 155 kcal/mol for
to oxygen adsorption on fcc(111) surfaces for which Ni(111), Pd(111), and Pt(111) (see above).
the experimental binding energies and, therefore, the
UBI–QEP values of QO are best determined (the close
packed fcc(111) surfaces are the least prone to surface Table 8. CH2 binding energies: UBI–QEP vs. ab initio/DFTa
reorganization after chemisorption). These selected
data are shown in Table 3 and compared with the rele- Metal DFT
vant (fortunately, still plentiful) DFT values. θ UBI–QEP
surface periodic cluster
Not surprisingly, the more numerous and diverse are
the DFT calculations (using different functionals and Ni(111) 0.25 75b 67–88c 83d (71)
approximations as well as different coverages) the Pt(111) 0.25 92e 68d (71)
wider is the calculated QO range, often comprising tens Rh(111) 0.25 100f 78h
of kcal/mol. This makes agreement with experimental 0.11 101f 81h
data (which usually also form a range) and an evalua-
tion of the DFT accuracy uncertain. For example, as Ru(0001) 0.25 100g 78h
seen from Table 3, the DFT values QO for Pt(111) are 0.11 103g 85h
spread within a range of 75–102 kcal/mol, which is 0.25 99i 85h
much wider than the reported experimental range is 85– Co(0001) 0.25 92i 77h
96 kcal/mol. Actually, the latter may be even smaller,
namely QO = 85–87 kcal/mol, because the cited QO = Notes: a All energies are in kcal/mol. b [74]. c Cluster-type ab ini-
96 kcal/mol is the initial (zero coverage) value which tio/DFT values (their range) cited in [74]. d From [13, Table 8].
might be overestimated (see above). If so, many DFT The value in parentheses is for the modified QC =
values of QO on Pt(111) are not accurate. Another 155 kcal/mol (see above). e Evaluated from [16, Fig. 4].
example is O/Ni(111), where two ab initio calculations, f [57]. g [75]. h The UBI–QEP values of Q
CH 2 are calcu-
DFT [21] and HF/CI [22], are consistent with the nar- lated for the DFT values of QC on those surfaces (see Table 8
row experimental range of QO = 112–115 kcal/mol but in the text). i [76].
two other ab initio DFT calculations, predicting QO =
125 kcal/mol [43] and even 137 kcal/mol [46] are wide
off the mark (see Table 3). One should add that various compared with the assumed UBI–QEP values of QN
DFT calculations may produce not only a spread of the and QC based on experimental data existed at the time and
absolute values of QA but also the different order in the some extrapolations of those data (see [13, Table 1]).
periodic series, which makes specific projections for a
given metal surface problematic. For example, as seen A serious complication is that the nature and
from Table 3, most DFT studies find, in agreement with strength of the metal-adsorbate bonding differ greatly
experiment, that the oxygen binding energy on Cu(111) for the d-band and sp-band metals, so that the regulari-
is distinctly larger than that on Pd(111). However, in the ties found for the VIII group (Co, Rh, Ir, Ni, Pd, Pt)
DFT calculations by Wang et al. [43] the opposite may be different for the IB group (Cu, Ag, Au). From
order Pd > Cu was found (because of a strong overes- Tables 3 and 4 we see that for all the VIII group metals
timation of QO on Pd which was calculated to be over the binding energy QA order follows the atomic
102 kcal/mol whereas the experimental value is QO = valences, namely O < N < C. Until recently, the com-
86–87 kcal/mol), which makes the theoretical projec- mon assumption was that for the IB metals the order
tions for Cu and Pd surfaces and comparisons with should be the same, which determined the choice of the
experiment not convincing. UBI–QEP atomic parameters QA [9–13]. However, this
The accuracy of DFT values QO might give an idea assumption might need correction. A conspicuous
about the probable accuracy of the calculated values of example is the QA order on Ag(111) found in the most
QN and QC, where experimental values are unavailable. recent state-of-the-art ab initio calculations [50]. These
Some data for fcc(111) and hcp(0001) surfaces (having calculations were cluster-type but the cluster size
the same M3 unit mesh) are given in Table 4, which are appears to be large enough (three layers of 91 atoms) to

RUSSIAN JOURNAL OF PHYSICAL CHEMISTRY B Vol. 1 No. 4 2007


316 SHUSTOROVICH

provide reliable estimates. Although for all atoms the (111) [29, 30] is overestimated, at least by 15 kcal/mol.
preferred coordination site is the fcc hollow, the QA This will be consistent with an alternative experimental
order did not follow the atomic valences, the maximum estimate of QC ≤ 160 kcal/mol [30].
being for O (74 kcal/mol), the minimum for N
(8 kcal/mol) and C in-between (37 kcal/mol). The sug- The recommended UBI–QEP numerical values of
gested reason was that the sp-metal band of Ag could QA may and should change with time when the more
not fully utilize the adsorbate unpaired electrons (a accurate estimates become available (see the last UBI–
maximum of two metal electrons were shared). This is QEP set [32, Table 4]). From said above, it would be
fundamentally different from the metal d-band which prudent to consider separately the parameters for the
can utilize more or even all the adsorbate unpaired elec- VIII and IB metals.
trons, as has been illustrated by comparing the electron (1) For the VIII metals, for the carbon binding ener-
distributions in the similar calculations for Pt(111) gies QC now we assume the periodic series Co ≥ Rh > Ir
[50]. Below we will also see that the VIII and IB metals and Co > Ni ~ Pd ~ Pt. Specifically, we recommend to
might show the reversed trends in the binding energies take the same value of QC = 155 kcal/mol for Ni(111),
of C versus CH and N versus NH. Pd(111), and Pt(111), compared to the previous UBI–
Unfortunately, unlike the systematic DFT calcula- QEP values of QC = 171, 160, and 150 kcal/mol,
tions for the VIII group metals, the ab initio/DFT calcu- respectively [9, 11, 13]. The changes are slight, how-
lations for the IB metals are scarce, mostly limited to ever, and do not warrant re-calculations of the UBI–
oxygen adsorption (see Tables 3 and 4), which shows QEP studies having used the previous QC parameters.
the common periodic trends. But from the discussed For QN, only few DFT versus UBI–QEP comparisons
calculations on Ag(111) [50] it appears that for N and can be made. The projections are mutually consistent
C adsorption the periodic QA regularities for the IB with the exception of Ir and Pd, namely the (conven-
metals might be different. With no hope for experimen- tional) DFT periodic order Rh > Ir [17, 18] and Rh > Pd
tal data, the relevant QN and QC parameters are in urgent [59] versus the reversed UBI–QEP order Rh < Ir and Rh
need of accurate ab initio determination. < Pd, based on the experimental data (see [13, Table 1]),
Within the VIII group metals, as seen from Tables 3 albeit of the uncertain accuracy. Because there were no
and 4, the periodic trends in the calculated values of QA UBI–QEP studies involving these (experimental) QN
are as follows: parameters for Ir and Pd, their accuracy has not been
tested. So, as an exception we suggest for Ir and Pd two
(1) For a given metal surface, the absolute values of sets of the QN values, old experimental and new DFT
QA increase with an increase of the atomic valence ones. It would be interesting to see in the future which
along the period. This increase is best documented for set would perform better.
the series mono-H < di-O < tri-N < tetra-C. (Please note
that the cross-period comparisons are not valid. For (2) For the IB metals, only the oxygen QO parame-
example, divalent sulfur is stronger bound than not only ters are reliable. The nitrogen QN and carbon QC param-
divalent oxygen but also trivalent nitrogen.) eters should wait for their determination.
(2) For a given atom A, the periodic trends are often The current recommended UBI–QEP parameters QA
obscured because the calculated ranges of QA of neigh- for the VIII and IB metals are summarized in Table 5.
boring metal atoms usually overlap. For H, the periodic
changes of QH are small (few kcal/mol) [14] and there-
fore insignificant, whether they are monotonic or not.
For oxygen, the values of QO distinctly decrease from 4. MOLECULAR ADSORPTION
left to right along the period/row and from top to bot- Compared to the UBI–QEP atomic formalism, the
tom along the column. For nitrogen, by and large the molecular formalism includes additional assump-
periodic trends are similar although sometimes it is dif- tions/approximations which makes direct verification
ficult to say whether this periodic decrease is mono- by ab initio/DFT calculations less transparent. Indi-
tonic or not. For carbon, the periodic trends appear to rectly, the UBI–QEP validity can be tested along two
be more complex, in particular suggesting no change or lines: (1) projected qualitative regularities of QAB, and
even a slight increase in the series Ni ≤ Pd ≤ Pt [4b]. (2) the numerical accuracy of values of QAB. Consider
Compared to oxygen, the relevant DFT databases first the projected qualitative UBI–QEP vs. DFT regu-
for nitrogen and carbon are not only more limited but larities.
also the QA ranges strongly overlap, with all uncertain-
ties in comparing (the accuracy of) different DFT cal-
culations. Nevertheless, one can distinguish some peri- 4.1. Major Qualitative Regularities
odic QA trends, in particular Pt < Ir < Rh for N or Ru > of Molecular Binding Energies
Rh > Pd, Rh > Ir and Ni ~ Pd ~ Pt for C (see Table 4).
As far as the absolute values of QA are concerned, one Mono-coordination (AB)–Mn. The molecular AB
computational result appears to be unambiguous: the binding energies QAB depend on (details of) the coordi-
putative experimental value of QC = 171 kcal/mol on Ni nation (AB)–Mn mode [9, 13]. For mono-coordination

RUSSIAN JOURNAL OF PHYSICAL CHEMISTRY B Vol. 1 No. 4 2007


CHEMISORPTION ENERGETICS AND SURFACE REACTIVITY 317

via A the analytic expressions for QAB have a general the typical contact atom A is oxygen, in the second
form case, carbon.
2 Let us start with the O coordination. In order to get
Q AB = Q A / ( k D AB + Q A ), (10)
the most unambiguous results, consider the simplest
where DAB is the gas-phase A–B binding energy (con- case of mono-O-coordination of radicals OX. The sim-
stant), k is some numerical coefficient depending on the plest radical, hydroxyl OH, is known to form hydrogen
type of bonding (strong or weak) [9, 13]. Equation (10) bonds which might strongly contribute to the value QOH
rigorously projects that of two possible contact atoms, (see discussion in [9]). (Hydrogen bonds already affect
A and B, A will be preferred if it has a larger atomic the DFT binding energies of H2O (see, e.g., [66]). The
binding energy, QA > QB. Indeed, all DFT studies cor- next best case for a model analysis is methoxy CH3O,
roborate this regularity always finding mono-coordi- whose adsorption has been extensively studied. The
nated CO and CN via C, NO via N, etc. (see, for exam- binding energies Q CH3 O obtained by periodic DFT cal-
ple [2, 3] and references therein). culations on various fcc(111) metal surfaces (individual
Insensitivity of QAB to coordination sites. Atoms values and their ranges) are shown in Table 6, where
have a strong preference for the highest coordination they are compared with the UBI–QEP values. The latter
sites (see Eq. (5)). Molecular radicals AB will show the were calculated for the H3C–O bond energy of DC–O =
stronger preference for higher coordination the more D CH3 O – D CH3 = 383–293 = 90 kcal/mol (see [13,
the contact atom A retains its free-atomic features
(unpaired electrons are substantially localized on A). Table 8]).
This is the case of the strong molecular bonding [9, 13]. Given the variety of the DFT calculations, the agree-
If AB is a closed-shell molecule (such as CO) or a rad- ment with the UBI–QEP values is remarkable. This is
ical with delocalized unpaired electrons (such as NO) another illustration of the well documented conclusion
or just one localized electron (such as CH3 or NH2), the [9, 11, 13] that the more accurate are the atomic binding
molecular bonding is weaker, and the UBI–QEP for- energies QA (here oxygen QO) the more accurate are the
malism [9, 13] project relative insensitivity of QAB to a molecular binding energies QAB, which testifies to the
coordination site. soundness of the UBI–QEP model.
Indeed, DFT studies find such insensitivity for many The second case (the lack of experimental data on
molecules when the preferred coordination sites might QA of the contact atom) includes many N-coordinated
be hollow, bridge and/or atop, the energy differences and all C-coordinated species. For a model analysis, the
being rather small, which often makes it difficult to cor- simplest species, which are also the most studied, are
rectly determine the preferred coordination site. The mono-C-coordinated CHx fragments. The binding ener-
example, which became a “cause celebre,” is CO on gies of CH3, CH2, and CH on selected fcc(111) surfaces
Pt(111), when all conventional DFT studies incorrectly calculated by the DFT (both periodic and cluster-type)
assigned CO to the hollow site (experimentally, the pre- and UBI–QEP methods are shown in Tables 7, 8, and 9,
ferred site is atop), and only a few specially designed respectively.
DFT calculations could make the correct assignment
and reproduce the CO binding energy (see a detailed The UBI–QEP values of QAB for the mono-A-coor-
discussion and references in [11]). Other examples are dinated molecule AB follow the same order as the val-
DFT studies of NO (see, e.g., [59–62] and references ues of QA (cf. Eq. (10)). The DFT calculations typically
therein) and CH3 (see, e.g., [57, 63–65] and references show the same patterns. Because the periodic changes
therein), where the preferred coordination sites was for the atomic carbon binding energy QC are compli-
found to vary depending on a metal surface, coverage, cated (see above), the DFT values of Q CH3 show a com-
and other factors. plicated picture as well. For example, from Table 4 one
can deduce the Q CH3 order Ni > Pd > Pt or Ni ~ Pd ~ Pt.
4.2. Molecular UBI–QEP Versus DFT Binding Energies: (Moreover, there is one cluster-type DFT study [73]
Numerical Comparisons which projects the periodic Q CH3 relations Pt > Pd and
The numerical accuracy of the UBI–QEP molecular Ir > Rh.) Coming back to the UBI–QEP projections, for
binding energies has already been tested for a wide the assumed values of QC decreasing from 171 to 160
variety of adsorbed species and surface reactions by to 150 kcal/mol along the series Ni > Pd > Pt [13], the
comparing with experimental data (see, e.g., the
reviews [9, 11, 13]). So, below we will focus on com- values of Q CH3 decrease the same way from 48 to 42 to
paring the UBI–QEP and DFT results for entities for 38 kcal/mol, in good agreement, even quantitatively,
which there are no (at all or accurate enough) experi- with the periodic DFT calculations giving Q CH3 = 47,
mental data. All molecular species are coordinated via 41, and 30 kcal/mol, respectively [63]. If we assume
some contact atoms, and we will distinguish situations that there is no change in QC = 155 kcal/mol along the
whether the relevant atomic binding energies QA are
known or not known from experiment. In the first case series Ni ~ Pd ~ Pt, the values of Q CH3 will be the same,

RUSSIAN JOURNAL OF PHYSICAL CHEMISTRY B Vol. 1 No. 4 2007


318 SHUSTOROVICH

Table 9. C and CH binding energies: UBI–QEP vs. periodic calculations, and they are in reasonable although not
DFTa that good (as for CH3) agreement with the UBI–QEP
DFT UBI–QEPb
values.
Metal Cover-
surface age, θ For CH species, the divergence of the cluster-type
C CH C CH and periodic binding energies becomes more pro-
Ru(0001)c 0.25 165.1 155.1 165.1 111 nounced. For cluster-type calculations of CH on
Ni(111) the ab initio/DFT values form a range from
0.11 174.6 160.1 174.6 119 QCH = 72 kcal/mol [77] to 120 kcal/mol [78, Table 1.9]
Co(0001)d 0.25 162.6 154.0 162.6 109 to 128 kcal/mol [55], which envelops the UBI–QEP value
Rh(111)e 0.25 165.0 155.9 165.0 111 of QCH = 116 kcal/mol (assuming QC = 171 kcal/mol
0.11 168.5 158.1 168.5 114 [13]). However, with the DFT periodic slab calcula-
tions the disagreement is conspicuous. As seen from
Ni(111)d 0.25 154.0 148.7 154.0 101 Table 9, the relevant DFT values of QCH exceed the
Ni(111)f 0.25 146.5 128.5 146.5 94 UBI–QEP ones by a factor of around 3/2.
Pd(111)g 0.33 147.5 136.0 147.5 95 The recurring feature of all periodic DFT calcula-
Pt(111)h 0.25 144.6 136.1 144.6 93 tions is that the binding energies of atom C and methy-
0.25 155.7 148.3 155.7 102 lidine CH are very close whereas the UBI–DFT values
Pt(111)i 0.17 157 150 157 104
differ significantly. This disagreement (accepting that
the periodic DFT calculations of the binding energies
Ag(111)j 36.9 67.4 are more accurate than the cluster-type ones) points out
Notes: a All energies are in kcal/mol. b The UBI–QEP values of to possible deficiency in the UBI–QEP treatment of the
QCH are calculated for the DFT values of QC. c [75]. d [52]. CH radical. The reason should be found in a quantum
e [57]. f [54]. g [58]. h The two values are for the RPBE and mechanical analysis (of the density of states, etc.)
explaining why formation of CH reduces the binding
PW91 functionals, respectively [15]. i Evaluated from [16,
energy of the contact C atom only slightly (binding of 3
Fig. 4]. j Ab initio non-DFT cluster-type calculations [50].
or 4 unpaired electrons makes little difference). Qualita-
tively, the UBI–QEP order should always be QC > QCH,
a and this indeed was found in ab initio calculations
Table 10. Binding energies Q (in kcal/mol) of C2H4 and C 2 H 2 on the VIII metal surfaces. For the IB metals the
order might be different. Table 9 contains an example
Metal Calculated Q Experi- of Ag(111) where the calculated order is reversed,
Adsorbate
surface ab initio UBI–QEP mental Q QC < QCH [50]. We will return to this point below.
di-σ-C2H4 Pd(111) 15b 13 13–14
Within the UBI–QEP model, symmetric hydrocar-
bons, ethylene C2H4 and acetylene C2H2, are treated as
Pt(111) 36c 11 10–17 homonuclear quasi-diatomic A2 molecules formed by
40d quasi-atoms CH2 and CH, respectively. It might be
Cu(111) 7 ~7 interesting to see whether the peculiarities of the CHx
di-σ-C2H2 Pd(111) 17 ~18 binding affect the UBI–QEP binding energies of C2H4
Cu(111) 6e 9 ~16 and C2H2 (in the strongest di-σ mode). They have been
0–41f calculated within two different approximations, namely
with and without bond-energy partitioning, giving sim-
Notes: a All UBI–QEP and experimental data are from [32, Table 2]. ilar results, in broad agreement with experiment albeit
b Periodic DFT calculations [47, 79]. c Cluster-type DFT cal-
scarce (see a discussion in [32]). The scarcity is mainly
culations [80]. d Cluster-type DFT calculations [81]. e Clus- caused by the fact that at low coverage (which is of rel-
ter-type HF/CI calculations [82]. f Cluster-type HF/CI cal- evance to our discussion) ethylene and acetylene can be
culations [83]. adsorbed intact only on few metal surfaces where the
binding is intermediate between too weak (no adsorp-
tion) and too strong (easy decomposition). Some data
Q CH3 = 40 kcal/mol, for all those metal surfaces, which are shown in Table 10 where they are compared with
the proper ab initio values.
would also be within the ranges of Q CH3 obtained by
For C2H4 on Pd(111), the agreement between the
various DFT calculations, as seen from Table 7.
UBI–QEP and (periodic) DFT binding energies and
Table 8 compares the binding energies of CH2. The those with experiment is excellent, all the values being
best studied case is CH2 on Ni(111) where the UBI– within 13–15 kcal/mol. For C2H4 on Pt(111), the UBI–
QEP and (cluster-type) DFT binding energies differ
QEP values of Q CH2 are in good agreement with both greatly, and the latter strongly overestimates the exper-
periodic and cluster-type ab initio/DFT calculations. imental value by 25–30 kcal/mol. For C2H2 on Cu(111),
For other surfaces we cite only results of periodic DFT the UBI–QEP binding energy lies within the wide

RUSSIAN JOURNAL OF PHYSICAL CHEMISTRY B Vol. 1 No. 4 2007


CHEMISORPTION ENERGETICS AND SURFACE REACTIVITY 319

range of the (cluster-type) HF/CI binding energies, Table 11. Binding energies of NHx species on Pt(111): DFT
which is too wide (up to 40 kcal/mol) to make meaning- versus UBI–QEPa
ful comparisons with experiment.
Species DFTb UBI–QEPc Expd
Within the UBI–QEP model, the binding energies of
C2H4 should follow the order of the binding energies of N 107.3 (116) 116
the atomic carbon, that is Ni ~ Pd (see Table 4). Usually NH 92.5 71 71–74
the same trend is found in ab initio studies. Therefore it NH2 71.2 47 >42
looks inconsistent that in the periodic DFT calculations NH3 16.2 14 12
of C2H4 on Ni(111), the binding energy was found to be
4 kcal/mol [84], which is much smaller than the value 13.9 15e
of 15 kcal/mol obtained in the similar calculations of Notes: a All energies are in kcal/mol. b [27]. Only the value of
C2H4 on Pd(111) [47, 79]. However, here the apparent Q NH = 13.9 kcal/mol is with the zero-point energy correc-
inconsistency might be because in the Ni(111) study 3

tions. c [86]. d [87]. e [85].


the C2H4 binding energy was calculated for the on
top π-mode but in the Pd(111) study for the bridge
di-σ-mode, the former being usually much weaker than
the latter [47, 79]. Still, the question remains why the tern was found for N versus NH, namely the normal
on top π-mode for C2H4 on Ni(111) is preferred. order N > NH for Pt(111) (see Table 11) but the
reversed order N < NH (7.6 and 39.4 kcal/mol, respec-
One may wonder whether the CH behavior is unique tively) for Ag(111) [50]. Again, the reason for this
or there are similar cases. The closest analogs are NHx reversal might be the limit of the amount (maximum
species. However, interrelations (if any) between the two) of the shared metal electrons in the metal-adsor-
DFT versus UBI–QEP binding energies of NHx species bate binding [50].
are complex. A representative example is the For the VIII group metals, the CH radical appears to
NHx /Pt(111) system for which one can compare the be the only example where the UBI–QEP formalism to
DFT, UBI–QEP and experimental values, as illustrated calculate the molecular binding energies is conspicu-
in Table 11. In the DFT study [27], the experimental ously deficient. This deficiency is quantitative, how-
binding energies were cited only for NH3 [85]. In the ever. A more serious, conceptual challenge to the
UBI–QEP study [86], the experimental binding ener- UBI−QEP modeling might occur if the discussed pat-
gies were cited for all NHx species, x = 0–3 [87]. For N, terns of the atomic and molecular binding energies on
the DFT binding energy of QN = 107 kcal/mol is close Ag(111) [50] will be found also on other metals.
to the experimental estimate of QN = 116 kcal/mol,
which is taken as the UBI–QEP value of QN. For NH3,
5. COVERAGE EFFECTS
the calculated DFT and UBI–QEP binding energies
( Q NH3 = 14–16 and 14 kcal/mol, respectively) are very 5.1. Atomic Coadsorption
close, especially for the zero-point energy corrected So far we considered the binding energies of iso-
DFT value of Q NH3 = 14 kcal/mol, and they all are in lated adsorbates. Now we move to the energetics of co-
excellent agreement with the cited experimental range adsorption. There is a solid experimental and theoreti-
cal evidence that, at low enough coverage when there is
of Q NH3 =12–15 kcal/mol. For NH and NH2, the DFT no steric hindrance, the direct (through space) A–A
binding energies (given only without the zero-point interaction is much weaker than the indirect (mediated
energy corrections) differ greatly from the UBI–QEP by the surface metal atoms) A–M–A interaction (see
ones but the latter are in much better agreement with references in [12]; for recent work, see, e.g., [88–91]).
experiment. This fact is a cornerstone of the UBI–QEP modeling
We saw that the atomic binding energies QA of A = which treats explicitly only surface-mediated bond
O, N, and C (with the increasing atomic valence) might energies.
have different orders for the VIII (d-band) versus IB M–A bond energy competition in the metal atom
(sp-band) metals. These metals might also show differ- sharing A–M–A: What is conserved? The fundamental
ent orders of the molecular AB binding energies QAB. feature of metal-adsorbate bonding in Mn–A and M–Am
The most striking difference is the relative values of the fragments is competition between the M–A–M and
binding energies of A and AH, A = N, C. Within the A−M–A bonds. This competition is omnipresent and
UBI–QEP model, there should always be QA > QAB. As manifests in weakening binding energies, bond length-
seen from Table 8, for the VIII group metals with the ening, changes in preferred coordination sites under
adsorbates in the same three-fold hollow sites, although coadsorption, etc. These manifestations are invariably
the binding energies of C and CH are very close, the found in all ab initio/DFT calculations. In particular, it
order is always C > CH. By contrast, for the IB metal was said that this competition “is a manifestation of the
Ag with the same adsorbate coordination, the bond order conservation principle: The more the
reversed order C < CH was found [50]. The same pat- valence electrons of a metal atom at the surface become

RUSSIAN JOURNAL OF PHYSICAL CHEMISTRY B Vol. 1 No. 4 2007


320 SHUSTOROVICH

Table 12. Coverage dependence of QO for O/fccAg(111): analytic formalism to calculate QA(θ) [12] has recently
DFT versus UBI–QEPa been extended to better quantify the metal sharing
under co-adsorption. Specifically, two major correc-
DFT QO(θ)/QO(0) ratio
θ Expc tions were introduced [97]:
QO(θ)b QO(0) DFTb UBI–QEPd (1) Rescaling two-center M–A bond energies under
0.11 83.3
co-adsorption to account for the fact that in the M–Am
cluster each A already sits in the hollow site, so the
0.25 81.2 lengthening of the shared A–M–A bonds will weaken
0.33 78.0 other existing bonds and therefore will be resisted.
80.8e 80 (2) An adjustment of the binding energies of the co-
0.50 67.4 0.83 0.83 adsorbed A atoms left on a surface after removal/de-
0.66 59.8 0.74 sorption of the atom A.
0.75 52.4 0.65 The first correction increases QA(θ), the second cor-
rection decreases it. Some simple cases are transparent
1.00 45.7 0.57 0.56 but in general the quantitative estimates require the use
Notes: a All energies in kcal/mol. b [35]. c From Table 3. d From of specially designed computing programs [97].
[12, Table 13]. e Average DFT value within the range of θ = Although these calculations revealed some new inter-
0.11–0.33. esting details, the qualitative UBI–QEP projections did
not change, and the numerical corrections were small.
So, for clarity and simplicity, below we will use the
distributed in bonds with neighboring atoms, the original UBI–QEP QA(θ) values [12].
weaker the individual bond becomes” ([92, p. 231]. The UBI–QEP QA(θ) values change (decrease) dif-
This formulation, however, is a qualitative statement ferently for different n-coordination sites—atop, bridge
about antithetical relations between strengths of the and hollow. For the range of 0 ≤ θ ≤ 1, the atop binding
competing bonds but not a quantitative conservation energies do not change at all, the bridge binding ener-
relationship (explicitly, what is conserved?). By con- gies do not change up to θ = 1/2 and then begin to
trast, the UBI–QEP model quantifies the bonding com- decline, and the hollow binding energies do not change
petition in the form of a genuine conservation stating up to θ = 1/n but then decline even faster than the bridge
what is conserved (namely, explicitly defined bond energies. Because of this steep decline, at some cover-
orders/bond indices) and how this conservation age θ > 1/n the hollow binding energy formally
(namely, at unity) can be quantitatively verified. The ab becomes smaller than the bridge one. Originally it was
initio/DFT studies that find manifestations of the bond assumed that at this point the bridge sites might start to
competition can be used to verify/falsify the quantita- be occupied [12, 94, 95]. Later, however, this assump-
tive UBI–QEP conservation relationships. For the his- tion was abandoned [97] when it was realized that even
tory of the bond order conservation principle, the inter- at high coverage the hollow sites remain the only
ested reader is referred to [9, 12, 93]. energy minima. Indeed, the atomic occupancy of the
Coverage dependence of QA. Consider now quanti- bridge (let alone the on-top) sites have never been
tative aspects of the metal atom sharing. In a regular found experimentally (see a discussion and references
unit mesh Mn–A all the A atoms occupy the n-fold hol- in [97]). Although some ab initio/DFT calculations
low sites. The zero-coverage UBI–QEP binding energy found the bridge site preferred at high enough coverage
QA(θ = 0) does not change within the coverage range (see, e.g., [44, 98–100], most of them find the hollow
0 < θ ≤ 1/n when (in the ground chemisorption states at sites preferred regardless of coverage. So below we will
0 K) there are no co-adsorbates in the adjacent Mn–A discuss the UBI–QEP projections for a coverage-
cells. When the coverage increases at θ > 1/n (and may dependent decrease of QA(θ) only for the hollow sites.
be earlier at finite temperatures), some adjacent cells Only a few DFT calculations were made for a wide
will be occupied, and some atoms M will be coordi- enough coverage range. The periodic slab calculations
nated to more than one adatom A being shared in the of the O/fccAg(111) system [35] are unique by cover-
formed A–M–Am fragments, 1 ≤ m ≤ n. As a conse- ing the whole range of θ = 0.11, 0.25, 0.33, 0.50, 0.66,
quence of the UBI conservation, each initial two-center 0.75, 1.00. For all coverages, the preferred coordination
A–M bond energy Q0A = QA(n)/(2 – 1/n) (cf. Eq. (5)) site is fcc-hollow. The results are shown in Table 12.
decreases, the more so the larger is m. For a given m, the One can see that the values of QO(θ) monotonically
decrease in QA is the more pronounced the more metal decrease from the start, slowly in the beginning and
M atoms are shared. All these lead to the new and precipitously after θ > 0.33. These data allow one to
smaller binding energy QA(θ) < QA(θ = 0) [12]. calculate the QO(θ)/QO(0) ratios, which are also shown
The UBI–QEP analytic formalism allows one to cal- in Table 12 and compared with the UBI–QEP results.
culate QA(i)(θ) for any atom A(i) in any specific configu- Because the UBI–QEP value QO(θ) does not change for
ration of p co-adsorbed atoms A within any (super)cell the range of θ = 0–0.33, the proper DFT value of QO(0)
Ap Mq (θ = p/q) [9–12, 94–96]. The original UBI–QEP for comparisons was taken as the average of three

RUSSIAN JOURNAL OF PHYSICAL CHEMISTRY B Vol. 1 No. 4 2007


CHEMISORPTION ENERGETICS AND SURFACE REACTIVITY 321

QO(θ) values for θ = 0.11, 0.25, 0.33, namely QO(0) ~ Table 13. Coverage dependence of QO for O/fccRh(111):
81 kcal/mol, which is in excellent agreement with the DFT versus UBI–QEPa
experimental value of QO(0) = 80 kcal/mol (see Table 3).
DFT QO(θ)/QO(0) ratio
Within the cited DFT coverage range, the UBI–QEP θ Expc
ratios QO(θ)/QO(0) were calculated for θ = 0.50 and QO(θ)b QO(0) DFTb UBI–QEPd
1.00 (see [12, Table 13]). The agreement with the DFT
ratios is amazingly (better: amusingly) close. 0.25 120.5 120.5 102
Other examples are the DFT periodic calculations of 0.50 113.8 0.94 0.83
O/Rh(111) [38] and O/Pt(111) [48], both starting with 1.00 101.6 0.84 0.56
θ = 0.25. The calculated values of QO(θ) and the Notes: a All energies in kcal/mol. b [38]. c From Table 3. d From
QO(θ)/QO(0) ratios for θ = 0.25, 0.50, and 1.00, are [12, Table 13].
shown in Tables 13 and 14, respectively, and compared
with the UBI–QEP values. In the both cases, the agree-
ment is quantitatively close for θ = 0.50 but only qual- Table 14. Coverage dependence of QO for O/fccPt(111):
itatively consistent for θ = 1.00. DFT versus UBI–QEPa
To summarize, the coverage dependence of the DFT QO(θ)/QO(0) ratio
binding energies is qualitatively similar in the UBI– θ Expc
QEP and DFT approaches for the whole range 0 ≤ θ ≤ 1, QO(θ)b QO(0) DFTb UBI–QEPd Expc
and the agreement is even quantitatively close up to θ =
1/2. After θ > 1/n, the UBI–QEP binding energies typi- 0.25 88.3 88.3 80.0
cally decrease with coverage faster than the DFT ones. 0.50 82.0 0.93 0.83 0.94
Therefore it was suggested to use the DFT values of 1.00 65.9 0.75 0.56
QA(θ) to re-scale the UBI–QEP coverage-dependent
Notes: a All energies in kcal/mol. b [48]. c [36]. d From [12, Table 13].
energetics. This re-scaling, combined with Monte
Carlo simulations, was used in kinetic modeling of var-
ious surface reactions, most exhaustively by Neurock
et al. [91, 100–105]. where
The UBI–QEP analytic formalism [9–12, 94–96] ∆H = D AB + Q AB – Q A – Q B (12)
allows one to calculate also the metal sharing effects on
the binding energies of different co-adsorbates A and B,
is the enthalpy of the dissociation reaction. The interpo-
A ≠ B, both atomic and molecular. This UBI–QEP for-
malism is more complex than that for identical co- lation is accomplished by choosing the proportionality
adsorbates, considered above, and practically can be coefficient k < 1 as
applied only in computer simulations. The primary
importance of such co-adsorption effects is that they ∆E ABa, int = k [ ∆H + Q A Q B / ( Q A + Q B ) ], (13)
affect the coverage-dependent activation barriers of the
dissociation and recombination reactions, ABa which in the simplest (arithmetic mean) form k = 1/2
Aa + Ba. Below we will mention some results, referring give the standard UBI–QEP expression
the reader to the original papers for computational
details. ∆E ABa, int = 1/2 [ ∆H + Q A Q B / ( Q A + Q B ) ] (14)

6. INTRINSIC ACTIVATION BARRIERS or


OF DISSOCIATION AND RECOMBINATION
∆E ABa, int = 1/2 [ D AB + Q AB – Q A – Q B
6.1. Energetic Structure of the UBI–QEP Barrier: (15)
Reminder + Q A Q B / ( Q A + Q B ) ].
In order to compare with the DFT projections let us In the case of a homonuclear diatomic A2, the dissocia-
recall how the UBI–QEP model considers the intrinsic
tion barrier is simply
activation barriers of dissociation ∆E ABa, int and recom-
bination ∆E A–Ba, int reactions, ABa Aa + Ba. ∆E A2a, int = 1/2 [ D A2 + Q A2 – ( 3/2 )Q A ]. (16)
Because the UBI–QEP model calculates only potential
energy minima (binding energies), the saddle For the recombination Aa + Ba ABa, the intrinsic
point/maximum energy (the activation barrier) can only barrier ∆E A–Ba, int is determined by the conservation of
be interpolated within a certain range, whose maximal
value is [9, 13] energy, namely

∆E ABa, max = ∆H + Q A Q B / ( Q A + Q B ), (11) ∆E A–Ba, int = ∆E ABa, int – ∆H. (17)

RUSSIAN JOURNAL OF PHYSICAL CHEMISTRY B Vol. 1 No. 4 2007


322 SHUSTOROVICH

The intrinsic dissociation barrier ∆E ABg, int from the is: not much quantitatively (few kcal/mol, cf. Eq. (14))
and not at all qualitatively (Ni is more active than Pd
gas-phase state is obviously
and Pt), because the oxygen binding energies QO show
∆E ABg, int = ∆E ABa, int – Q AB a very distinct periodic trend decreasing from Ni(111)
(18) to Pt(111) by 30 kcal/mol (see Table 3).
= 1/2 [ ∆H g – Q AB + Q A Q B / ( Q A + Q B ) ], Another periodic UBI–QEP projection was the
where ∆Hg is the dissociation enthalpy from the gas effect of (low-coverage) pre-adsorption of oxygen on
phase the X–H bond cleavage, X = O, N, and C [9, 108]. By
comparing the UBI–QEP activation barriers for direct
∆H g = D AB – Q A – Q B . (19) X–H and oxygen-assisted X–H…O dissociation it was
predicted that the effect of pre-adsorbed oxygen will be
In principle, the observed activation barrier may reversed along the periodic series (e.g., Ag, Cu, Ni, W),
also include a contribution from diffusion of adsorbates from facilitating X–H bond cleavage on the least active
to and from the transition state [9, 106]. One can argue, metals such as Ag and Au to inhibiting this process on
however, that typically this diffusion contribution can the more active metals such as Ni and W. This UBI–
be ignored (although for different reasons for small and QEP projection, being in broad agreement with
large diffusion barriers) [11]. Thus, the intrinsic barrier experiment [109], was corroborated by various DFT
can usually be identified with the apparent (observable) calculations (see, for example, the latest DFT studies
barrier, ∆Eint = ∆E. Indeed, this identification was quite of the oxygen-assisted bond cleavage, N–H…O [27],
successful in numerous UBI–QEP studies of various O−H…O [43], and C–H…O [110], and references
reactions including Fischer–Tropsch chemistry, ammo- therein).
nia decomposition and synthesis, methanol synthesis, Linear relationships: is it a good approximation?
WGS reaction, oxygen-assisted X–H bond cleavage, Many DFT studies revealed a “universal” linear depen-
etc. (see [9, 13, 107] and references therein). This iden- dence of the dissociation barriers on the reaction
tification also justifies direct comparisons of the UBI– enthalpy (see particularly [6–8, 56, 111]). If the refer-
QEP ∆Eint with both the experimental and ab ini- ence point is AB in the gas phase, the claimed depen-
tio/DFT barriers (the latter are also the intrinsic barri- dence (relationship) is
ers). Below we will discuss both general regularities
and numerical values. ∆E ABg, int = k∆H g + C, (20)
where the proportionality coefficient k and the constant
6.2. UBI–QEP Versus DFT: Qualitative Comparisons C may be different for different classes of dissociating
Periodic regularities. Since the atomic binding bonds and for different surface types (say, on close-
energies are much larger than the molecular ones, packed surfaces and on steps). The values of k were
QA, QB  QAB, the UBI–QEP dissociation barriers are usually found to be close to unity, k = 0.75–0.97, which
mainly determined by the values of the atomic binding was often said to be indicative of the product-like or
energies QA and QB. When the latter show distinct peri- “late” transition states (see, for example, [7, 56]). The
odic regularities (see Section 3), the periodic regulari- dependence of Eq. (20) is similar to the BEP (Brøn-
ties of the dissociation barriers immediately follow. In sted–Evans–Polanyi) relationship
particular, if the values of QA and QB decrease from left ∆E AB = α∆H, (21)
to right along the rows and from top to bottom along the
columns, the dissociation barriers ∆EAB increase along where the proportionality coefficient α is typically
the same lines. These regularities have already been assumed to be around α ~ 1/2 (namely, α < 1/2 if the
thoroughly discussed [9–11, 13, 32], and they are in full reaction is exothermic but α > 1/2 if the reaction is
agreement with the results of the ab initio/DFT calcula- endothermic [92]).
tions [1–4]. Within the UBI–QEP framework, the activation dis-
Here a comment might be useful. Compared with sociation barriers ∆E ABg (Eqs. (18)–(19)) are deter-
the original UBI–QEP studies [9, 11, 13, 32] the only mined not only by the reaction enthalpy ∆Hg but also by
substantial difference is that if earlier we assumed the the term QAQB/(QA + QB) – QAB, which may be small
periodic decrease of the carbon binding energies QC or large depending on the relevant Q values. Thus, in
along the series Ni(111) > Pd(111) > Pt(111) with QC =
170, 160, and 150 kcal/mol, respectively, now (see principle, the dissociation barrier ∆E ABg should deviate
above) we assume that there are no changes in QC along from a linear relationship. The ∆EAB versus ∆H rela-
this series, Ni ~ Pd ~ Pt with QC = 155 kcal/mol. The tionship has been found to be non-linear within other
question is how much these changes affect the previous theoretical models as well, in particular for dissociative
UBI–QEP conclusions concerning the periodic trends oxygen adsorption on various surfaces of Rh, Ag, and
in the dissociation barriers ∆ECA of the C–A bonds Au [112]. Still, some approximate linear relationships
along the series Ni Pd Pt, in particular (and (with different ad hoc values of k and C) may some-
most importantly) for the CO dissociation. The answer times be obtained. In particular, the UBI–QEP dissoci-

RUSSIAN JOURNAL OF PHYSICAL CHEMISTRY B Vol. 1 No. 4 2007


CHEMISORPTION ENERGETICS AND SURFACE REACTIVITY 323

ation barrier for a homonuclear molecule A2 (neglect- For the purposes of the present review, a more infor-
ing a small term Q A2  QA = QB) has a form of mative example is the DFT studies of the ethanol
decomposition on Pt(111) by Alcala et al. [111]. For the
∆E A2g ∼ D A2 – ( 3/2 )Q A . (22) dissociation of relevant intermediates, the authors
found the linear relationship, Eq. (20), with the slope of
k = 0.97 ± 0.03 and the constant C = 1.50 ± 0.15 eV. The
Thus, for a given molecule A2 ( D A2 = const) the barrier
∆E ABg versus ∆Hg relations, calculated by the UBI–
differences on different metal surfaces will be propor-
tional to the atomic A binding energy QA, which is a QEP formalism, fits Eq. (20) with k = 0.94 ± 0.04 and
variation of the linear Brønsted–Evans–Polanyi rela- the constant C = 0.85 ± 0.20 eV [113]. Pragmatically,
tionship, Eq. (21). this close agreement with the numerical DFT data man-
ifests the comparable accuracy of the UBI–QEP model
While claiming the existence of the universal linear for quantitative estimates of binding energies and acti-
dependence, Eq. (20), it was also stated that the “accu- vation barriers (see more details below). At the same
racy of the DFT calculations is not such that we can time, because the analytical UBI–QEP model projects
give a quantitative treatment” [6]. Thus, one should that the ∆EAB versus ∆H relationship should in princi-
exercise caution because the conclusions may depend ple be non-linear (Eq. (14)), it would be interesting to
not only on the accuracy of calculations but also on see whether the reported DFT linear relationship will
their interpretation. In particular, the interpretation hold if the electronvolts are replaced by kJ/mol.
might sometimes be affected by the resolution of the
data set depending on the choice of energy units. For As said above, Eqs. (14) or (15) with the UBI–QEP
example, the linear ∆E versus ∆H relationship in elec- proportionality coefficient k = 1/2 is the standard form
tronvolts has been reported in the DFT studies of the corresponding to the simplest interpolation: the arith-
NHx dehydrogenation (dissociation) reactions by metic mean between two extremes [13]. (In this respect
Michaelidis et al. [7, 56], who found the slope close to k is similar to the BEP coefficient α ~ 1/2 in Eq. (21)).
unity (k ~ 1) implying the product-like or “late” transi- Of course for particular cases the coefficient k < 1 may
tion states. However, when Offermans et al. [27] re-cal- be fine-tuned. For example, in the UBI–QEP studies of
culated the DFT reaction energetics with zero-point CO and H2 oxidation and the WGS reaction on Rh, a
energy corrections they found large deviations from the better fit with experiment was found for the value of
proposed linear relationship, being around –20, +50, k = 0.8 [114].
and +20 kJ/mol for the NH3, NH2, and NH dissociation,
respectively. Moreover, when Offermans et al. [27] Geometry of transition states. The intrinsic UBI–
simply magnified the resolution of the data set by QEP activation barriers, being determined completely
Michaelidis et al. [7, 56] by using the energy units in by the energetics of initial (reactants) and final (prod-
kJ/mol instead of eV (1 eV ~100 kJ/mol), similar devi- ucts) states, do not require knowledge of geometry of
ations were found also in the original set, making the transition state (TS). It was claimed [46] that the
meaningless the conclusions concerning the linear rela- UBI–QEP method underestimates the activation barri-
tionship, its slope and “lateness” of the transition state. ers because it ignores interaction energies associated
Because most of DFT studies reported the linear ∆E with coadsorption. This is a plain misunderstanding.
versus ∆H relationship in electronvolts, such a magnifi- The TS energy (activation barrier) is determined by the
cation of the resolution appears to be warranted in all UBI–QEP constraint which automatically includes
those studies, not to mention a necessity of evaluating changes in coadsorption energies from the initial state
the zero-point energy corrections. when the reactants are infinitely separated on a surface
to the TS where the reactants diffuse close enough to
Another example is a DFT study of methane disso- interact. Thus, the UBI–QEP model is not totally blind
ciation on Rh(111) by Bunnik and Kramer [57]. From to the TS geometry but can qualitatively project its
an analysis of the transition state energies and geome- some major features determined by the preferred geom-
tries the authors concluded that, unlike the claim by etry of co-adsorption of non-equivalent species. In par-
Michaelidis et al. [7], the different CHx steps do not fall ticular, the UBI–QEP model projects that the stronger
in the same class, so that there is no universal BEP rela- bound species prefer a minimal change from their
tionship and no simple way to predict the “lateness” of ground-state adsorption site [9, 12]. For example, in the
the transition states. TS of the reactions C + H CH or C + O CO,
A further example is the DFT calculations of H2O the hollow site will be occupied by C but H and O will
dissociation on clean and oxygen-pre-adsorbed metal be close to the bridge site. However, in the TS for the
surfaces [43]. It was demonstrated that even if the O–H oxidation reaction CO + O CO2, the hollow site
dissociation barrier on the clean surfaces might approx- will be occupied by O but CO will be close to the bridge
imately fit the BEP relationship the latter fails for the or atop site. Indeed, this general pattern and the cited
oxygen-pre-adsorbed surfaces, which makes inconclu- specific TS arrangements are found in most DFT calcu-
sive any guess concerning the nature of the transition lations providing the co-adsorption geometries along
states. the (minimum energy) reaction path including those of

RUSSIAN JOURNAL OF PHYSICAL CHEMISTRY B Vol. 1 No. 4 2007


324 SHUSTOROVICH

the TS (see, e.g. a discussion and references in [16, 44, GGA functionals are appropriate for the description of
56, 115]). the C–H bond scission [124].
(2) The zero-point energy effects generally reduce
6.3. UBI–QEP Versus DFT: Quantitative Comparisons the DFT activation barriers (see, e.g., [27, 71]) but
many DFT calculations neglect these effects.
Hybrid UBI–QEP/DFT approach. Recall that the (3) In order to reduce computational efforts, varying
UBI–QEP formalism to calculate the activation barri- constraints on the geometry of the transition state are
ers, Eqs. (11)–(19), requires only the knowledge of the common. The commonest constraint is a fixed adsor-
relevant atomic and molecular binding energies Q. bate–adsorbate distance but it might lead to the incorrect
They should be as accurate as possible but their origin TS and therefore to the incorrect activation barrier. For
does not matter: they might be calculated (by UBI– example, in the DFT calculations of the CO + O
QEP or ab initio/DFT or other proper techniques) or CO2 reaction on Pt(111) this constraint lead to the TS
taken from experiment. Numerous examples have with oxygen in the bridge site while the unconstrained
already been discussed [9, 11, 13] confirming the gen- TS has oxygen in the hollow site (see discussion in [16,
eral rule that the accurate (enough) values of Q lead to 44, 56, 115]).
the accurate (enough) values of ∆E.
(4) The coverage effects even at low coverage
This corroboration of the UBI–QEP formalism appear to be distinct already for atoms (see above), and
prompted a development of some “hybrid” approaches they are more important for molecules. Accordingly,
where the ab initio/DFT binding energies have been the choice of the periodic unit cell (which determines
plugged in the UBI–QEP formulas for the barriers, the coverage) might strongly affect the geometry and
Eqs. (11)–(19). For example, such an approach was the binding energy of the molecular precursor AB* and
employed in the analysis of methanol oxidation [116] the AB dissociation activation barrier. For example, the
and the WGS reaction [117–119], in broad agreement common (2 × 2) unit cell of the fcc(111) surface might
with experiment. Various examples of this kind are con- be too small to properly treat the AB A + B dis-
sidered in the accompanying article in this issue [120]). sociation even for such a small molecule as O2. The
True, there was also one DFT study of chemisorp- A–(2 × 2) unit cell corresponds to θA = 0.25 (<1/n)
tion energetics (for low CO coverage on Ni(111)) without A–M–A sharing but the A2–(2 × 2) cell leads to
where the authors used their data to calculate the UBI– θA = 0.50 (>1/n) with A–M–A sharing and therefore
QEP activation barriers and claimed that the latter are with much smaller binding energies QO. Moreover, in
highly inaccurate [46]. This claim, however, is a result
of obvious errors in calculations of the UBI–QEP val- the molecular O 2* precursor the O–O bond length
ues. The calculated DFT binding energies were QC, QO, increase might be such that, due to the periodic 2 × 2
QCO = 156, 137, 44 kcal/mol, respectively. For this boundary conditions, oxygen atoms form a chain struc-
DFT energy set, the UBI–QEP barrier for CO disso- ture on the surface, so that the dissociation process can
ciation (cf. Eq. (15)) is 39 kcal/mol but the authors mys- be properly studied only using a larger surface cell
teriously cited the UBI–QEP barrier of ∆ECO = [125]. Specifically, this is the case of O2 on Pd(111)
5 kcal/mol. If one uses the assumed experimental when the DFT dissociation barrier for the 2 × 2 cell was
energy set QC, QO, QCO = 171, 115, 27 kcal/mol, respec- found to be about 23 kcal/mol [126] but with the
tively, the UBI–QEP barrier is ∆ECO = 33 kcal/mol (slightly larger) 3 × 2 cell the dissociation trajectory
(cf. [13, Table 9]). Clearly, the problem in the work [46] changed and the barrier reduced to 20 kcal/mol and
is not with using the DFT energy numbers (with the would certainly drop further for a larger cell [125].
experimental Q values there would be very close For the larger (polyatomic) molecules the size of the
results) but with a gross arithmetic error in calculating unit cell is even more critical. An example is the
the UBI–QEP activation barrier. CH3OH/Pt(111) system, where in similar DFT periodic
DFT activation barriers. Ab initio/DFT calculations calculations for slightly differing coverage, θ = 0.11
of the intrinsic barriers are usually obtained by using and 0.25 (the 3 × 3 and 2 × 2 unit cells), the barriers for
the Nudge Elastic Band (NEB) method [121] with fur- O–H bond scission differ by 15 kcal/mol, namely 19
ther improvements and modifications [69, 122, 123]. [71] and 34 kcal/mol [72], respectively.
The NEB method is designed to search for the mini- The list of possibly varying details, often hidden in
mum energy path between the given initial and final the chosen assumptions and approximations, is long.
states, where the potential maxima are the saddle points As a result, even the apparently similar DFT calcula-
and the highest saddle point, corresponding to a transi- tions often give dissimilar activation barriers.
tion state, TS, gives the reaction activation barrier ∆E.
All those factors are irrelevant for the UBI–QEP
But the accuracy, as the proverbial devil, is in model. Recall that its AB A + B dissociation bar-
details. Let us mention some of them. rier (the TS energy) is determined by the ground-state
(1) The common GGA exchange-correlation func- binding energies of the reactant AB and the products, A
tionals may strongly overestimate the barriers for C–H and B (see Eqs. (11)–(19)), so the model is not con-
bond scission, which lead to questioning whether the cerned with the explicit TS geometry and the reaction

RUSSIAN JOURNAL OF PHYSICAL CHEMISTRY B Vol. 1 No. 4 2007


CHEMISORPTION ENERGETICS AND SURFACE REACTIVITY 325

path. For both the reactant and products, there are no Table 15. Intrinsic UBI–QEP activation barriers (kcal/mol):
coverage and steric constraints artificially imposed by comparison with the DFT and experimental valuesa
the boundary conditions. Furthermore, one can often
Calculated ∆EintExperi-
ignore the effect of the atomic, A and B, diffusion bar- Reaction
Metal
mental
riers on the resulting coordination sites when the high surface DFT UBI–QEP ∆E
barriers might cause A and B to stay close enough to be
energetically penalized for the forced A–M–B metal H2O + O 2OH Pt(111) 8c 11 10
sharing. Indeed, if the AB dissociation is exothermic H+O OH Pt(111) 23c 10 <13
(which is typical for O2 on most metal surfaces) the 20d
forming A and B atoms are “hot” enough to travel sev- H + OH H2O Pt(111) 23c 11 16 ± 5
eral lattice distances to find the most favorable coordi- 20d
nation sites minimizing the A–M–B competition (see,
CO + O CO2 Pt(111) 23b 23 25
for example, Monte Carlo/UBI–QEP simulations of the 26e
O2 dissociation on fcc(111) [127] and fcc(100) [97])
surfaces. Anyway, whatever the final mutual A and B Notes: a All UBI–QEP and experimental data are taken from [11,
position, the UBI–QEP coverage-dependent binding Table 6]. b [128]. c [129]. d [56]. e [115].
energies automatically take care of the coverage-depen-
dent barrier. (The higher the coverage the smaller are
the product, A + B, binding energies and therefore the culations [111] and diverse experiment. Although DFT
larger AB dissociation barrier.) values have been considered as a reference for the UBI–
Because there is no direct interrelation between the QEP model, it was stated that because “ab initio frame-
DFT and UBI–QEP barriers, it is not clear when they works can fail in the treatment of some interactions…
might be similar or different. The comparison is further discrepancies between results from the two frameworks
obscured by the fact that for the same system usually are therefore not necessarily indicating a failure of the
there is a variety of DFT calculations producing dis- simple [UBI–QEP] model.”
tinctly different numerical results. Not surprisingly, The tested UBI–QEP quantities were key reaction
there are examples where there is good agreement parameters such as adsorption energies (of reactants,
between the UBI–QEP and DFT calculated barriers, intermediates, and products) and activation barriers of
and there are examples where the agreement is not possible elementary steps. Simulation of the whole
good. At present, a systematic comparison of the DFT decomposition reaction, from CH3CH2OH to CO, CH4,
versus UBI–QEP barriers can hardly be made, and it is and H2, was made. Temperature programmed desorp-
beyond the scope of this review. tion (TPD) spectra and surface coverage of the stable
Below we will give only few examples to illustrate species as a function of the temperature have been sim-
these points. After all, the most important is how con- ulated to provide a direct comparison with experimen-
sistent are the UBI–QEP and DFT projections with tal data. The major findings are as follows:
experiment. So, we will limit our illustrations to reac-
tions when the UBI–QEP barriers are in good agree- (1) With practically no exception, there is close
ment with experiment and will compare them with the agreement between the DFT and UBI–QEP values of
relevant DFT values. Table 15 shows several such the adsorption energies, all in all for 24 species (see
examples which, in hope to discern some regularities, [113, Fig. 1]).
are further limited to the same reaction type (recombi- (2) Qualitatively, the both theoretical approaches
nation) on the same surface, Pt(111). No regularities project practically the same reaction mechanism. For
were discerned: in some cases the agreement is good most part, there is also good agreement between the
but in others, not. activation barriers of various elementary steps. When
The next example, however, is a case of remarkable poor agreement between the DFT and UBI–QEP acti-
correspondence between the UBI–QEP and DFT pro- vation barriers was found, the UBI–QEP projections
jections, and this is for a multi-step process including a usually were in better agreement with experiment, and
variety of reactions—dissociation, recombination, and these were projections concerning such diverse and
desorption. important features as temperatures of the C–C and C–O
bond dissociation, temperature dependence of surface
DFT vs. UBI–QEP: detailed comparison. The concentration of the stable species, and TPD spectra of
process is ethanol decomposition on Pt(111) which reactants and products.
comprises bond dissociation (O–H, C–H, C–O, C–C)
and recombination, and desorption of the reactant The importance of this example is that the DFT and
(CH3CH2OH) and products (CO, CH4, and H2). It was UBI–QEP barriers closely correlate, qualitatively and
studied by Vesselli et al. [113] who applied the UBI– even quantitatively, for a variety of reactions. This
QEP method and used the results as a benchmark for might be taken as a corroboration of the UBI–QEP con-
evaluating the accuracy of the UBI–QEP model by servation along the minimal potential energy path for
comparing with both self-consistent periodic DFT cal- molecular dissociation.

RUSSIAN JOURNAL OF PHYSICAL CHEMISTRY B Vol. 1 No. 4 2007


326 SHUSTOROVICH

7. CONCLUSIONS approaches for the whole range 0 ≤ θ ≤ 1, and the agree-


ment is even quantitatively close up to θ = 1/2. After
The ab initio/DFT and UBI–QEP approaches treat
θ > 1/n, the UBI–QEP binding energies typically
chemisorption energetics and reactivity on metal sur-
decrease with coverage faster than the DFT ones. One
faces in a general and complementary way. The abun-
can use the DFT values of QA(θ) to re-scale the UBI–
dance of the DFT calculations makes it possible to eval-
QEP coverage-dependent energetics, and this re-scal-
uate their accuracy and systematically compare projec-
ing was made in modeling of various surface reactions.
tions of the DFT and UBI–QEP approaches. In this
review we made representative comparisons dealing All other UBI–QEP projections use the absolute
with both qualitative and quantitative aspects of the binding energies QA. Based on the experimental and
energetics and reactivity. theoretical data of the time, in the original UBI–QEP
First, we considered DFT verification of the UBI– studies we assumed a monotonic decrease of QA for all
QEP projections concerning atomic A binding energies atoms (H, O, N, and C) along the VIII–IB rows and
at low coverage (no metal sharing A–M–A). These from top to bottom along the column. Now, based on
model projections are most general and rigorous the systematic DFT calculations (no new experimental
because they are determined by symmetry of the n-fold data) for the VIII group metals, we re-evaluated the car-
coordination site Mn–A and independent of the nature bon binding energies QC and assumed the same value
of A and M. The UBI–QEP projections of the relative of QC = 155 kcal/mol for Ni, Pd, and Pt, differing
energies of the hollow, bridge and atop sites (the hollow from the original (non-equal) UBI–QEP values by 5–
site is preferred) and of the diffusion barriers (along the 15 kcal/mol. These slight changes, however, practically
hollow-bridge-hollow path) are quantitatively corrobo- did not affect the previous UBI–QEP projections. There
rated by numerous DFT calculations of various atoms were also modifications for some nitrogen QN binding
(A = O, N, C, S) on various M surfaces (M = Rh, Ir, Ni, energies, particularly for Ir and Pd, but so far there were
Pt, Cu, Ag, etc.). This quantitative agreement between no UBI–QEP calculations for those metals and there-
the DFT and UBI–QEP values is the most direct cor- fore no projections to correct. For the IB metals the ab
roboration of the validity of the UBI–QEP model. initio/DFT calculations are scarce, mostly limited to
oxygen adsorption. But it looks probable that the peri-
Other UBI–QEP projections include some addi- odic QA regularities for the IB metals might be differ-
tional assumptions and parameters which make the ent. Accordingly, our revision of QA for the IB metals is
model results less rigorous and therefore their DFT ver- more substantial. Now we recommend using only the
ification less certain. The basic UBI–QEP parameter is oxygen parameters QO whereas the nitrogen QN and
the (zero coverage) atomic binding energy QA in the carbon QC parameters need further ab initio corrobora-
hollow site. It’s convenient to distinguish the UBI–QEP tion. The recommended values of QA are summarized
projections by whether they are concerned with the rel- in Table 5.
ative binding energies (expressed in terms of QA) or
they employ the absolute (numerical) value of QA. The The absolute values of QA are used in calculations of
projections of the first kind include coverage effects on molecular QAB binding energies, and here the UBI–
the A binding energies under atomic co-adsorption. The QEP validity has been tested along two lines: (1) pro-
projections of the second kind deal with molecular QAB jected qualitative regularities of QAB, and (2) numerical
binding energies and the AB reactivity patterns (activa- accuracy of values of QAB. The qualitative UBI–QEP
tion barriers ∆EAB). In any model projection there are projections for molecular AB coordination concern
qualitative and quantitative aspects. It is easy to see the with the preferred contact atom A (it has a larger atomic
qualitative agreement or disagreement but it is more binding energy, QA > QB) and the preferred coordina-
difficult to judge the quantitative closeness. The fact is tion site (the higher coordination for the atomic-like
that the DFT (and experimental) values of QA, QAB, and molecular radicals but no preference for weakly bound
∆EAB usually are not unique but each entity forms a molecules). DFT studies fully corroborate these regu-
range of calculated (and measured) values. The UBI– larities. Typically, there is a good correlation between
QEP parameter QA is taken from experiment or theoret- the numerical UBI–QEP and DFT values of QAB. The
ical calculations. For all practical purposes, the UBI– most conspicuous exception is methylidine CH: while
QEP QA should be uniquely numerically defined but the (periodic slab) DFT values of QC and QCH are close
such a single value can hardly be chosen if the theoret- the corresponding UBI–QEP values are distinctly dif-
ical and/or experimental values QA constitute a wide ferent. A quantum mechanical explanation of why the
range. Nevertheless, we tried to make reasonable com- atom C and the radical CH have very close binding
parisons. energies is necessary.
The UBI–QEP projections using the relative bind- Secondly, the UBI–QEP projections using the abso-
ing energies (expressed in terms of QA) include cover- lute binding energies QA and QB (and QAB) deal with
age effects revealing the M–A bond energy competition the AB A + B reactivity patterns judged by the cal-
(metal sharing A–M–A) under atomic co-adsorption. culated activation barriers, dissociation ∆EAB and
The coverage dependence of the binding energies is recombination ∆EA–B. The atomic binding energies are
qualitatively similar in the UBI–QEP and DFT the major contributors to the barriers, such that a

RUSSIAN JOURNAL OF PHYSICAL CHEMISTRY B Vol. 1 No. 4 2007


CHEMISORPTION ENERGETICS AND SURFACE REACTIVITY 327

decrease (increase) of QA and QB results in an increase 11. E. Shustorovich and A. V. Zeigarnik, Russ. J. Phys.
(decrease) of ∆EAB. The qualitative projections include Chem. 80, 4 (2006).
the periodic regularities (changes of ∆EAB along the 12. E. Shustorovich, Surf. Sci. Rep. 6, 1 (1986).
VIII–IB group rows and from top to bottom along the 13. E. Shustorovich, Adv. Catal. 37, 101 (1990).
column, both for the direct A–B and oxygen-assisted 14. J. Greeley and M. Mavrikakis, J. Phys. Chem. B 109,
A–B…O dissociation), the barrier vs. reaction enthalpy 3460 (2005).
relationships (in principle, it should be non-linear 15. D. C. Ford, Y. Xu, and M. Mavrikakis, Surf. Sci. 587,
although sometimes it may be approximated as linear), 159 (2005).
geometry of the transition state (strongly bound adsor- 16. A. Michaelides and P. Hu, J. Chem. Phys. 114, 5792
bates prefer to keep their ground state adsorption site), (2001).
etc. All these qualitative UBI–QEP projections are cor- 17. W. P. Krekelberg, J. Greeley, and M. Mavrikakis,
roborated by DFT calculations. J. Phys. Chem. B 108, 987 (2004).
As far as quantitative aspects are concerned, there is 18. M. Mavrikakis, J. Rempel, J. Greeley, et al., J. Chem.
no explicit interrelation between the DFT and UBI– Phys. 117, 6737 (2002).
QEP barriers, so it’s not clear when they might be close 19. Y. Xu and M. Mavrikakis, Surf. Sci. 494, 131 (2001).
or different. Still, the DFT and UBI–QEP barriers often 20. Y. W. Wang, L. Jia, W. Wang, and K. Fan, J. Phys.
correlate, which may be considered as a corroboration Chem. B 106, 3662 (2002).
of the UBI–QEP conservation along the molecular dis- 21. S. Yamagishi, S. J. Jenkins, and D. A. King, Surf. Sci.
sociation path. 543, 12 (2003).
Given the variety of covered areas of chemisorption 22. P. E. M. Siegbahn and U. Wahlgren, Int. J. Quant.
phenomena and uncertainties of the DFT and experi- Chem. 42, 1149 (1992).
mental numbers, the correlations found between the 23. M. Lynch and P. Hu, Surf. Sci. 458, 1 (2000).
DFT and UBI–QEP projections are remarkably good, 24. T. Jacob, R. P. Muller, and W. A. Goddard, III, J. Phys.
arguably better than one could expect for those comple- Chem. B 107, 9465 (2003).
mentary but totally different approaches. Further com- 25. G. Volpihac, H. F. Busnengo, W. Dong, and A. Salin,
parisons and testing are desirable but they will be real- Surf. Sci. 544, 329 (2003).
istic only under two conditions: (1) experimental data 26. N. Perron, N. Pineau, E. Arquis, et al., Surf. Sci. 599,
should be more complete and definitive, and (2) the ab 160 (2005).
initio/DFT values should be accurate enough and com- 27. W. K. Offermans, A. P. J. Jansen, and R. A. van Santen,
prise a narrow range consistent with experimental data. Surf. Sci. 600, 1714 (2006).
28. J. T. Stuckless et al., J. Chem. Phys. 106, 2012 (1997).
29. L. Isett and J. Blakely, Surf. Sci. 47, 645 (1975).
ACKNOWLEDGMENTS 30. L. Isett and J. Blakely, Surf. Sci. 58, 397 (1976).
I wish to express my gratitude to Dr. A.V. Zeigarnik 31. D. J. Klinke, II, S. Wilke, and J. Broadbelt, J. Catal. 178,
for stimulating discussions and help in literature search. 540 (1998).
32. E. Shustorovich and A. V. Zeigarnik, Surf. Sci. 527, 137
(2003).
REFERENCES 33. G. Cipriani, D. Loffreda, A. Dal Corso, et al., Surf. Sci.
1. R. A. van Santen and M. Neurock, Catal. Rev. Sci. Eng. 501, 182 (2002).
37, 557 (1995). 34. S. Kurth, J. P. Perdew, and P. Blaha, Int. J. Quantum
2. B. Hammer and J. K. Nørskov, Adv. Catal. 45, 71 Chem. 75, 889 (1999).
(2000). 35. W.-X. Li, C. Stampfl, and M. Scheffler, Phys. Rev. B:
3. J. Greeley, J. K. Nørskov, and M. Mavrikais, Annu. Condens. Matter 65, 075407 (2002).
Rev. Phys. Chem. 53, 319 (2002). 36. Y. Y. Yeo, L. Vatuone, and D.A. King, J. Chem. Phys.
4a. J. R. B. Gomes, S. Gonzalez, D. Torres, and F. Illas, 106, 392 (1997).
Russ. J. Phys. Chem. B 1 (2007) (this issue). 37. C. T. Campbell et al., Surf. Sci. 107, 220 (1981).
4b. R. A. van Santen and M. Neurock, Russ. J. Phys. 38. M. V. Ganduglia-Pirovano and M. Scheffler, Phys. Rev.
Chem. B 1 (2007) (this issue). B: Condens. Matter 59, 15533 (1999).
5. B. Hammer and J. K. Nørskov, Surf. Sci. 343, 211 39. Y. Xu and M. Mavrikakis, J. Chem. Phys. 116, 10846
(1995). (2002).
6. J. K. Nørskov et al., J. Catal. 209, 275 (2002). 40. CRC Handbook of Chemistry and Physics, Ed. by
D. R. Lide, 75th ed. (CRC, Boca Raton, FL, 1994–
7. A. Michaelidis, Z.-P. Liu, C. J. Zang, et al., J. Am. 1995), pp. 9–55.
Chem. Soc. 125, 3704 (2003). 41. B. Hammer, L. B. Hansen, and J. K. Nørskov, Phys.
8. T. Bligaard et al., J. Catal. 224, 206 (2004). Rev. B: Condens. Matter 59, 7413 (1999).
9. E. Shustorovich and H. Sellers, Surf. Sci. Rep. 31, 1 42. X. Guo, A. Hoffman, and J. T. Yates, Jr., J. Chem. Phys.
(1998). 90, 5787 (1989).
10. H. Sellers and E. Shustorovich, Surf. Sci. 504, 167 43. G.-C. Wang, S.-X. Tao, and X.-H. Bu, J. Catal. 244, 10
(2002). (2006).

RUSSIAN JOURNAL OF PHYSICAL CHEMISTRY B Vol. 1 No. 4 2007


328 SHUSTOROVICH

44. K. Bleakley and P. Hu, J. Am. Chem. Soc. 1212, 7644 76. Q. Ge, M. Neurock, H. A. Wright, and N. Srinivasan,
(1999). J. Phys.Chem. B 106, 2826 (2002).
45. A. Bogisevic and K. C. Haas, Surf. Sci. 506, L237 77. H. Yang and J. I. Whitten, Surf. Sci. 255, 193 (1991).
(2002). 78. P. E. M. Siegbahn and U. Wahlgren, in Metal–Surface
46. T. Li, B. Bhatia, and D. S. Sholl, J. Chem. Phys. 121, Reaction Energetics, Ed. by E. Shustorovich (VCH,
10241 (2004). New York, 1991).
47. M. Neurock and R. A. van Santen, J. Phys. Chem. B 79. V. Pallassana and M. Neurock, J. Catal. 191, 301
104, 11127 (2000). (2000).
48. P. Legare, Surf. Sci. 580, 137 (2005). 80. J. Kua and W. A. Goddard, III, J. Phys. Chem. B 102,
49. P. Legare, private communication. 9492 (1998).
50. D. B. Kokh, R. J. Buenker, and J. L. Whitten, Surf. Sci. 81. R. M. Watwe, B. E. Spiewak, R. D. Cortright, and
600, 5104 (2006). J. A. Dumesic, J. Catal. 180, 184 (1998).
51. X.-Q. Gong, R. Raval, P. Hu, J. Chem. Phys. 122, 82. K. Herman and M. Witko, Surf. Sci. 337, 205 (1995).
024711 (2005). 83. A. Clotet and G. Pacchioni, Surf. Sci. 346, 91 (1996).
52. D. J. Klinke, II, D. J. Dooling, and L. J. Broadbelt, Surf. 84. R. T. Vang et al., Surf. Sci. 600, 66 (2006).
Sci. 425, 334 (1999). 85. W. I. Guthrie, J. D. Sokol, and G. A. Somorjai, Surf. Sci.
53. E. Shustorovich and A.T. Bell, Surf. Sci. 289, 127 Lett. 109, 390 (1981).
(1993). 86. E. Shustorovich and A. T. Bell, Surf. Sci. Lett. 259,
54. R. M. Watwe, H. S. Bengaard, J. R. Rostrup-Nielsen, L791 (1991).
et al., J. Catal. 189, 16 (2000). 87. J. J. Vajo, W. Tsai, and W. H. Weinberg, J. Phys. Chem.
55. H. Burghgraef, A. P. J. Jansen, and R. A. van Santen, 89, 3243 (1985).
Surf. Sci. 324, 345 (1995). 88. H. S. Kato, H. Okuyama, J. Yoshinobu, and M. Kawai,
56. A. Michaelides and P. Hu, J. Am. Chem. Soc. 122, 9866 Surf. Sci. 513, 239 (2002).
(2000). 89. J. F. Weaver, J.-J. Chen, and A. L. Gerrard, Surf. Sci.
57. B. S. Bunnik and G. J. Kramer, J. Catal. 242, 309 592, 83 (2005).
(2006). 90. W. V. Glassey, Surf. Sci. 600, 173 (2006).
58. J.-F. Paul and P. Sautet, J. Phys. Chem. B 102, 1578 91. D. H. Mei, P. A. Sheth, M. Neurock, and C. M. Smith,
(1998). J. Catal. 242, 1 (2006).
59. W. A. Brown and D. A. King, J. Phys. Chem. B 104,
92. R. A. van Santen and J. W. Niemantsverdriet, Chemical
2578 (2000).
Kinetics and Catalysis (Plenum, New York, 1995).
60. D. Loffreda, D. Simon, and P. Sautet, J. Chem. Phys.
93. P. Blowers and R. Masel, Surf. Sci. 417, 238 (1998).
108, 6447 (1998).
61. L. D. Kieken, M. Neurock, and D. Mei, J. Phys. Chem. 94. A. V. Zeigarnik, L. A. Abramova, S. P. Baranov, and
B 109, 2234 (2005). E. Shustorovich, Surf. Sci. 541, 76 (2003).
62. V. A. Ranea et al., Surf. Sci. 600, 2663 (2006). 95. A. V. Zeigarnik, L. A. Abramova, S. P. Baranov, and
E. Shustorovich, Surf. Sci. 548, 342 (2004).
63. G.-C. Wang, J. Li, X.-F. Xu, et al., J. Comput. Chem. 26,
871 (2005). 96. H. Sellers, Surf. Sci. 524, 29 (2003).
64. W. Lai, D. Xie, and D. H. Zhang, Surf. Sci. 594, 83 97. S. P. Baranov, L. A. Abramova, A.V. Zeigarnik, and
(2005). E. Shustorovich, Surf. Sci. 555, 20 (2004).
65. S.-G. Wang et al., Surf. Sci. 600, 3226 (2006). 98. Q. Ge, P. Hu, D. A. King, et al., J. Chem. Phys. 106,
1210 (1997).
66. Y. Cao and Z.-X. Chen, Surf. Sci. 600, 4572 (2006).
99. E. W. Hansen and M. Neurock, Surf. Sci. 441, 410
67. G.-C. Wang, Y.-H. Zhou, and J. Nakamura, J. Chem. (1999).
Phys. 122, 044707 (2005).
100. E. W. Hansen and M. Neurock, Chem. Eng. Sci. 54,
68. Z. X. Chen, K. M. Neyman, K. H. Lim, and N. Rosch, 3411 (1999).
Langmuir 20, 20 (2004).
101. E. W. Hansen and M. Neurock, J. Catal. 196, 21 (2000).
69. P. Maragakis, S. A. Andreev, G. Henkelman, et al.,
J. Chem. Phys. 113, 9901 (2000). 102. E. W. Hansen and M. Neurock, Surf. Sci. 464, 91
(2000).
70. J. Greeley and M. Mavrikakis, J. Am. Chem. Soc. 124,
7193 (2002). 103. E. W. Hansen and M. Neurock, J. Phys. Chem. B 105,
9218 (2001).
71. J. Greeley and M. Mavrikakis, J. Am. Chem. Soc. 126,
3910 (2004). 104. M. Neurock and D. H. Mei, Top. Catal. 20, 5 (2002).
72. S. K. Desai, M. Neurock, and K. Kourtakis, J. Phys. 105. D. H. Mei, E. W. Hansen, and M. Neurock, J. Phys.
Chem. B 106, 2559 (2002). Chem. B 107, 798 (2003).
73. J. Kua, F. Faglioni, and W. A. Goddard, III, J. Am. 106. E. Shustorovich, Russ. J. Phys. Chem. 68, 1996 (1994).
Chem. Soc. 122, 2309 (2000). 107. A. V. Zeigarnik and E. Shustorovich, Russ. J. Phys.
74. A. Michaelidis and P. Hu, J. Chem. Phys. 112, 6006 Chem. B 1 (2007) (this issue).
(2000). 108. E. Shustorovich and A. T. Bell, Surf. Sci. 268, 397
75. I. M. Ciobica, F. Frehard, R. A. van Santen, et al., Chem. (1992).
Phys. Lett. 311, 185 (1999). 109. M. A. Henderson, Surf. Sci. Rep. 46, 1 (2002).

RUSSIAN JOURNAL OF PHYSICAL CHEMISTRY B Vol. 1 No. 4 2007


CHEMISORPTION ENERGETICS AND SURFACE REACTIVITY 329

110. C.-T. Au, C.-F. Ng, and M.-S. Liao, J. Catal. 185, 12 120. A. V. Zeigarnik and E. Shustorovich, Suppl. Russ.
(1999). J. Phys. Chem. 81, 26 (2007) this issue.
111. R. Alcala, M. Mavrikakis, and J. A. Dumesic, J. Catal. 121. G. Mills, H. Jonsson, and G. K. Schermer, Surf. Sci.
218, 178 (2003). 324, 305 (1995).
112. E. German and I. Efremenko, J. Mol. Struct. 122. G. Heinekelman and H. Jonsson, J. Chem. Phys. 113,
(Theochem) 711, 159 (2004). 9978 (2000).
113. E. Vesselli, G. Coslovich, G. Comelli, and R. Rosei, 123. K. Brumer, D. R. Reichman, and E. J. Kaxiras, J. Chem.
J. Phys.: Condens. Matter 17, 6139 (2005). Phys. 117, 4651 (2002).
114. A. B. Mhadeshwar and D. G. Vlachos, J. Catal. 234, 48 124. C. Sendner, S. Sakong, and A. Gross, Surf. Sci. 600,
(2005). 3258 (2006).
125. K. Honkala and K. Laasonen, J. Chem. Phys. 115, 2297
115. I. N. Yakovkin and N. V. Petrova, Surf. Sci. 600, 2600 (2001).
(2006).
126. A. Eichler, F. Mittendorfer, and J. Hafner, Phys. Rev. B:
116. Y. Ishikava, M.-S. Liao, and C. R. Cabrera, Surf. Sci. Condens. Matter 62, 4744 (2000).
463, 66 (2000).
127. L. Abramova, A. V. Zeigarnik, S. P. Baranov, and
117. G. C. Wang, L. Jiang, et al., J. Phys. Chem. B 107, 557 E. Shustorovich, Surf. Sci. 565, 45 (2004).
(2003). 128. A. Alavi, P. Hu, T. Deutsch, et al., Phys. Rev. Lett. 80,
118. G. C. Wang, L. Jiang, et al., Surf. Sci. 543, 118 (2003). 3650 (1998).
119. L. Jiang, G.-C. Wang, Z.-S. Cai, et al., J. Mol. Struct. 129. A. Michaelidis and P. Hu, J. Am. Chem. Soc. 123, 4235
(Theochem) 710, 97 (2004). (2001).

RUSSIAN JOURNAL OF PHYSICAL CHEMISTRY B Vol. 1 No. 4 2007

You might also like