Grand canonical rate theory for electrochemical and ...
Transcript of Grand canonical rate theory for electrochemical and ...
doi.org/10.26434/chemrxiv.8068193.v5
Grand canonical rate theory for electrochemical and electrocatalyticsystems I: General formulation and proton-coupled electron transferreactionsMarko Melander
Submitted date: 10/06/2020 • Posted date: 12/06/2020Licence: CC BY-NC-ND 4.0Citation information: Melander, Marko (2019): Grand canonical rate theory for electrochemical andelectrocatalytic systems I: General formulation and proton-coupled electron transfer reactions. ChemRxiv.Preprint. https://doi.org/10.26434/chemrxiv.8068193.v5
A generally valid rate theory at fixed potentials is developed to treat electrochemical and electrocatalyticpotential-dependent electron, proton, and proton-coupled electron reactions. Both classical and quantumreactions in adiabatic and non-adiabatic limits are treated. The applicability and new information obtainedfrom the theory is demonstrated for the gold catalyzed acidic Volmer reaction.
File list (3)
download fileview on ChemRxivappendix.pdf (649.78 KiB)
download fileview on ChemRxivmain_article.pdf (1.12 MiB)
download fileview on ChemRxivsupporting_info.pdf (384.50 KiB)
Appendix to Grand canonical rate theory for
electrochemical and electrocatalytic systems I:
General formulation and proton-coupled electron
transfer reactions
Marko M. Melander
Nanoscience Center, P.O. Box 35 (YN) FI-40014, Department of Chemistry,
University of Jyvaskyla, Finland
E-mail: [email protected]
10 June 2020
Abstract. In this appendix the rate constants for non-adiabatic reactions within the
grand canonical rate theory are presented.
1. Non-adiabatic ET and PCET reaction rates within GCE
As shown in the main paper to this appendix, computation of fixed potential rates for of
electronically adiabatic reactions does not yield any fundamental difficulties as compared
to the canonical case; after finding the barrier, one can simply use a simple TST-like
expression to compute the reaction rate using grand free energies. Besides the reaction
barrier, rates depend on the prefactor which is crucial for modelling ET, PT, and PCET
reactions which often feature (non-)adiabatic tunneling. Given the importance of ET,
PT, and PCET in both applications and fundamental studies, the prefactors are treated
here for consistency of the general framework while the computational applications will
be presented separately[1].
Including the prefactor beyond TST presents some difficulties and care is needed.
In particular, the treatment of non-adiabatic processes is difficult; electronic transition
matrix elements are not defined for states with different number of electrons when
only particle conserving operators are used. In other words, the transition needs to be
particle conserving. The non-adiabatic flux-side correlation function utilizes a projection
operator which is explicitly depends on the particle number.[2] Hence, developing a
transition probability which is independent of the particle number is not straight-
forward and therefore one cannot directly use the effective GCE-EVB states to compute
the electronically non-adiabatic rate. Instead, the electronic transition matrix element
needs to be computed separately for each canonical transition which preserve particle
Grand canonical rate theory 2
number upon transition. Afterwards, a summation over the canonical rates is performed
to express the non-adiabatic ET/PCET rate as a GCE expectation value.
To obtain the non-adiabatic TST rate, the Golden-rule approach is used herein.
In the canonical ensemble, the Golden-rule formulations are well established.[3, 4, 5, 6]
Below the theory of non-adiabatic ET and PCET rates within GCE is developed. It
is stressed that the non-adiabatic approach is inherently quantum mechanical and ET,
PT, and PCET reactions describe quantum mechanical tunneling processes.
1.1. Non-adiabatic ET rate
To start with, localized electronic states |iN〉 are specified as eigenstates corresponding
to the electronic Hamiltonian HelN . Electronic states are defined for initial (i) and final
(f) states with a fixed number of particles (N). Then the electronic energies for the
initial and final states at fixed particle number at nuclear geometry Q are
〈iN |HelN |iN〉 = εiN(Q) and 〈fN |Hel
N |fN〉 = εfN(Q) (1)
Within the Born-Oppenheimer approximation (BOA), the nuclear wave functions
and their energies ε in the initial (|mN〉) and final (|nN〉) electronic states are obtained
from
[TQ + εiN(Q)] |mN〉 = εmN |mN〉 and
[TQ + εfN(Q)] |nN〉 = εnN |nN〉(2)
where TQ is the nuclear kinetic energy. Within BOA, the total vibronic wave
function and the corresponding energy factorize as
|imN〉 = |iN〉 |mN〉 and EimN = εiN + εmN (3a)
|fnN〉 = |fN〉 |nN〉 and EfnN = εfN + εnN (3b)
As the different energy contributions are additive, the canonical partition functions
can be factorized
QNi = exp[−βεiN ]
∑m
exp[−βεmN ] and QNf = exp[−βεfN ]
∑n
exp[−βεnN ] (4)
At this point all relevant canonical quantities have been defined and the focus turns
to the GCE formulation of the Golden-rule rate. The GCE partition function for the
initial state is
Ξi =∑N
exp[βµN ]QNi (5)
This equation is inserted in the general GCE rate expression. For the non-adiabatic
limit, the Golden rule rate expression is used. The Golden rule expression is consistent
Grand canonical rate theory 3
with the general rate theory based on the flux approach if a non-adiabatic Hamiltonian
and suitable flux operator are utilized. The GCE-NATST rate constant is then
kGCE−NATST =2π
hΞi
∑N
e−β(εiN−µN)∑m,n
e−βεmN
∣∣∣ 〈Nnf |VN |imN〉∣∣∣2δ(EimN − EfnN)
=2π
h
∑N
∑m,n
pimN
∣∣∣ 〈Nnf |VN |imN〉∣∣∣2δ(EimN − EfnN)(6)
where pimN is the population of the vibronic state |imN〉. Next, a significant
simplification is made; it is assumed that the vibrational part of the canonical partition
function does not depend on the number of electrons in the system. This assumption
directly implies that the reorganization energy is potential independent which should
be a reasonable assumption for electronically non-adiabatic reactions. As a result
QNi = exp[−βεiN ]
∑m exp[−βεmN ] ≈ exp[−βεiN ]
∑m exp[−βεm] = exp[−βεiN ]Qm and
the GCE partition function becomes
Ξi ≈ Qm
∑N
exp[−β(εiN − µN)] = QmΞi (7)
Inserting this approximation in the GCE-NATST rate expression gives
kGCE−NATST ≈2π
hΞi
∑N
e−β(εiN−µN)∑m,n
e−βεmN
Qm
∣∣∣ 〈Nnf |VN |imN〉∣∣∣2δ(EimN − EfnN)
=2π
h
∑N
piN∑m,n
pmN
∣∣∣ 〈Nnf |V |imN〉∣∣∣2δ(EimN − EfnN)
(8)
where piN,el = exp[−β(εiN − µN)]/Ξi,el and pmN = exp[−βεmN)]/Qm.
This equation has the structure of the canonical Golden rule rate weighted by the
GCE probability of being in the initial electronic state iN . To simplify the notation, one
can momentarily concentrate only on the canonical part of the above rate expression.
As shown in Section 2.2, using the Fourier transform presentation of the delta function,
gives
kGCE−NATST ≈∑N
V 2N,fi
2h2 piN
∫dtC(t) (9)
where C(t) is an energy autocorrelation function (see Section 2.2). The
autocorrelation function maybe extracted from time-dependent quantum or classical
dynamics. However, to obtain a closed form for the rate equation, herein the
autocorrelation function is expressed using a cumulant expansion[7]. Using the second
order cumulant expansion, assuming that all solvent degrees of freedom are classical
and taking the short time approximation[8] to the correlation function results in (see
Section 2.2):
Grand canonical rate theory 4
kGCE−NATST ≈∑N
piNV 2N,if
h√
4πkBTλexp
[−
(∆ENfi + λ)2
4kBTλ
](10)
The reorganization and reaction energies are defined as λ = Eim(QF ) − Ein(QI)
and ENfi = EN
fn(QF )− ENim(QI). While the above development rests on the assumption
of constant reorganization energy, the solvent structure depends on the potential and
charge state of the electrode and reorganization energies for reactions near the electrode
may not be constant. To account for this, the reorganization energy can be further
separated to inner and outer sphere components as discussed in Section 2.4. If this
separation is invoked, one can alleviate the assumption that the total reorganization
is independent of the particle number and instead assume that only bulk solvent
(outer sphere) reorganization is a constant while the inner-sphere reorganization energy
depends on the particle number i.e. is potential-dependent.
1.2. PCET kinetics within GCE
The PCET kinetics is based on the PCET rate theory of Soudackov and Hammes-
Schiffer. Within the canonical ensemble the relevant rate expressions were derived in
Refs. [9, 10, 11, 12] and here this treatment is extended to the GCE yielding PCET
rate constants at fixed electrode potentials. The PCET rate constant derivation follows
a similar procedure as the one used above for the ET rates. In the case of PCET, an
additional geometric variable q for the transferring proton is introduced. Within BOA,
the total vibronic wave function is then
|iumN〉 = |iN(q,Q)〉 |uN(Q)〉 |mN〉 (11)
where it is explicitly written that the electronic wave function |iN〉 depends
explicitly on the proton q and system coordinate Q while the proton wave function
|uN(Q)〉 depends on the system coordinate Q. The wave functions and corresponding
energies are solved using equations similar to the ET case
〈iN |HelN |iN〉 = εiN(q,Q) and
〈fN |HelN |fN〉 = εfN(q,Q)
(12a)
[Tq + εiN(q,Q)] |iuN〉 = εiuN |iuN〉 and
[Tq + εfN(q,Q)] |fvN〉 = εivN |fvN〉(12b)
[TQ + εiuN ] |mN〉 = EmN |mN〉 and
[TQ + εfvN ] |nN〉 = EnN |nN〉(12c)
where Tq and TQ are the kinetic energy operators for the proton and other nuclei,
respectively. Within BOA, the total energy of the at fixed N is written as a simple sum
of the three contributions:
Grand canonical rate theory 5
EiumN = εiN + εiuN + EmN (13)
and similarly for the final diabatic state. Furthermore, coupling constant is given
as
〈Nnvf |V (R)N |iumN〉 ≈〈Nvf |V (R)N |iuN〉q 〈Nn|mN〉Q = V (R)NuvS
Nnm
(14)
The SHS treatment of PCET rates is valid for reactions ranging from vibronically
non-adiabatic to vibronically adiabatic scenarios[13] and rate expressions for various
well-defined limits have been achieved. The SHS PCET rate theories are derived
following a path analogous to the derivation of ET rates and extension to the GCE
is rather straightforward. As done by SHS, the Golden rule formulation is used. The
details of this derivation are presented in the Section 2.1 . The simplest GCE-PCET
rate is given for the short time approximation of the energy gap correlation is valid in
the high-temperature limit and static proton donor-acceptor R distance as
k =∑N,u
piu∑v
∣∣V (R)Nuv∣∣2
h√
4πkBTλuvexp
[−(∆EN
uv + λuv)2
4kBTλuv
](15)
where the reaction energy between vibrational states iuN and fvN is ENuv =
EfvnN(qF , QF ) − EiumN(qI , QI). The state-dependent reorganization energy λuv =
Eium(qF , QF ) − Eivn(qI , QI) is assumed independent of the particle number. If some
vibrational modes (besides the R mode) are sensitive to changes in the particle number,
they can be separated from the total reorganization energy by decomposing the total
reorganization energy to inner- and outer-sphere components as shown in Section
2.5. Depending on the form of the prefactor, both electronically and vibronically
adiabatic and non-adiabatic limits of PCET can be reached within the semiclassical
treatment[14, 15, 16] of the prefactor.
1.3. Hybrid GCE-NA-EVB model
Bridging the the NA and adiabatic GCE-EVB models developed above and in the main
paper, vibronic or electronic diabatic states along the reorganization are considered as
shown in Figure 1. As in the NA model, the hybrid model takes the reorganization
of the unreactive nuclei Q as the reorganization and reaction coordinate. However,
unlike in the NA model, the hybrid GCE-NA-EVB model treats the reorganization
coordinate at a fixed electrode potential, as done in the GCE-EVB model. The effective
barrier is then ∝ exp[−β(Λ + ∆Ω)2/4Λ]. At the crossing point of the initial and
final states along Q coordinate the vibronic/electronic diabatic states are brought in
to resonance so that nuclear/electron tunneling can take place. The contribution of
tunneling between different canonical diabatic states depends on the applied potential.
The GCE prefactor can be approximated as a GCE expectation value of canonical
Grand canonical rate theory 6
Figure 1. Schematic picture for electrochemical PCET. The pink (green) line depicts
an initial (final) vibronic diabatic state as a function of the environment reaction
coordinate. The insets show two types of proton curves along the proton transfer
coordinate at the initial, transition and final solvent coordinates. The top (bottom)
inset shows the electronically non-adiabatic (adiabatic) proton curves. The dashed
solid (dashed) electronically non-adiabatic curve corresponds to electron localized at
the initial (final) state. In each inset the black and orange curves correspond to two
different electrode potentials or states with different number of electrons.
prefactors ∝ V 2µ,if ≈
∑N piNV
2N,if =
⟨V 2N,if
⟩µ. Combining the the barrier and prefactor
leads to the NA-GCE-EVB rate constant
kNA−GCE−EV B ≈
⟨V 2N,if
⟩µ√
4kBTΛexp
[−β (Λ + ∆Ω)2
4Λ
](16)
Unlike the NA-TST rates in the previous section, the hybrid model makes of the
GCE free energies. Also, unlike the GCE-EVB model, the hybrid includes a well-defined
way to compute the prefactor as detailed in Section 3 of the SI. Using the Landau-Zener
approach[17], the hybrid model offers a transparent way to interpolate between adiabatic
and non-adiabatic reactions in a single framework. This hybrid model also provides a
tempting way to include and estimate non-adiabatic and tunneling effects in large scale
DFT studies and material screening with kinetics.
Grand canonical rate theory 7
1.4. Analysis of the non-adiabatic GCE rates
In this section the computation of prefactors was considered to go beyond the
TST limit. In particular (non-adiabatic) tunneling in ET and PCET was treated.
The main difficulty in the GCE non-adiabatic rate theory is the treatment of the
electronic/vibronic coupling constant; this term is defined only for particle conserving
transitions. This precludes the straightforward use of initial and final GCE diabatic
states which have different number of electrons at the same geometry. The NA-GCE-
EVB hybrid offers a well-defined way to approximate the non-adiabatic effects using
GCE diabatic states but the prefactors still need to be computed using canonical states
and GCE averaging. Only at the thermodynamic limit when the particle number
fluctuation approaches zero can the GCE diabatic states be used for computing the
coupling constant. However, at this limit the GCE-NATST is equal to the canonical
NATST as only a single particle number state is populated i.e. pi becomes a delta
function at some particle number. At the thermodynamic limit either using fixed
potential GCE states or fixed particle number canonical states will give equivalent
results, as they should.
Even at the thermodynamic limit the present treatments differ from the traditional
Dogonadze-Kutzetnotsov-Levich[5], Schmikler-Newns-Anderson[18, 19, 20], and SHS
approaches. The crucial difference is that the present formulation does not rely on the
separation of the total interacting wave function to non-interacting or weakly interacting
fragments. Also, in the present approach, the applied electrode potential does not only
affect the electrode potential is self-consistently treated to affect all electrode, reagent,
and solvent species. This way the inherent complexity of the electrochemical interface
is naturally included in the Hamiltonian and the wave functions in a self-consistent
manner. For instance, the work terms entering Marcus[21] or other electrochemical
rate theories[22, 23] do not need to be computed in the present formalism. Another
crucial difference is that the charge transfer kinetics are not decomposed into single
electron orbital contributions. Instead, the work herein formulates the kinetics in terms
of many-body diabatic wave functions. In the canonical ensemble, such an approach has
been shown[24] to provide accurate barriers, prefactors, and overall kinetics for electron
transfer reaction in battery materials.
For small systems where particle number fluctuations are pronounced the
summation over particle numbers need to be performed. While straightforward in
principle, the amount of calculations can seem daunting at first. However, as the
populations depend exponentially on the energy and target chemical potential, piN ∼exp[−β(EiN − µN)], only a limited number of states will contribute to the summation.
It is expected that the infinite summation can be safely reduced to summation over a
small number (5–10) of different charge states covering the electrode potential range of
interest. Again, at the thermodynamic limit only a single calculation per potential is
needed.
Grand canonical rate theory 8
2. Supporting information for the appendix
2.1. PCET kinetics within GCE
Continuing the PCET scheme set up in the paper section 1.2., the PCET rate constant
is written using the Golden Rule formulation. This gives
kGCE−PCET =2π
hΞi
∑N,u,v,m,n
e−β(EiumN−µN)∣∣∣ 〈Nnvf |VN |iumN〉∣∣∣2δ(EiumN − EfvnN)
=2π
h
∑N
∑u,v
∑m,n
piumN
∣∣∣ 〈Nnvf |VN |iumN〉∣∣∣2δ(EiumN − EfvnN)(17)
The obtained form is analogous to the GCE-ET theory developed herein and shares
the structure of the canonical PCET rate of SHS. As assumed for ET part, it is
expected that the vibrational part of the system does not depend on the number of
particles. However, no such assumption is made for the transferring proton i.e. the
proton potential depends on the charge state. This is written as
Ξi =∑N,u,m
e−β(EiumN−µN) ≈ Qm
∑N,u
e−β(εiN+εiuN−µN) = QmΞiu (18)
At this point it is important to stress that the vibronic coupling depends sensitively
on the proton donor-acceptor distance R which is included in the rate expression. It is
assumed that the coupling can be decomposed as
〈Nnvf |V (R)N |iumN〉 ≈ 〈Nvf |V (R)N |iuN〉q 〈Nn|mN〉Q = V (R)NuvSNnm (19)
Inserting these two approximations result in PCET rate constant of the form
kGCE−PCET ≈2π
h
∑N,u,v
e−β(εiN+εiuN−µN)
Ξiu
∑m,n
e−βEmN
Qm
∣∣V (R)Nuv∣∣2∣∣SNmn∣∣2δ(EiumN − EfvnN)
=2π
h
∑N,u,v
piuN∑m,n
pm∣∣V (R)Nuv
∣∣2∣∣SNmn∣∣2δ(EiumN − EfvnN)
(20)
This form is amenable to the direct treatment as performed by SHS. Depending
on the treatment of the R coordinate, several appropriate limits may be considered
each yielding a different canonical rate constant. The derivations for the R-dependent
PCET rates follow a similar (but more complex [11]) cumulant expansion as performed
above for ET. Hence, the GCE-PCET rate can be obtained by extending the approach
presented above for the ET. The extension of PCET in GCE is straight-forward and here
I present only the most simple result valid under the same conditions as the Marcus-
like expression derived above for ET. Specifically, one assumes that[25] i)the short time
Grand canonical rate theory 9
approximation of the energy gap correlation is valid, ii) high-temperature limit is taken,
and iii) that the R coordinate gives Eq. 15.
2.2. Fourier transform and cumulant expansion for ET reactions
Here it shown how the non-adiabatic rate constants are obtained from cumulant
expansion to the energy autocorrelation function. First, the autocorrelation function is
related to the delta function as
∑m,n
pimN
∣∣∣ 〈Nnf |VN |imN〉∣∣∣2δ(EimN − EfnN)
=1
2πh
∑m,n
pimN
∣∣∣ 〈Nnf |VN |imN〉∣∣∣2 ∫ dteit(EimN−EfnN )/h
=1
2πh
∑m,n
pimN 〈fmN |VN |inN〉 〈inN |VN |fmN〉∫dteit(EimN−EfnN )/h
≈ 1
2πh
∑m,n
pimN
∣∣∣ 〈fN |VN |iN〉∣∣∣2 ∫ dt 〈mN |nN〉 〈nN |mN〉 eit(EimN−EfnN )/h
=1
2πh
∑m,n
pimNV2N,if
∫dt∣∣∣〈nN |mN〉q∣∣∣2eit(EimN−EfnN )/h
=V 2N,if
2πh
∫dt⟨eit(EimN/he−it(EfnN )/h
⟩q
=V 2N,if
2πh
∫dtC(t)
(21)
where C(t) is an energy autocorrelation function. The last two equations are
amenable to two different ways of computing the rate constant. The last can be used
with a cumulant expansion approach, while the second last has the form of a thermally
averaged Franck-Condon treatment is presented in Section 2.3 for completeness.
While nuclear quantum effects maybe important and can be included in the
computation of C(t)[26], in the present work, nuclear degrees of freedom are treated
classically. Following either Geva[26] or Marcus[27], the autocorrelation function can be
expressed using a cumulant expansion[7]. Use of the second order cumulant expansion
results in
〈exp[iEfnN t/h] exp[iEimN t/h]〉i ≈
exp
[−ith
⟨∆EN
fi
⟩− 1
h2
∫ t
0
dτ1
∫ τ1
0
dτ2C(τ1 − τ2)
](22)
where⟨∆EN
fi
⟩is the average free energy gap between the final and initial electronic
diabatic states. Also C(τ1−τ2) =⟨δ∆EN
fi(τ)δ∆Efi(0)⟩
where δ∆ENfi = ∆EN
fi−⟨∆EN
fi
⟩.
C(τ1−τ2) is directly linked to the vibrational spectral density of the system[27, 28, 11, 6].
To obtain a manageable expression for the rate, the short time approximation or slow
Grand canonical rate theory 10
fluctuation limit[8] to the correlation function is used: C(τ1− τ2) ≈ C(0) =⟨δ(∆EN
fi)2⟩.
Inserting this in Eq. (22) yields
exp
[− 1
h2
∫ t
0
dτ1
∫ τ1
0
dτ2C(τ1 − τ2)
]≈ exp
[− t
2
h2 (⟨δ(∆EN
fi)2⟩]
(23)
This is inserted in Eq. (21) to give
∑m,n
pimN
∣∣∣ 〈Nnf |VN |imN〉∣∣∣2δ(EimN − EfnN)
≈V 2N,if
2πh
∫ ∞−∞
dt exp
[it
h
⟨∆EN
fi
⟩− t2
h2 (⟨δ(∆EN
fi))2⟩]
=V 2N,if
2πh
√2πh2⟨
δ(∆ENfi)
2⟩ exp
[−⟨∆EN
fi
⟩2
2⟨δ(∆EN
fi)2⟩]
≈V 2N,if
2π
√π
kBTλexp
[−
(∆ENfi + λ)2
4kBTλ
](24)
where on the last line it has been assumed that the free energy surfaces are quadratic
along the energy gap coordinate. The reorganization and reaction energies are defined as
λ = Eim(QF )−Ein(QI) and ENfi = EN
fn(QF )−ENim(QI). A generalization to asymmetric
GCE-diabatic energy curves can be made following Mattiat and Richardson[29].
Furthermore, it is assumed that the curvature of the quadratic surfaces is the same
for all particle numbers N in which case the reorganization energy does not depend on
N . This should be to a rather good approximation as the reorganization is related to
the reorientation of the surrounding medium which is expected be rather insensitive to
the number of electrons in the system. For example, in the spin-boson model, which in
the canonical ensemble yields the Marcus rate, the reorganization energy is only related
to the bath frequencies in thermal equilibrium.[6] If the spin-boson model is applied
to the present GCE case, the vibrational, bosonic Hamiltonian would be assumed to
be independent of the number of electrons and yield directly the reorganization energy
which is indepenedent of the number of particle for the GCE. The assumption that
the reorganization energy is independent on the particle number can also be reinforced
by doing a re-derivation of the rate using the thermalized Franck-Condon approach as
shown in Section 2.3.
2.3. Franck-Condon derivation of the non-adiabatic rate
The Franck-Condon treatment starts from the second-last line of Eq. (21) by noticing
that
Grand canonical rate theory 11
1
2πh
∑m,n
pimNV2N,if
∫dt|〈nN |mN〉|2eit(EimN−EfnN )/h =
V 2N,if
2πFC(∆E)i
(25)
where FC(∆E)i is the thermalized Franck-Condon factor. In general case, the
thermalized Franck-Condon factor can be computed by Fourier transforming it and
using generating functions.[30] As shown in Ref. [8] chapter 6,the FC-factor can be
written using the spectral density function Jfi(ω) to give
FC(∆E)i =1
2πhexp[G(0)]
∫ ∞−∞
dt exp[it∆EN
fi/h+G(t)]
≈∫ ∞−∞
dt
2πhexp
[it
∆ENfi − λh
]exp
[− λt
2
βh2
]
=
√1
4πkBTλexp
[−
(∆ENfi + λ)2
4kBTλ
]
where G(t) =
∫ ∞0
dω cos(ωt)(1 + 2n(ω))JIF (ω)− sin(ωt)JIF (ω)
≈∫ ∞
0
dω(ωt)2
βhωJIF (ω)− i
∫ ∞0
dωtωJIF (ω)
(26)
using the high-temperature approximation (1 + 2n(ω) ≈ 2kBT >> 1), slow-
fluctuating Debye solvent assumptions, and∫∞
0dωωJIF (ω) = λ/2. Hence, if the spectral
density not sensitive to the number of electrons, the reorganization energy is independent
on the number of electrons in the systems. For practical purposes this is expected to be
a good approximation. When the approximate FC factor is introduced, Eq. (25) gives
the Marcus rate in the GCE.
2.4. Decomposition of the reorganization energy to inner- and outer-sphere
contributions
The total reorganization energy is often[9, 31, 21, 32] modified differentiate between
inner- and outer-sphere contributions. This is achieved by partitioning the surrounding
molecules to tightly bound ligands or inner-solvent solvent molecules and the bulk
solvent. While this is not necessary in the approach taken in this work, separating the
effect the nearby atoms or molecules and the solvent might be useful for a understanding
the role of different constituents on the overall reaction. In both computational and
theoretical studies this separation occurs naturally if the bulk solvent is presented as a
continuum as in the work of Dogonadze et.al.[5, 4] for ET and SHS[9] for PCET.
To single out the solvent reorganization energy, a solvent polarization coordinate
Q is introduced. As detailed in Ref. [9] this coordinate introduces a new parametric
Grand canonical rate theory 12
dependence to the electron, proton, and vibrational Hamiltonians, wave functions and
energies. Here it is shown how an additional solvent coordinate modifies the ET reactions
and the PCET kinetics can be treated analogously.
First, a solvent coordinate Q is introduced. The solvent coordinate is
orthogonal to other coordinates which allows writing the wave function as |imaN〉 =
|iN(q,Q)〉 |mN(Q)〉 |aN〉 where |aN〉 is the wave function related to solvent
polarization. All other quantities obtain a parametric dependence on Q. The initial
state solvent wave functions are eigenfunctions obtained from
[TQ + εmN ] |aN〉 = EaN |aN〉 (27)
and similarly for the final state. Above, TQ is the kinetic energy operator for the
outer-sphere species. Then the total energy is given by
EimaN = εiN + εimN + EaN (28)
and the total coupling between the initial and final states is
VimaN,fnbN = 〈fmbN |VN |imaN〉≈ 〈fN |VN |iN〉 〈nN |mN〉q 〈bN |aN〉Q= Vif,NSnm,NSab,N
(29)
Assuming that the outer-sphere free energy related to the solvent reorganization
is independent of the particle number allows separating its contribution from the total
grand partition function
Ξi =∑m,a,N
exp[−β(EimaN − µN)]
≈ Qa
∑m,N
exp[−β(EimN − µN)] = QaΞim
(30)
Note that inner-sphere energies and partition function explicitly depend on the
particle number. Inserting the last two equations in the golden rule expression yields
k =2π
hΞi
∑Nabmn
e−β(εiN−µN+βEiaN+εmN )
∣∣∣ 〈Nnvf |VN |iumN〉∣∣∣2δ(EimaN − EfnbN)
≈ 2π
h
∑N
∑m,n
pimN∑a,b
paNV2if,NS
2nm,NS2
ab,Nδ(EimaN − EfnbN)(31)
where pimN = exp[−β(εiN + εmN − µN)]/Ξim and paN = exp[−βEaN/Qa]. As
done above, representing the delta function as a Fourier transform allows writing
Grand canonical rate theory 13
k =∑N
V 2if,N
h2
∫dt⟨eit(εmN/he−it(εnN )/h
⟩q×⟨eit(EaN/he−it(EbN )/h
⟩Q
=∑N
V 2if,N
h2
∫dtGmn,N(t)gab,N(t)
(32)
where auxiliary correlation functions Gmn,N(t) and gab,N(t) are introduced providing
a connection to the work of SHS[9, 11]. To be specific, Gmn,N(t) characterizes the inner-
sphere contributions while gab,N(t) is related to the outer-sphere solvent polarization.
Different approximations for the correlation functions presented by SHS in Ref. [9, 11]
can be readily used here as well to derive various well-defined limits of the rate equation.
For example, assuming that the intra-molecular modes can be neglected leads to Eq.(21)
with a/b replacing the m/n indices. Within this assumption and repeating the steps
leading to Eq. 10 shows that resulting reorganization energy is the solvent reorganization
energy and the inner-sphere interactions contribute only to the reaction energy.
If the intra-sphere contributions cannot be neglected, the rate equations become
rather cumbersome in general. However, the case Gab,N(t) ≈ Gab(t) i.e. that the
outer-sphere contribution to rate is independent of the particle number, deserves some
attention. For this, the inner- and outer-sphere components are separated by rewriting
Eq.(31) using a convolution[32]
k =∑N
piN2πV 2
if,N
h
∫dEf(x)F (∆EN
fi − x) (33)
with f(x) =∑
mn pmNS2nm,Nδ(ε
imN − εinN + E) and
F (ENfi − x) =
∑ab paNS2
ab,Nδ(EaN − EbN + ∆ENfi − x) as shown for single N in
Ref.[32]. f(x) and F (ENfi − x) represent inner- and outer-sphere contributions to the
activation energy. Again various forms for both terms can be derived[32]. To retain
consistency, a high-temperature approximation for quadratic solvent modes is used.
This gives[32, 9, 11]
F (ENfi − x) =
1
h√
4πkBTλNoexp
[−
(∆ENfi + λNo )2
4kBTλNo
](34a)
f(x) = FC(∆E − x)i (34b)
where FC(∆E − x)i is a modified Franck-Condon factor given in (26) and λNo is
recognized as the outer-sphere reorganization energy. Making the high-temperature and
slow-fluctuating Debye solvent approximations as done in Eq (26) allows performing the
convolution integral. This yields [32]
k =∑N
piN2πV 2
if,N
h
1
h√
4πkBT (λNo + λNi )exp
[−
(∆ENfi + λNo + λNi )2
4kBT (λNo + λNi )
](35)
Grand canonical rate theory 14
Finally the assumption that the outer-sphere contributions do not depend on the
particle number can be applied to give
k =∑N
piN2πV 2
if,N
h
1
h√
4πkBT (λo + λNi )exp
[−
(∆ENfi + λo + λNi )2
4kBT (λo + λNi )
](36)
From this form it can be seen that the total reorganization energy can be separated
to a particle number independent solvent contribution λo and a reorganization energy
of the inner sphere component λNi which depends explicitly on the particle number.
3. Declaration of interest
Declarations of interest: none
4. References
[1] Melander M M 2020 Grand canonical rate theory ii: Addressing non-adiabaticity and tunneling
in gold-catalyzed Volmer reaction from density functional theory in preparation
[2] Richardson J O and Thoss M 2014 The Journal of Chemical Physics 141 074106
[3] Hammes-Schiffer S and Stuchebrukhov A A 2010 Chemical Reviews 110 6939–6960
[4] Dogonadze R and Kuznetsov A 1975 Progress in Surface Science 6 1 – 41
[5] Dogonadze R 1971 3. theory of molecular electrode kinetics Reactions of Molecules at Electrodes
ed Hush N (Wiley-Intersciences) pp 135–228
[6] Nitzan A 2006 Chemical Dynamics in Condendsed Phases: Relaxation, Transfer, and Reactions
in Condensed Molecular Systems (Oxford University Press)
[7] Kubo R 1962 Journal of the Physical Society of Japan 17 1100–1120
[8] May V and Kuhn O 2011 Charge and Energy Transfer Dynamics in Molecular Systems vol 3rd
(WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim)
[9] Soudackov A and Hammes-Schiffer S 2000 The Journal of Chemical Physics 113 2385–2396
[10] Soudackov A and Hammes-Schiffer S 1999 The Journal of Chemical Physics 111 4672–4687
[11] Soudackov A, Hatcher E and Hammes-Schiffer S 2005 The Journal of Chemical Physics 122 014505
[12] Venkataraman C, Soudackov A V and Hammes-Schiffer S 2008 The Journal of Physical Chemistry
C 112 12386–12397
[13] Hammes-Schiffer S 2012 Energy Environ. Sci. 5(7) 7696–7703
[14] Goldsmith Z K, Lam Y C, Soudackov A V and Hammes-Schiffer S 2019 Journal of the American
Chemical Society 141 1084–1090
[15] Georgievskii Y and Stuchebrukhov A A 2000 The Journal of Chemical Physics 113 10438–10450
[16] Skone J H, Soudackov A V and Hammes-Schiffer S 2006 Journal of the American Chemical Society
128 16655–16663
[17] Newton M D 1991 Chemical Reviews 91 767–792 (Preprint
https://doi.org/10.1021/cr00005a007) URL https://doi.org/10.1021/cr00005a007
[18] Schmickler W 1986 Journal of Electroanalytical Chemistry and Interfacial Electrochemistry 204
31 – 43 ISSN 0022-0728
[19] Schmickler W 2017 Russian Journal of Electrochemistry 53 1182–1188
[20] Santos E, Lundin A, Potting K, Quaino P and Schmickler W 2009 Phys. Rev. B 79(23) 235436
[21] Marcus R A 1965 The Journal of Chemical Physics 43 679–701
[22] Lam Y C, Soudackov A V, Goldsmith Z K and Hammes-Schiffer S 2019 The Journal of Physical
Chemistry C 123 12335–12345
Grand canonical rate theory 15
[23] Nazmutdinov R R, Bronshtein M D and Santos E 2019 The Journal of Physical Chemistry C 123
12346–12354
[24] Park H, Kumar N, Melander M, Vegge T, Garcia Lastra J M and Siegel D J 2018 Chemistry of
Materials 30 915–928
[25] Hammes-Schiffer S, Hatcher E, Ishikita H, Skone J H and Soudackov A V 2008 Coordination
Chemistry Reviews 252 384 – 394 the Role of Manganese in Photosystem II
[26] Sun X and Geva E 2016 The Journal of Physical Chemistry A 120 2976–2990
[27] Georgievskii Y, Hsu C P and Marcus R A 1999 The Journal of Chemical Physics 110 5307–5317
[28] Hynes J T 1989 Chemical Physics Letters 162 19 – 26
[29] Mattiat J and Richardson J O 2018 The Journal of Chemical Physics 148 102311
[30] Englman R and Jortner J 1970 Molecular Physics 18 145–164
[31] JOM Bockris S K 1979 Quantum Electrochemistry (Plenum Press)
[32] Kestner N R, Logan J and Jortner J 1974 The Journal of Physical Chemistry 78 2148–2166
download fileview on ChemRxivappendix.pdf (649.78 KiB)
Grand canonical rate theory for electrochemical andelectrocatalytic systems I: General formulation andproton-coupled electron transfer reactions
Marko M. Melander
Nanoscience Center, P.O. Box 35 (YN) FI-40014, Department of Chemistry,University of Jyvaskyla, Finland
E-mail: [email protected]
10 June 2020
Abstract. Electrochemical interfaces present a serious challenge for atomisticmodelling. Electrochemical thermodynamics are naturally addressed within thegrand canonical ensemble (GCE) but the lack of fixed potential rate theoryimpedes fundamental understanding and computation of electrochemical rateconstants. Herein, a generally valid electrochemical rate theory is developed byextending equilibrium canonical rate theory to the GCE. The extension providesa rigorous framework for addressing classical reactions, nuclear tunneling andother quantum effects, non-adiabaticity etc. from a single unified theoreticalframework. The rate expressions can be parametrized directly with self-consistentGCE-DFT methods. These features enable a well-defined first principles routeto address reaction barriers and prefactors (proton-coupled) electron transferreactions at fixed potentials. Specific rate equations are derived for adiabaticclassical transition state theory and adiabatic GCE empirical valence bond(GCE-EVB) theory resulting in a Marcus-like expression within GCE. FromGCE-EVB general free energy relations for electrochemical systems are derived.The GCE-EVB theory is demonstrated by predicting the PCET rates andtransition state geometries for the adiabatic Au-catalyzed acidic Volmer reactionusing (constrained) GCE-DFT. The work herein provides the theoretical basisand practical computational approaches to electrochemical rates with numerousapplications in physical and computational electrochemistry.
Keywords: electrochemical kinetics, grand canonical, free energy relations, Volmerreaction, constrained DFTSubmitted to: J. El. Chem. Soc.
Grand canonical rate theory 2
1. Introduction
Electrochemical reactions and especially electrocatal-ysis are at the forefront of current green technologiesto mitigate climate change. To realize and utilize thefull potential of electrocatalysis, selective and activecatalysts are needed for various applications and re-actions including e.g. oxygen and hydrogen reduc-tion/evolution reactions, nitrogen reduction to ammo-nia and CO2 reduction.[1] These and other electrocat-alytic/electrochemical reactions are based on succes-sive proton-coupled electron transfer (PCET), electrontransfer (ET), and proton transfer (PT) reactions; theunique aspect of electrochemistry is the ability to di-rectly control PCET, ET, and PT kinetics and ther-modynamics by the electrode potential.[2]
Besides the catalyst material, electrocatalytic per-formance is controlled by the electrolyte compositionand electrode potential. To translate these to mi-croscopic, computationally treatable quantities, it isthe combination of the electrolyte and electron elec-trochemical potentials which determine and controlthe (thermodynamic) state of electrochemical systems.Therefore, an atomic-level computational model needsto provide an explicit control and description of thesechemical potentials as depicted in Figure 1. In ther-modynamics fixing the chemical potentials is achievedthrough a Legendre transformation from a canonicalensemble to a grand-canonical ensemble (GCE) forboth electrons and nuclei.[3] This calls for theoreti-cal and computational methods to treat systems wherethe particle numbers are allowed to fluctuate and thechemical potentials are fixed.
The theoretical basis for fixed potential electronicstructure calculations was developed by Mermin whoformulated electronic density functional theory (DFT)within GCE.[4, 5]. Later, GCE-DFT has been gener-alized for treating nuclear species either classically orquantum mechanically [3, 6, 7, 8, 9]. The GCE-DFTprovides a fully DFT, atomistic approach for comput-ing free energies of electrochemical and electrocatalyticsystems at fixed electrode and ionic/nuclear chemicalpotentials.[3] Importantly, the free energy from a GCE-DFT calculation is in theory exact and unique to agiven external potential. In practice, the (exchange-)correlation effects in both quantum and classical sys-tems need to be approximated. The thermodynamicGCE framework has already been adopted by the elec-tronic structure community to model electrocatalytic
thermodynamics at fixed electrode[10, 11, 12, 13, 14,15, 16, 17, 18, 19, 20, 3] and ion potentials[3, 14, 12].Based on the large number of theoretical and compu-tational works utilizing GCE-DFT, the computationalframework for thermodynamics within GCE seems gen-erally accepted. The thermodynamic approach hasprovided fundamental atomic level insight on reactionsat complex electrochemical interfaces and enabled com-putational catalyst screening using free energy rela-tions, Volcano curves, and scaling relations.[1]
Figure 1. Pictorial description of a proper electrochemicalinterface at fixed electron µ and solvent/electrolyte µ± chemicalpotentials.
However, it has been shown that a purely thermo-dynamic perspective on electrocatalysis is not sufficientfor understanding and predicting activity, selectivity,or catalytic trends.[21, 22, 23, 24] Besides applicationsin catalysis and material science, electrochemical ki-netics are fundamentally important and provide a wayto understand complex solvent effects, electron andnuclear tunneling, and non-adiabatic reactions. Ide-ally both fundamental and applied kinetic computa-tional/theoretical studies should make use of generaland self-consistent first principles Hamiltonians withinGCE. This has, unfortunately, remained unattainabledue to theoretical and methodological difficulties andomissions.[25] Surprisingly, a general GCE rate theoryhas not yet been established; mending this deficiencyis the central goal of the present work.
Before diving to the development of the GCErate theory, it is worth considering what new andimportant information can be obtained from a general
Grand canonical rate theory 3
electrochemical rate theory. First and foremost, thetheory needs to accurately capture the intricacies ofET, PT, and PCET reactions as function of theelectrode potential. Therefore, a general treatment ofelectrochemical reaction rates needs to be applicableto 1) both inner-sphere and adiabatic as well asouter-sphere and non-adiabatic reactions, 2) sequentialET/PT or decoupled reactions as well as simultaneousPCET reactions, 3) tunneling of both electrons andnuclei, and 4) be combined with general first principlesGCE Hamiltonians. The motivation for including eachof these four requirements is discussed next.
First, adiabatic inner-sphere reactions present alarge and important class of electrocatalytic reactionsas demonstrated by a large body of computationalworks aiming to evaluate rate constants for this classof reactions[11, 12, 20, 19, 21, 26, 27, 28, 29, 30,31, 32]. However important adiabatic reactions are,all electrochemical reactions are certainly not inner-sphere nor adiabatic. In particular, both vibronicand electronic non-adiabatic effects are frequentlyencountered in outer-sphere and long-range ET, PT,and PCET reactions.[25, 33] Even for electrocatalyticreactions, non-adiabaticity may be present and theimportance and contribution of non-adiabaticity maydepend on the electrode potential.[34, 35] As a concreteexample, it has been shown that only the inclusion ofvibronic non-adiabaticity in electrochemical hydrogenevolution reaction can explain experimentally observedTafel slopes and kinetic isotope effects.[34]
Second, there are several reactions where thePT and ET are decoupled for kinetic reasons.For example, in alkaline ORR pure ET has beenproposed as the rate determining step[36, 37, 38, 39].Recent experiments of ORR on carbon-based materialsconclusively demonstrate that ET is the rate- andpotential-determining step.[40, 41]. Also solution pHcan alter the reaction mechanism and e.g. CO2
reduction can proceed through simultaneous PCETin acidic and through decoupled PCET (ET-PT) inalkaline solutions[42, 43]. In general, decoupled ETand PT are expected to play an important role onweakly bonding electrode surfaces in oxygen, CO2,CO, alcohol etc. reduction reactions.[44] In suchreaction-catalyst combinations long-range ET/PT maytake place warranting the inclusion non-adiabaticityeffects. From an applied perspective, decoupledsteps may enable circumvention of thermodynamicscaling relations and lead to identification of novelelectrocatalysts.[45]
Third, ET, PT, and PCET include the transferof very light particles and therefore quantum effectsmay be very important. Especially nuclear tunnelinghas a long tradition in electrochemistry[46] andexperiments have conclusively demonstrated that
room-temperature hydrogen tunneling takes placeduring ORR on Pt, and at low over-potentialstunneling is the prevalent reaction pathway.[47]Tunneling contributions are rarely considered inthe field of computational electrocatalysis which ismainly due to tradition and methodological difficulties;the computational electrochemistry community hasadopted tools and classical transition state theory(TST) from computational heterogenous catalysiswhere reactions take place at high temperatures andquantum effects are considered negligible. On the otherhand, the theoretical electrochemistry community hastraditionally considered ET, PT, and PCET in thenon-adiabatic, tunneling framework[36, 48, 49, 50,51, 52, 53, 34, 54, 55, 56, 57, 58, 59, 60, 61, 62].The computational community has been slow inadopting the language and approaches developed inthe theory community which has resulted scarcity offirst principles study of tunneling in electrochemicalenvironments.
Fourth, theoretical electrochemistry has a longtradition of using model Hamiltonian formulationsto understand reaction kinetics. For instance,Marcus[62], Dogonadze-Kutzetnotsov-Levich[48, 49],Schmickler-Newns-Anderson[63, 64], or Soudackov-Hammes-Schiffer[33, 34, 53, 60, 61, 65] theories haveprovided the basis for understanding electrochemicalkinetics. The main drawback of these methods is thatthey are difficult to parametrize in a self-consistentmanner and require effective parameters obtainedfrom either experiments, simple DFT calculations,or a mixture of these. Yet, widely differentparametrizations for the same reaction can resultin similar rates. For instance, differences as largesas ∼ 3-4 eV in reorganization energy and thecoupling matrix elements[66, 67, 35, 68] lead topractically identical reaction rates; it is clear thatsome unphysical error or parameter cancellation takesplace. The difficulty of parameter estimation anderror cancellation limits the physical/chemical insightobtained from model Hamiltonians. Furthermore,model Hamiltonians are static and (usually) not self-consistent. Typically, the electrode potential servesto role of changing the Fermi-level in an otherwisestatic electronic structure. Even when potential-dependent electrostatic interactions and work termsare included [35, 51, 67, 69], most parameters suchas the solvent reorganization energy, chemical bondingcharacterized by Morse potentials, electrode structure,tunneling matrix elements etc. remain unchanged bythe electrode potential. As such, it is unlikely thatmodel Hamiltonians can quantitatively capture thecomplexity of electrochemical reactions. Besides issuesrelated to self-consistency, model Hamiltonians studiesof non-adiabatic reactions implicitly rely on the single
Grand canonical rate theory 4
orbital picture which is highly problematic for firstprinciples Hamiltonians as discussed in the SupportingInformation Section 1. Instead, modern fixed potentialfirst-principles methods explicitly incorporate theeffect of electrode potentials on the interfacialproperties and bonding. Especially the GCE-DFThas proven to provide a well balanced and rigorousdescription electrochemical interfaces. However,using general first principles methods for addressingET/PCET kinetics in general have remained largelyelusive thus far.
The above discussion highlights how differentreactions and phenomena have been and can beaddressed in the theoretical and computationalcommunities. Computational works utilize high-quality ab initio Hamiltonians but rate constants arebased on tools derived from heterogeneous catalysisand electrocatalytic reactions have been studiedonly using classical adiabatic TST theory. Thesecomputational studies describe the electrochemicalinterfaces in a self-consistent way and there is noneed for empirical parametrization of the TST rateequation. Thus far, these methods have only givenaccess to the reaction barrier but not the prefactorbeyond the TST approximation. Estimates onimportance of the prefactor has relied on perturbativerate theories with model Hamiltonians at the non-adiabatic limit to describe electron/proton tunneling.Other theoretical works extend the Newns-Anderson-Schmickler model Hamiltonian to study both classicaladiabatic TST and non-adiabatic tunneling reactions.While both barriers and prefactor have been computed,the models are evaluated using non-self-consistentparametrization. Therefore researchers have beenbe faced with a difficult choice: Should the studyinclude all the complexity addressed in a self-consistentmanner using an ab initio approach but with therestriction of classical TST approximation withoutgeneral prefactors? Or should the studies includeprefactors to reflect non-adiabaticity or tunneling butwith a empirically-parametrized model Hamiltonian?
In this work this difficulty is resolved bydeveloping a generally valid electrochemical rate theorywhich can be directly combined with fixed-potentialab initio methods. This is achieved by deriving agrand canonical rate theory building on Miller’s generalequilibrium (micro)canonical rate theory [70, 71, 72].As Miller’s theoretical framework is equally validfor adiabatic and non-adiabatic as well as quantum,semiclassical, and classical rate expressions[73] and canutilize both model or first principles Hamiltonians[33,57, 58, 59, 60, 61, 74, 75] the presented novel GCEextension provides a generally valid electrochemicalrate theory; the developed GCE rate theory enablesusing all canonical rate theories in constant potential
simulations. In particular, the work herein provides aunified rate theory for computing reaction barriers aswell as the prefactors making the theory applicable totreat adiabatic and non-adiabatic reactions, classicaland tunneling reactions, and PT, ET, and PCET onequal footing using GCE-DFT methods.
Besides developing a general and exact GCE ratetheory, approximate techniques for adiabatic reactionsare developed; non-adiabatic reactions are treatedusing the same formalism in a future publications.First, for adiabatic ET, PT and PCET reactionsa generalized GCE transition state theory (TST) isderived. Second, adiabatic Marcus-like[62] empiricalvalence bond theories (GCE-EVB) are developed.These lead to well-defined non-linear free energyrelationships ideally suited for materials’ screeningpurposes with kinetic information as demonstrated forthe acidic Volmer reaction on Au(111) in Section 4.Crucially, the developed rate theories can be seamlesslycombined with modern computational methods basedon (GCE-)DFT to facilitate self-consistent evaluationof rate constants without experimental parameters.The fixed potential rate theory will expand the type ofsystems, conditions, and phenomena in electrocatalysisamenable for first principles modelling.
The paper is organized as follows. In Section2 a general rate theory and TST within GCE aredeveloped. Rest of the paper focuses on ET andPCET kinetics within GCE. Section 3 shows howthe adiabatic barrier and rate of ET and PCETreactions are computed using GCE-EVB and freeenergy perturbation theory to developed a fixedpotential version of Marcus theory. Tafel slopes andother useful quantities as extracted from GCE-EVBare analyzed. A simple computational demonstrationof the GCE-EVB for Au-catalyzed Volmer reaction ispresented in Section 4. Next, additional computationalaspects for evaluating the rate constants are discussedin 5. Finally, the advances and results are summarized.
2. Rate theory in the grand canonicalensemble
2.1. Ensemble considerations
The GCE is open and the system exchanges matterwith its surroundings. The thermodynamics ofGCE are well understood[76] and we have recentlyshown that both electrochemical thermodynamicquantities for both classical and quantum particlescan obtained rigorously from GCE multi-componentDFT[3]. GCE provides a rigorous and natural wayto compute all thermodynamic expectation values atfixed electrode potentials by including the electrodepotential explicitly in the ab initio Hamiltonian. Thisis also the case for rate constants and fixed potential
Grand canonical rate theory 5
rate constants are GCE expectation values of canonicalrate constants as shown below.
To address the GCE rate constants, one needsto consider the dynamics of open (quantum) systemswhich is still an active area of research.[77, 78]. Thetreatment of open system dynamics directly affects theGCE rate theory. First, GCE phase space volume isnot globally conserved and the Liouville theorem doesnot hold in general and computed ensemble propertieswill depend on time if the system is not in equilibriumor is non-stationary.[79, 78, 80]. For the presentwork it is important that equilibrium and short-timeproperties are unique and time-independent in theGCE[79, 81]. At other times the expectation valuesdepend sensitively on the coupling between the systemwith the particle reservoir and introduces the reservoirtime scales.[80] As a result, time-dependent quantitiessuch as particle fluxes and correlation functionsentering the general flux formulation of rate theory(see below) would require extensive sampling andcareful computation.[78, 80] To avoid the treatmentof explicitly time-dependent quantities, the GCEtheory developed herein only utilizes equilibrium andinstantaneous quantities. Therefore, non-equilibriumprocesses cannot be treated using the approach takenhere. Neglecting the bath time-scale and coupling alsomeans that electron transfer kinetics from the electron”bath” to system (see Figure 1) are assumed fast,a condition satisfied by well-conducting electrodes.Neither of the the above restrictions on treating thebath coupling and time scale are expected to greatlyaffect the use or validity of the developed GCE ratetheory in electrochemical and electrocatalytic systems.
A related consideration from on the treatmentof open systems is particle conservation. If a quan-tum system is characterized by particle conserv-ing operators (H Hamiltonian, S entropy, and Nparticle number), even time-dependent observablesare obtained as ensemble weighted (pn) expecta-
tion values from O(t) = Tr[ρU(t0, t)O(t)U(t, t0)
]=∑
n pn 〈ψn|U(t0, t)O(t)U(t, t0)|ψn〉. Note, that changesbetween states with different number of particles arenot included in the propagator when both the propa-gator U and the operator O are particle conserving.[82]Hence, even explicit propagation of the wave func-tion does not allow sudden jumps in particle numbers.Therefore, in the extension of (micro)canonical ratetheory to the GCE, only particle conserving reactionsare considered. Then, all equilibrium quantities arealways well-defined but jumps between states with un-equal number of particles are suppressed. While thisis not an issue for adiabatic reactions with smoothchanges in the number of particles, the prefactors en-tering e.g. non-adiabatic rate constants need to beformulated so that particle conservation is respected.
Therefore, all rate expressions derived herein will onlyutilize particle conserving operators.
2.2. General grand canonical rate theory
After establishing the particle conserving and equilib-rium nature of the rate constants, the GCE rate con-stants can be formulated. To allow various types of re-actions to be described, the exact equilibrium canonicalrate expression due to Miller[70, 71, 72, 83] is adopted:
k(T, V,N)QI =
∫dEP (E) exp[−βE] = lim
t→∞Cfs(t)
(1)where QI is the canonical partition function of
the initial state, and β = (kBT )−1. The firstexpression is written in terms of transition probabilityat a given energy P (E). The second expressionutilizes a canonical flux-side correlation function
Cfs(t) =1
(2πh)f∫dpfdqf exp(−βH)δ[f(q)]qh[f(qt)]
for f degrees of freedom. δ[f(q)] constrains thetrajectories to start from the dividing surface, q is theinitial flux along the reaction coordinate, and h[f(qt)]is the side function which includes the dynamicinformation whether a trajectory is reactive or not.
Based on the discussion in Section 2.1 on thedynamics of open systems, only the t → 0+
and t → ∞ should be considered for the flux-side correlation function in the equilibrium rateexpressions. Depending on the choice of P (E) orH and h[f ] non-adiabatic and adiabatic (nuclear)quantum effects are included in the rate.[84, 85, 86, 87].It is noteworthy that P (E) and Cfs are computedusing only particle conserving operators[71] and theconditions discussed above are satisfied when (1) isused as the starting point for formulating GCE rateconstants.
To compute reaction rates at fixed potentials astraight-forward, yet novel, extension of the canonicalrate theory to the GCE is made:
k(µ, V, T )ΞI =1
2π
∞∑N=0
exp[βµN ]
∫ ∞−∞
dE exp[−βEN ]P (EN )
=
∞∑N=0
exp[βµN ]k(T, V,N)Q0 = limt→∞
Cµfs(t)
(2)
where ΞI = exp[βµN ]QI is the initial state grandpartition function and k(T, V,N) was introduced in(1). Above, N is the number of species (nuclearor electronic) in the system and Cµfs is the GCEflux-side function. The previous equation showsthat all canonical rate equations can be applied to
Grand canonical rate theory 6
electrochemistry within GCE approach and that fixedpotential electrochemical rate constants are GCEaveraged canonical rates constants.
The above equations are completely general andvarious flavors of rate theories can be extractedby invoking different Hamiltonians and transitionprobabilities, but they are somewhat cumbersomefor computational purposes. Indeed, it would beconvenient if the GCE rates could be directly evaluatedwithout explicitly summing over different particlenumbers. One way to achieve this is to makethe transition state theory (TST) assumption[72, 71,70] but generalized to GCE herein. In canonicalTST, the instantaneous limt→0+ Cfs(t) is consideredcorresponding to the assumption that there areno recrossings of the dividing surface. Bothquantum/classical and adiabatic/non-adiabatic TSTsare written as [88, 89, 90, 91]
kTST (T, V,N)QI(T, V,N) = limt→0+
Cfs(t) (3)
and the exact rate is recovered by introducing acorrection
k(T, V,N) = limt→∞
κ(t)kTST (T, V,N)
with κ(t) =Cfs(t)
Cfs(t→ 0+)
(4)
where κ(t) is the time-dependent transmis-sion coefficient which at long times is κ =k(T, V,N)/kTST (T, V,N).[92] Inserting this equationin (2) results in the most general grand canonical rateconstant. Significantly simplified rate constants are ob-tained when focusing on classical nuclei and using TST.As derived in the SI Section 2, for classical nuclei theTST result is [71, 72]:
k(T, V, µ)ΞI =
∞∑N=0
exp[βµN ]
∫dEPcl(E) exp[−βE]
≈∑N
exp[βµN ]kBT
hQ† ≡ kBT
hΆ
(5)
where Pcl(E) denotes transition probability forclassical nuclei but the electrons are of course quantummechanical[75, 93] with details given in [72] and theSI Section 2. The previous equation shows that thestructure of GCE-TST and canonical TST are similarwhich is true for open system in general if memoryeffects are neglected[94]. To obtain the GCE rateconstant without invoking the TST approximation, onecan use the transmission coefficient κ to write
k(T, V, µ) =
∑∞N=0 exp[βµN ]κ(T, V,N)
kBT
hQ‡
ΞI
≈ 〈κµ〉kBT
h
Ξ‡
ΞI= 〈κµ〉
kBT
hexp[−β∆Ω‡
](6)
where it is assumed that an effective transitionprobability 〈κµ〉 can be used. To complete thederivation for the classical GCE rate constant, therate is expressed in terms of grand energies with thedefinition Ωi = − ln(Ξi)/β and ∆Ω‡ = Ω‡−ΩI for theGCE barrier. Above the only new assumption besidesgrand canonical equilibrium distribution and TST, isthat the flux out of the transition state 〈κµ〉 can betreated as an expectation value and separated from thebarrier. For large enough systems and small variationsin the particle number this is a justified assumption.
The above development establishes the generalfixed chemical potential rate theory. For classical,adiabatic reactions the rate constants in GCE areessentially the same as in the canonical ensemble.Within TST approximation the rate constant isdetermined by the grand free energy barrier andeffective prefactor. The transmission coefficient needsto be approximated but this depends on the case athand; examples for the adiabatic and non-adiabaticharmonic GCE-TSTs expression valid for fully opensystem are derived in Supporting Information section3. A more thorough treatment on the theoryand computation of non-adiabatic and tunnelingcorrections will be presented in forthcoming work.
2.3. Semi-grand canonical ensemble
The above development is valid when both nuclearand electronic subsystems are open. A significantsimplification results if one assumes that the reactionrate does not explicitly depend on the number of somenuclei in the system. In a typical first principlescalculation this simplification is often exploited whenthe system can be divided to two subsystem: 1)classical electrolyte species consisting of nuclei andelectrons and 2) electrode + reactants treated eitherclassically or quantum mechanically. Typically thenumber of nuclei constituting the electrode andreactant are fixed while the electrolyte and electronchemical potential are fixed. Fixing only theelectron and electrolyte chemical potentials definesa semi-grand canonical ensemble used for derivingthe thermodynamics of electrocatalytic systems withinGCE-DFT[3]. In this treatment is often utilizedin e.g. Poisson-Boltzmann type models where theelectrolyte is at a fixed chemical potential but theenergetics do not explicitly depend on the number of
Grand canonical rate theory 7
electrolyte species. Then, summation over the numberof electrode/reactant nuclei or the electrolyte speciesis not needed.
Herein the semi-GCE is applied to derive rateconstants as a function of the electrode potential.From now on, I assume that the reaction rates dependexplicitly only on the number and/or chemical potentialof electrons in the system. Then, the state of thesystem is determined by T , V , number of nuclei ofthe electrode+reactant NN , chemical potential of theelectrolyte, chemical potential of the electrons µn, andnumber of electrons in the system N unless explicitlyspecified otherwise. Electroneutrality is maintainedby the electrolyte. A widely utilized harmonic TSTrate for constant number of nuclei and constantelectrochemical potentials are derived in section 3 ofthe Supporting Information.
3. Adiabatic barriers and rates fromGCE-EVB
To compute the GCE-TST rate at a given electrodepotential, the grand energy barrier in (6) needs tobe obtained. For electronically adiabatic reactionsmethods like the constant-potential[20] nudged elasticband[95] can be used. An alternative method forcomputing the grand energy barrier is to formulatea Marcus-like[62] approach or empirical valence bond(EVB) theory[96] within GCE. Such models arecommonly utilized in electron[62] and proton transfertheories.[65, 96, 97, 98, 99] Here, the treatmentis based novel development of GCE diabatic statesand the extension of the canonical thermodynamicperturbation theory to the GCE to facilitate derivationof a GCE-EVB rate theory (see SI sections 3 and 4).The GCE-EVB theory provides a theoretically well-justified and computationally affordable way to obtainfixed potential barriers at various electrode potentials;the adiabatic barrier needs to be explicitly computedonly at a single electrode potential while barriers atother potentials can be obtained using well-definedextrapolation of (17). The utility of the GCE-EVBtheory is demonstrated in Section 4.
In canonical EVB and Marcus theories usediabatic states, effective wave functions and freeenergies[62]. This can be extended to GCE byusing two fixed potential, diabatic ground statesurfaces which represent a GCE-statistical mixtureof states with probabilities given by the densityoperator in GCE[3]. Importantly, the diabatic statesobtained using the GCE density operator naturallyinclude many-body effects of the coupled electrode-reactant-solvent system and the complexity of theelectrochemical interface is explicitly included in themodel. Also, there is no need to decompose the
rate constants to orbital dependent quantities(seeSection 1 in the Supporting Information for additionaldiscussion). Then, two grand canonical diabatic all-electron wave functions are used to form an effectivediabatic GCE Hamiltonian. This is analogous tomolecular Marcus theory utilizing a canonical diabaticHamiltonian containing an initial (oxidized) I andfinal(reduced) molecule F .
Following the treatment in the SupportingInformation Section 3, a diabatic 2×2 grand canonicalHamiltonian in (7) can be formed from two diabaticGCE states. The resulting form is analogous to thecanonical EVB methods[96], electron[62], proton[98,99] and proton-coupled electron[65] theories. Thepresent form is, however, crucially different from itspredecessors; based on the approach developed in thiswork, all quantities are defined and computed at fixedelectrode potentials. In the basis two GCE diabaticstates the GCE Hamiltonian is
HGCE−dia =
[ΩII ΩIFΩFI ΩFF
](7)
as derived in Supporting Information Section 3.Here the diagonal elements are the grand energies ofthe initial (II) and final (FF) systems. The off-diagonalelements account for the interaction and mixingbetween the initial and final states. They can becomputed as GCE expectation values of contributionsfrom different N-electron states ψNi as ΩFI =∑N pN (µ)
⟨ψNF∣∣HN
∣∣ψNI ⟩. This is rather straight-forward for ET reactions using e.g. constrainedDFT discussed in Sections 4 and 5. For PT andPCET reactions computing these matrix elementswould require computing the vibronic matrix elementsusing e.g. the semiclassical approach of Georgievskiiand Stuchebrukhov[100] This direct computation isparticularly useful for non-adiabatic rate constantswhich are investigated in future work.
For adiabatic reactions, the direct calculation canbe replaced by the diagonalization of the 2×2 diabaticHamiltonian in Eq. (7). This diagonalizating producesthe adiabatic ground and first excited states as
Ω±ad =1
2
(ΩII + ΩFF ±
√(ΩII − ΩFF )2 + 4Ω2
IF
)(8)
As shown below, the diabatic states cross (ΩII =ΩFF ) at the transition state. This makes it possible tocompute the coupling matrix element as the differencebetween the diabatic states and the adiabatic states.For the ground state one has ΩIF = ΩII − Ω−ad. Theadiabatic ground transition state grand free energycan be computed using e.g. NEB calculations andthe coupling matrix element is simply the difference
Grand canonical rate theory 8
between the diabatic transition state grand energy andthe adiabatic one as shown below in Eq. (18).
Finally, the (diabatic) grand canonical statescorrespond to a single electron density which isguaranteed by the Hohenberg-Kohn-Mermin[4, 3] tobe unique for a given electrode potential. Bydefinition, GCE diabatic states are unique groundstates. Such diabatic statas also include the interactionand exchange between all the electrons in the systemand for adiabatic, ground state reactions there is noneed to include addition excited states despite thecontinuum of (single-electron) states of the electrode.In principle it is possible to add other, possibly excitedstates as basis states but here the focus in on treatingadiabatic reaction and excited states beyond the firstexcited state are neglected. If a general quantummechanical Hamiltonian is used, bond breaking isnaturally included in the GCE-EVB model. Theonly disambiguity is the definition of diabatic states.In practice the GCE diabatic energies, (ΩII andΩFF ), can be computed directly by applying usinge.g. cDFT[101, 102, 103] with fixed potential DFTas discussed in Section 5 and shown in Section 4.
3.1. Computation of diabatic GCE energy surfacesand barriers
An approach often used in molecular simulationsfor constructing the diabatic free energy curves isto sample the diabatic potentials along a suitablereaction coordinate. For canonical ET, PT, andPCET reactions the reaction coordinate is the en-ergy gap between the two diabatic states as shownby Zusman[104] and Warshel[105]: ∆Egap(R) =EF (R) − EI(R).[76, 106] From the sampled energygap, free energy curves are obtained as A(R) =−kBT ln(p(Egap(R))) + c. If the distribution is Gaus-sian
(p(Egap(R)) = c exp
[−(∆Egap − 〈∆Egap〉)2/2σ2
]),
the resulting free energy curves a parabolic. The dia-batic barrier in EVB or Marcus theory is then obtainedfrom the intersection of the initial and final diabaticcurves[106, 107, 108, 109].
Within GCE, the energy gap is simply Egap(R;µ) =∑N,i pN,iEgap(Ri, N). As shown in the SI section
4, the gap distributions can be formulated and com-puted by generalizing Zwanzig’s[110] canonical free en-ergy perturbation theory to the GCE. This provides arigorous way to derive the reaction barrier in termsof diabatic states and energies as presented in theSupporting Information Section 4. The reaction en-ergy rate can be computed from the initial-final stateenergy gap distribution functions using a well-knownformula[105, 111, 112, 113, 114, 115, 116]
kIF = κexp[−βgI(∆E‡)
]∫d∆E exp[−βgI(∆E)]
= κpI(∆E‡) (9)
where gi(∆E) is the free energy curve in statei as a function of the energy gap, pI(∆E
‡) isthe gap distribution at the transitions state, and κdenotes an effective prefactor. The reaction rate isdetermined by the energy gap distribution functionpI(∆E) = 〈δ(∆E(R)−∆E)〉I from equation (S30) ofthe Supporting information.
While the approach is general and valid forcomplex reactions, assuming that Egap(R;µ) isGaussian leads to a closed form equation. In thiscase the GCE-diabatic states are parabolic and theMarcus barrier in GCE is given by (13). As shownin the Sections 4 of the SI, the (Gaussian) gapdistribution may be derived using a second ordercumulant expansion resulting in
pI(∆E) =1√
2πσIexp
[−
(∆E − 〈∆E〉I)2
2σ2I
](10)
where 〈∆E〉I is the energy gap expectationvalue in the initial state obtained from equa-tion (S27) in the Supporting Information andσI =
⟨(∆EI − 〈∆EI〉)2
⟩Iis the gap variance. The
Marcus relation is then obtain after standardmanipulations[106, 112] yielding
pI(∆E‡) =
1√4kBTΛ
exp
[−β (∆Ω + Λ)2
4Λ
](11)
where σ2I = σ2
F = 2kBTΛ = kBT (〈∆E〉I −〈∆E〉F ), Λ is the fixed potential reorganization energyand ∆Ω = (〈∆E〉I + 〈∆E〉F )/2 is the reactiongrand energy as depicted in Figure 2. These gapidentities are valid for symmetric reactions and havebeen previously established well for the canonicalensemble[112] and generalized here to the GCE. Inpractice, the reorganization energy is computed asan average of the reorganization energies which aredifferences of the diabatic free energy at the finalgeometries
Λ =1
2[ΩII(RF )− ΩII(RF )I + ΩFF (RI)− ΩFF (RI)]
=1
2[ΛI + ΛF ]
(12)
shown in Figure 2. Finally, the GCE-EVB rateequation using the above assumptions results in anexpression analogous to Marcus equation
Grand canonical rate theory 9
k =κ√
4kBTΛexp
[−β (∆Ω + Λ)2
4Λ
](13)
Figure 2. Schematic depiction of the important GCE-EVBquantities. The blue (orange) dashed lines is initial (final)diabatic surface while the black solid line is the adiabatic surface.
The energy barrier of (13) is the diabatic energybarrier. The adiabatic barrier is estimated from (7)using the methods discussed in Section 3.2 below.One caveat to keep in mind is the more involvedcomputation of κ within the GCE as discussed inSection 2 and forthcoming work for non-adiabaticreactions. The above result may safely be used whenκ ≈ 1 for all particle numbers meaning that thereaction is always fully adiabatic and classical.
3.2. Implications of the canonical GCE-EVB ratetheory
If the diabatic grand energy surfaces are symmetricand quadratic they have the same curvature andreorganization energy. In this case, the diabatic grandenergy barrier is estimated from (13). The assumptionon equal curvature can be relaxed[117] (see also SIsection 5). One easy approach to realize this is toutilize an asymmetry parameter αas as[118]
αas =ΛI − ΛFΛI + ΛF
(14)
in terms of the reorganization energies for boththe initial and final states ΛI and ΛF , respectively.The transition state is located at the crossing point
x‡/ξ = − 1
αas+
1
αas
√1− αas
(αas −
4∆Ω
ΛI + ΛF
)(15)
With these definitions the asymmetric diabaticMarcus barrier and rate become
∆Ω‡ =1
4ΛI(x‡/ξ − 1
)2(16a)
k ≈ κ√4kBTΛI
1 + αas1 + αasx‡/ξ
exp[−β∆Ω‡
](16b)
When αas → 0, the regular Marcus barrier andcrossing point are obtained. In Figure 3 the effect ofasymmetry and reaction energy to the reaction barrierand location of the transition state are compared. Itcan be seen that both the barrier heights and itslocation are affected by the asymmetry and reactionenergy.
Figure 3. Left: EVB curves at different different asymmetriesαas. The final state reorganization energy is ΛF = 40 while theinitial state reorganization energy ΛI ∈ [10, 80]. The reactionenergy is ∆Ω = 0 for all curves. Right: EVB curves as afunction of the reaction energy: ∆Ω ∈ [−15, 15] and ΛF =40.The blue (red) curve corresponds to ΛI = 40 (ΛI=60). Both:The dashed line at x = 0 indicates the position of the transitionstate when ΛI = ΛF and ∆Ω = 0. The curve crossing point
equals ∆Ω‡dia.
While the Marcus-like equation results in adiabatic barrier, the adiabatic reaction barrier can beextracted from the diabatic barrier by diagonalizing(7). The adiabatic barrier can also be obtained from(13) using the Hwang-Aqvist-Warshel adiabaticitycorrection[119, 120]
∆Ω‡ad,EV B =(∆Ω + Λ)2
4Λ− ΩIF (x‡) +
(ΩIF (xI))2
∆Ω + Λ
= ∆Ω‡dia − ΩIF (x‡) +(ΩIF (xI))2
∆Ω + Λ(17)
where ΩIF is the off-diagonal matrix of the GCE-EVB Hamiltonian in (7). If the Condon approximationis used, the above equation is greatly simplified asΩIF ≈ ΩIF (x‡) ≈ ΩIF (xI) becomes a geometry-independent constant.
Next changes in the adiabatic GCE-EVB barrieras function of the parameters is analyzed. Fromthe schematics shown in Figures 2 and 3, one canobserve that changes of the minima along the reactioncoordinate correspond to horizontal displacements of
Grand canonical rate theory 10
the diabatic states and changes in Λ. Vertical changescorrespond to changes in the reaction grand energy∆Ω. In general, the reorganization energy of inner-sphere reactions taking place on or near the electrodesurface may depend on the electrode potential andinvestigations along this direction are on their way.
Under equilibrium conditions, ∆Ω = 0 and thecorresponding reorganization energy Λ0, the adiabaticbarrier is
∆Ω0,‡ad,EV B =
Λ
4− ΩIF +
(ΩIF )2
Λ≈ Λ0
4− ΩIF (18)
which leads to Λ0 = 4(∆Ω0,‡ad,EV B + ΩIF ) ≈
4∆Ω0,‡dia assuming that ΩIF << Λ0 (this is the case
for e.g. the Au-catalyzed Volmer reaction in Section4). At the equilibrium point, the overpotential is,η = ∆Ω = 0. Assuming for a moment that Λ ≈ Λ0 andreplacing the solution for Λ0 in (17) gives the diabaticbarrier as
∆Ω‡dia = ∆Ω0,‡dia +
∆Ω
2+
(∆Ω)2
16∆Ω0,‡dia
(19)
Inserting (19) in (17) results in the adiabaticreaction barrier as
∆Ω‡ad,EV B = ∆Ω0,‡ad,EV B +
∆Ω
2+
(∆Ω)2
16∆Ω0,‡dia
(20)
This result is well-known in the canonical EVBand Marcus theories. However, in GCE this relationis valid only when constant reorganization energyis assumed. In general, the driving force can bemanipulated easily by changing the electrode potentialwhich is in turn directly related to the absoluteelectron electrochemical potential as EM (abs) ∼ −µn.[3, 121, 122] An experimentally meaningful approach isto study −∂r(T, V, µn)/∂µn as done in a Tafel analysis,for example. Tafel analysis can also be understoodin a more general context of Brønsted-Evans-Polanyi(BEP) and other free energy relations measuring thechange of reaction rate when the reaction energy ischanged[123, 124, 125], as both Tafel and BEP analysesmeasure the reaction rate as a function the reactiondriving force - Tafel analysis focuses on the over-potential and BEP on the free energy. These twoquantities are linked by |∆η| = |∆µn| = |∂∆Ω/∂n|.Defining the rate constant as a function of the electrodepotential E as k(E) = k(E = 0)A(E) exp(−βαE) interms of the prefactor A and the Tafel-BEP coefficientα and the Tafel-BEP coefficient is [2, 123, 124]
d ln k(E)
dE≈ −βα→ α = −β−1 d ln k(E)
dE(21)
where constant α and prefactor are assumed.Within GCE-EVB α is obtained in terms of thereorganization and reaction energies
−βα =d ln k(∆Ω,Λ)
dE
= −∂ ln k
∂∆Ω
∂∆Ω
∂µn
∂µn∂E− ∂ ln k
∂Λ
∂Λ
∂µn
∂µn∂E
= −γ∆Ω′ − c∆Λ′
(22)
where the first term measures how the ratechanges as a function of the reaction energy, γ denotesa BEP coefficient and ∆Ω′ denotes the grand energychange as a function of the over-potential. Thesecond contribution is novel and unique to the GCEformulation. It measures the sensitivity of the rate tochanges in the reorganization energies as a functionalof the potential. This unconventional contribution canbe observed in e.g. the Volmer reaction treated in[67] with a model Hamiltonian and will be discussedin more detail in a future publication.
To facilitate understand the BEP term, onerecognizes that ∆Ω = (AF (〈NF 〉) − AI(〈NI〉) −µn(〈NF 〉 − 〈NI〉). For macroscopic systems, chemicalreactions have NF = NI while simple electrochemicalsteps have NF = NI ± 1. For chemical reactions∆Ω = ∆A and the variation ∆Ω′ is small. Within thecomputational hydrogen electrode (CHE) concept[126]the reaction energy ∆Ω ≈ ∆A0 ∓ η for PCET stepswith ∆A0 computed without any bias potential. Then,αCHE = γ for PCET steps and zero otherwise. Ingeneral such a simple relationship does not hold ingeneral and models such as GCE-DFT can be usedfor computing ∆Ω′ explicitly. Thus far, ∆Ω′ has beenreported in only few studies[20, 21, 127]. In these worksand in Section 4, ∆Ω is found to exhibit a roughlylinear dependence on the applied potential.
Next the BEP γ of Eq (22) is analyzed. Using thediabatic barriers in (19) (obtained assuming constantreorganization energy and constant prefactor), oneobtains
γ =∂ ln k(T, V, µn)
∂∆Ω
∣∣∣∣T,V
≈ −− β 1
2
[1 +
∆Ω
Λ0
](23)
which results in α = −∆Ω′(1/2 + ∆Ω/2Λ0).It is seen that γ is not a simple constant butdepends linearly on the reaction driving force. If thereorganization energy is small the dependence on thereaction grand energy becomes more pronounced asdemonstrated for the Au-Volmer reaction in Section 4.In general, non-linearity of the grand energy barrier hastwo contributions: non-linearity of the diabatic barrierand the potential-dependent reorganization energy.For macroscopic systems non-linearity is established by
Grand canonical rate theory 11
including the quadratic part of the diabatic barrier inthe model. Lately[29, 31, 20, 21] this has been observedcomputationally and it is pleasing that the GCE-EVBpicture seems qualitatively correct.
To summarize, the generalized BEP-Tafel rela-tionships have been derived from a microscopic per-spective using grand canonical rate theory. Both vari-ation in the reaction energy barrier and the transitionstate location as a function of the potential can bepredicted using just a few parameters. This is demon-strated for the acidic Volmer reaction in Section 4. Thegeneral form of the BEP-Tafel relation is given in (22).For small over-potentials, the rate is expected to de-pend linearly on the applied potential. For larger over-potentials non-linear dependence is predicted when thereorganization energy is small. The GCE perspectivealso predicts a novel potential-dependent reorganiza-tion energy which is supported by model Hamiltoniancalculations[67] and will be addressed carefully in fu-ture publications.
4. Application of the GCE-EVB theory to theAu-catalyzed Volmer reaction
Here the first demonstration of the GCE ratetheory is provided. I consider the acidic Volmerreaction i.e. proton discharge which is arguablythe simplest and yet relevant electrocatalytic reactionfor hydrogen production and other electrocatalyticreactions. Similarly, gold can be considered as thesimplest electrode material. Yet, the Volmer reaction,even on gold, is not fully understood[128] and thereaction is considered to exhibit nuclear quantumeffects and even vibronic non-adiabaticity.[34, 35] Asa first application of the theory and methodologyderived and developed in this work, I consider mostlyan adiabatic and classical model for the acidic Volmerreaction – quantum effects and a non-adiabaticity arestudied separately in forthcoming publications. Theresults are discussed in the GCE-EVB framework ofSection 3.
The Volmer reaction is modelled as a singlehydronium ion on a 3x3x5 Au(111) surface as shownin Figure 5. The needed free energies were computedusing GCE-DFT as implemented in GPAW[129] withinthe surface-jellium approach[20] with a continuumsolvent model for water[130]. This approach givesall the thermodynamic quantities at a fixed electrodepotential. The potential-dependent minimum energypathways are computed using a nine image nudgedelastic band[95] (NEB) discretization. Geometries andNEB pathways were considered converged when themaximum force was below 0.05 eV/A.
Constant-potential diabatic states and reorganiza-tion energies were computed using constrained DFT
(cDFT) as implemented in GPAW[131]. As in thecanonical case, the constraining potential in GCE-cDFT is introduced as an external potential to theGCE-DFT giving
Ω[n(r), Vc;T, V, µn]cDFT =
Ω[n(r);T, V, µn]DFT + Vc∑σ
(∫drwσc (r)nσ(r)−Nc
)(24)
where Ω[n(r);T, V, µn]DFT is the GCE-DFT energyfunctional[3] and n(r) is the electron density. wσi (r)is the weight function which defines how the charge isto be partitioned, i.e. the regions where charge is to belocalized, Nc is the desired number of excess electronswithin the constrained region, and Vc is the Lagrangemultiplier enforcing the charge/spin localization. Theintroduction of constraining terms in Eq. (24) leads toa new effective potential defined as
vσeff =δΩ[n(r), Vc]
δn(r)=δΩKS [n(r)]
δn(r)+∑c
Vcwσc (r) (25)
Thus, the cDFT potential is just the sum ofthe usual KS potential and the constraining potentialwhich is also used in the self-consistent calculation.The constraint is further enforced by introducing theconvergence criteria
C ≥∣∣∣∣∑σ
∫drwσc (r)nσ(r)−Nc
∣∣∣∣ ,∀ c (26)
The optimize grand canonical free energy underthe specified constraint is[131]
Ω[Vc] = maxVc
minnΩ[n, Vc;T, V, µn]cDFT (27)
To fulfill the cDFT constraints and to performfix potential calculations, the approach in Fig. 4 hasbeen utilized. An example of GPAW scripts used forperforming GCE-cDFT calculations is given in the SIsection 6.
The charge states to define diabatic states arechosen as Nc = +1 state for the hydronium (H3O
+)and neutral Nc = 0 for the final water and adsorbedhydrogen (H2O+H∗). The reorganization energies arecomputed from Eq. (12) at the equilibrium potentialwhich resulted in ΛF = 2.1 eV for the initial stategeometry, and ΛI = 3.2 eV for the final state geometry.The average Λ = 2.65 eV is used in calculating theMarcus barrier of Eq. (13). These cDFT computedreorganization values are in good agreement with the
Grand canonical rate theory 12
Figure 4. The two phase optimization loop used for obtainingfixed potential diabatic states. The cDFT optimization is giventhe outer loop of Eq. (27) and the SJM loop is the inneroptimization performed using the algorithm from Ref.[20] tochange the number of electrons N to satisfy the set electrodepotential
.
values used for model Hamiltonian parametrizations ofHuang[68] and Santos[66] but much larger than the oneused by Lam[67]. The reaction asymmetry from (14) is0.2 meaning that the transition state geometry alongthe reorganization coordinate is closer to the initialstate. Analysis of the TST geometries at differentpotentials shows that that reorganization energy is bestpresented either by the Au-O distance or the dihedralangle between the H2O and surface.
In Figure 5 the adiabatic reaction barrier asobtained from NEB calculation is plotted as a functionof the reaction energy corresponding to differentelectrode potentials. First, one observes that thebarriers are very small for all considered electrodepotentials. This is in line with explicit water DFTresults[20] at all electrode potentials and reactionenergies. The figure also shows the adiabatic TSTlocation as a function of the reaction energy fromboth NEB-DFT and extrapolation using (16a). Theextrapolation reproduces the TST geometries verywell, and captures the trends in the TST location.For comparison, explicit solvent calculations exhibit asimilar trend in the TST position as a function of thepotential[20] as the one found here using an implicitsolvent. This example demonstrates the Au-O distanceis good reorganization coordinate and that the TSTlocation is effectively captured by Eq. (16a). UnlikeDFT-NEB calculations, the GCE-EVB requires justone NEB and two reorganization energy calculationsto capture the TST geometry for a range of reactionenergies and electrode potentials.
Figure 5. Above: The reaction pathways at different potentials.The stars show the DFT-NEB computed TST geometry alongreorganization coordinate. The stars correspond to TSTgeometries predicted using equation (14a). Below:The fixedpotential Au(111) Volmer barrier as a function of the reactionenergy. NEB[*] refers to NEB calculations of the present paperwith an implicit solvent while NEB[20] are from [20] withexplicit, ice-like solvent. Also the barrier and reorganizationenergy used in the model Hamiltonian work of [67] are shownand extrapolated using (20).
The results in Figure 5 also show the GCE-EVBbarrier as obtained from Eq. (20) which is evaluatedusing the cDFT computed average reaorganizationenergy. The adiabatic equilibrium barrier (∆Ω0,‡
ad inEq. (20)) is computed from a NEB calculation close tothe equilibrium potential. As the results in Figure 5show, the extrapolation with Eq. (20) provides a veryaccurate way for computing the adiabatic potential-dependent energy barrier. It is also observed thatthe estimate for the reorganization energy used in themodel Hamiltonian work of [67] is very small (∼ 0.3eV) and cannot be used for predicting barriers usingGCE-EVB.
Next, the validity of the adiabatic assumptionis tested using the coupling matrix elements of Eq.
Grand canonical rate theory 13
(7). The coupling matrix element is obtained usingEq. (8) at the equilibrium potential transition stateand results in ΩIF = 0.37eV . The coupling constantmaybe used for estimating transmission coefficient 〈κµ〉of Eq. (6) the adiabaticity from the the Landau-Zenerfactor (PLZ) for PCET reaction [132, 35] adopted tothe GCE.
〈κµ〉 =2PLZ
1 + PLZ(28a)
PLZ = 1− exp[−2πγ] (28b)
2πγ =π3/2 |ΩIF |2
hνn√
ΛkBT(28c)
and 〈κµ〉 = 1 is a signature of an electronicallyadiabatic reaction whereas non-adiabatic reactionshave 〈κµ〉 << 1. Eq. (28a) is valid for reactions in thenormal region.[132, 35] νn is the vibrational frequencyalong the reaction coordinate.[133] As discussed above,the reaction coordinate n the Au-catalyzed Volmer isthe water reorientation and νn = 1/τL ≈ 0.5× 109s−1
computed from the water reorientation time τL ≈ 2ps[134]. The expression for γ in Eq. (28c) is validfor quadratic free energy surfaces as derived in Eq.S.15 of the Supporting Information. Evaluating thetransmission coefficient at room temperature usingthe computed average reorganization energy, couplingmatrix element, and water reorganization time gives〈κµ〉 = 1.0 - this shows that the reaction is adiabaticand justifies the treatment in this section.
Besides enabling a reliable prediction of reactionbarriers and TST geometries, the results demonstratethe first GCE-cDFT calculations. Besides showingthat GCE-cDFT is technically possible, the resultsshow that ab initio computed diabatic states offer newinsight to electrocatalytic reactions. In particular, theresults provide a proof-of-principle that GCE-EVB canbe used to accurately estimate barriers using just asingle NEB calculations and a few cDFT calculationswith an expense similar to a standard DFT calculation.
5. Discussion
The distinct advantage of the formalism and theory de-veloped in this paper is that all rate equations can bereadily evaluated with GCE-DFT or other first princi-ples approaches. The presented formalism enables thetreatment electrochemical and electrocatalytic thermo-dynamics and kinetics in terms of the prefactor andbarriers in the same self-consistent framework – theGCE-DFT. Therefore, the same DFT-based tools canbe used to address inner-sphere and outer-sphere kinet-ics and thermodynamics instead of modifying or chang-ing the theoretical and computational framework fordifferent reaction steps[36].
By construction the rate constants include theinterplay between the electronic structure, solvent,electrode potential etc. All quantities can becomputed using self-consistent DFT energies and”wave functions” to include exchange and correlationeffects between all the electrons in the system. TheFermi-Dirac distribution is fulfilled at the DFT leveland, therefore, there is no need to integrate overthe filled/empty orbitals weighted by the Fermi-Diracdistribution in the rate expression as done is traditionalsingle-orbital descriptions (see SI Section 1 and below).Also, the Kohn-Sham-Mermin theorem[3] guaranteesthat GCE-DFT and GCE-EVB states are unique toa given electrode potential and that the GCE-EVBdiabatic inlcude that interactions between all electronsin the metal and the reactants.
The electrode potential is self-consistently treatedand all free energies and prefactors depend explicitlyon the potential. This is in contrast with traditionalmodel Hamiltonian treatments where the electrode po-tential rigidly shifts the Fermi-level without modify-ing any interactions or prefactors [64, 69] or modifiesonly the electrostatic interactions[34, 67]. Also, sep-arate computation of work terms[67, 69, 68, 135] isnot needed because all relevant interaction can be di-rectly included in the general Hamiltonian. Evaluationof chemisorption functions entering adiabatic Newns-Anderson-based models[63, 66, 67, 64, 68, 135] is alsoavoided. Therefore, the current models are free of ap-proximate treatment of semi-elliptic DOSs[67, 66, 135]or fitting the chemisorption functions to a computedDOS[66, 135].
As the developed rate theory utilizes generalab initio Hamiltonians, bond formation/breakingare naturally included. This is again in contrastwith model Hamiltonians which require approximatepotential-independent terms to describe changes inatomic bonding [136, 137, 135]. Instead, asdemonstrated herein, ET, PT, or PCET and bondrupture/formation are naturally captured with GCE-DFT. Bond formation s is also captured by diabaticmodels using cDFT as demonstrated herein forthe Volmer reaction and previously for ET[138],PCET[139] and general chemical reactions[140, 141].
As all necessary terms can be computed fromGCE-(c)DFT, adoption and evaluation of the rate ex-pressions is straight-forward (but potentially labori-ous). While applicability and usefulness of combinedDFT and GCE-EVB was demonstrated for the Volmerreaction, it is worth discussing the additional compu-tational requirements in some detail. First, the simu-lation of charged systems is needed to sample the elec-trode potential. Electroneutrality can be enforced us-ing some variant of the Poisson-Boltzmann model, fordetails see [3]. Fixed potential calculations can be ac-
Grand canonical rate theory 14
complished within a single SCF cycle[10], or iteratively[142, 20]. Second, the solvent effects should be in-cluded in the model. In traditional TST-based modelsfor adiabatic reactions the main solvent contributionis thermodynamic and stems from (de)stabilization ofdifferent structures. GCE-EVB models need to involvethe solvent as the reaction barrier is directly related tothe solvent/environment reorganization energy and ne-glecting the solvent contributions will most likely leadto incorrect results.
Given a software capable of handling charged sys-tems and performing constant potential calculations,adiabatic TST rate constants can be readily evaluated.As shown for the Volmer reaction, reaction barriersand adiabatic prefactors at constant potential are ob-tainable using e.g. the NEB[95] method. Evalua-tion of GCE-EVB rate constants requires additionalsoftware capabilities for constructing charge/spin lo-calized diabatic states and to evaluate the electroniccoupling between these states either through directcalculation or diagonalization of the diabatic Hamil-tonian of Eq. (7). Also the reorganization energy,which is an excited state quantity, needs to be com-puted. One widely implemented and available toolfor evaluating the additional parameters is the cDFTmethodology[101, 102, 103] which is implemented inseveral DFT codes[131, 143, 144, 145, 146, 147, 148,149, 150, 151, 152]. The extension of cDFT to GCEwas, for the first time, demonstrated and applied inthis work. GCE-EVB should be accompanied with aconstant potential simulations to compute fixed poten-tial reaction and reorganization energies.
The general framework can also be extendedto treating reactions beyond adiabatic reactions toinclude e.g. non-adiabatic effects, nuclear tunneling,and solvent-dynamics controlled reactions - treatmentof these effects is under current study and willbe published separately. Non-adiabatic effects areexpected for e.g. outer-sphere ET reactions and severalPCET reactions[33]. Several PT and PCET reactionsare also likely include adiabatic or non-adiabaticnuclear tunnelling effects. Also, solvent dynamicsshould be included as these are likely to becomeincreasingly important or even dominant when thereaction is adiabatic and the reaction barrier becomesvery small or vanishes.[153] Under such conditionsthe reorganization will be the slowest process andthe reaction prefactor should reflect this. Last, well-defined interpolation[154, 155, 156] between adiabatic– solvent dynamic –non-adiabatic should be developedor adapted to the fixed-potential rate theory.
Finally, a spectacular feature of canonical Marcusand EVB theory is the observation of an inverted regioni.e. the rate constant starts to decline as the reactionbecomes more exothermic. However, the inverted
region has not been observed for electrochemicalreactions even at large over-potentials. The grandcanonical Marcus rate of (13) seems to predictan inverted region for highly exothermic conditionsand warrants additional discussion. First, theMarcus-like expression is based on linear response orsecond order cumulant treatment[157] which leads toquadratic free energies along the reaction coordinate- the prediction of the inverted region is a directconsequence of the linear response assumption. Togo beyond the quadratic free energy surfaces higherorder cumulants can be added (see SI section 5)to modify the existence of the inverted region.[158]For instance, the inverted region is not predicted forMorse potentials.[159] Secondly, the inverted regionis very sensitive to tunneling effects, excited states,and solvent effects.[160] For example, nuclear quantumeffects are needed to achieve accurate ET rates in theinverted region ET[161] while excited proton vibronicstates dominate PCET reaction rate[159]. Even ifthe inherent approximations on quadratic, classical,and adiabatic nature are acceptable, in GCE-EVBthe reorgnization energy is potential-dependent andthe inverted region is suppressed if ∆Ω′ ≈ 0. Alsothe reorganization energy should saturate at largeoverpotentials.
The prediction of non-existing inverted regionis inherent to several non-adiabatic single orbitalapproaches. In these treatments the inverted regionis avoided by taking the manifold of single-electronstates into account and integrating over the Fermi-Dirac weighted transition rates, as discussed inSupporting Information Section 1. This has resultedin e.g. Marcus-Hush-Chidsey[162], Dogonadze-Levich-Kuztnetsov[49, 48], Soudackov-Hammes-Schiffer[53]models of ET and PCET. On the other hand, adiabaticrate computed using the Newns-Anderson-SchmicklerHamiltonian [63, 64] does not the include the orbital-to-orbital contributions separately as the redox orbitalis coupled with all levels on the metal by definition.In the adiabatic Newns-Anderson-Schmickler model,the free energy surface becomes as single well and thereactant and product become indistinguishable. As thevanishing barrier cannot be treated[63, 154], the barrieris simply extrapolated to zero in the inverted regionand the rate is controlled by the prefactor.
Based on the above discussion, the inverted regionis more complicated than the normal region andrequires careful consideration of excited states andtunneling effect, for example. The vanishing invertedregion is for non-adiabatic single-orbital models andnot for adiabatic models where the exchange betweenall metallic states and the redox molecule are treated.In the GCE-EVB picture the inverted region is inherentto the quadratic grand energy surfaces and simply
Grand canonical rate theory 15
extrapolation of the barrier to zero removes theinverted region. However, this is not fully satisfactoryand more studies are needed to study non-quadraticfree energy surfaces, tunneling, and excited states inthe inverted region.
6. Conclusions
In this work a new theoretical formulation forcomputing electrochemical and electrocatalytic rateconstants at a fixed potential is developed. The rateexpressions are obtained by extending the universallyvalid and exact canonical rate theory[70, 71, 72] tothe grand canonical, fixed potential ensemble. Generalconditions and limitations for the fixed potentialrate theory are developed. It is shown that allrate theories developed within the canonical ensemblecan be transferred to the GCE; electrochemical rateconstants are ”just” GCE-weighted canonical rateconstants. This is conceptually important becausethe fixed-potential rate theory enables treating allpotential-driven reactions within a single formalisminstead of relying on separate theories for differentcases.
This provides a unified framework for computingand understanding both the barrier and prefactor froma single formalism towards modelling beyond adiabaticinner-sphere reactions to include e.g. non-adiabaticityand tunneling. Herein, specific rate expressions arederived for typical electrocatalytic adiabatic ET, PT,and PCET reactions. Fixed potential rate expressionhave been derived for i) general electrocatalyticreactions with ((5)) and without ((2)) the TSTapproximation, ii) electronically adiabatic ET, PT andPCET reactions using a grand canonical Marcus-likeGCE-EVB theory in (13).
The GCE-EVB formalism enables computing thegrand energy barrier in terms of fixed potentialreorganization energy and the reaction grand energyin analogy with the canonical EVB or Marcus theory.GCE-EVB can explain and predict the electrocatalytic”Marcus-like” behavior in energy barriers and TSTgeometries as a function of the thermodynamic drivingforce. GCE-EVB enables also the computation ofnon-linear energy relationships and Tafel slopes andgeneral BEP-Tafel relations in goo accuracy using justfew DFT parameters; this has shown for the Volmerreaction using fixed potential (constrained) DFTdeveloped here. Also, quantitatively accurate barriersand TST geometries can be predicted using a few self-consistent GCE-DFT calculations. These features areexpected to make the GCE-EVB approach particularlysuitable for electrocatalyst screening studies.
The developed theory can be directly combinedwith modern, solid-state ab initio methods to capture
the complexity of the electrochemical interface atconstant potential in a self-consistent manner. Inthis sense, the model is fully ab initio and allparameters can be directly computed. A set of widelyimplemented DFT-based tools suffices to compute allthe needed parameters in a self-consistent manner.This should enable the computational communityadopt the theoretical framework and to progress froma thermodynamics-based description of electrocatalysisto addressing also electrocatalytic kinetics underexperimentally realistic conditions.
The advances herein enable further developmentof theory and computational methods to address e.g.tunnelling pathways and non-adiabaticity in electro-chemical systems from first principles. Understand-ing and controlling (non-adiabatic) tunneling can openup new reaction pathways to avoid constraining scal-ing relations[163, 164, 165] encountered for adiabaticPCET reactions. Besides applications, the advancedrate theories will improve fundamental understandingof electrochemical kinetics in e.g. de-coupled and non-adiabatic ET, PT, and PCET. This contributions areespecially important for weakly-binding catalysts, butneglected in computational studies thus far.
7. Acknowledgements
I acknowledge support by the Alfred KordelinFoundation and the Academy of Finland (Project No.307853). Computational resources were provided byCSC – IT CENTER FOR SCIENCE LTD. I alsowish to thank the anonymous reviewer for constructivecriticism and for providing comments to improve themanuscript.
8. Declaration of interest
Declarations of interest: none
9. References
[1] Seh Z W, Kibsgaard J, Dickens C F, Chorkendorff I,Nørskov J K and Jaramillo T F 2017 Science 355
[2] Allen J Bard L R F 2001 Electrochemical Methods:Fundamentals and Applications, 2nd Edition (JohnWiley & Sons)
[3] Melander M M, Kuisma M J, Christensen T E K andHonkala K 2019 The Journal of Chemical Physics 150041706
[4] Mermin N D 1965 Phys. Rev. 137(5A) A1441–A1443[5] Pribram-Jones A, Pittalis S, Gross E K U and Burke
K 2014 Thermal density functional theory in contextFrontiers and Challenges in Warm Dense Matter edGraziani F, Desjarlais M P, Redmer R and Trickey S B(Springer International Publishing) pp 25–60
[6] Evans R 1979 Advances in Physics 28 143–200[7] Kreibich T, van Leeuwen R and Gross E K U 2008 Phys.
Rev. A 78(2) 022501
Grand canonical rate theory 16
[8] Capitani J F, Nalewajski R F and Parr R G 1982 TheJournal of Chemical Physics 76 568–573
[9] Hansen J P H J P 2006 Theory of Simple Liquids, 3rdEdition (Academic Press)
[10] Sundararaman R, GoddardIII W A and Arias T A 2017The Journal of Chemical Physics 146 114104
[11] Taylor C D, Wasileski S A, Filhol J S and Neurock M 2006Phys. Rev. B 73(16) 165402
[12] Goodpaster J D, Bell A T and Head-Gordon M 2016 TheJournal of Physical Chemistry Letters 7 1471–1477
[13] Otani M and Sugino O 2006 Phys. Rev. B 73(11) 115407[14] Jinnouchi R and Anderson A B 2008 Phys. Rev. B 77(24)
245417[15] Skulason E, Tripkovic V, Bjørketun M E, Gudmundsdottir
S, Karlberg G, Rossmeisl J, Bligaard T, Jonsson H andNørskov J K 2010 The Journal of Physical ChemistryC 114 18182–18197
[16] Letchworth-Weaver K and Arias T A 2012 Phys. Rev. B86(7) 075140
[17] Fang Y H and Liu Z P 2010 Journal of the AmericanChemical Society 132 18214–18222
[18] Skulason E, Karlberg G S, Rossmeisl J, Bligaard T,Greeley J, Jonsson H and Nørskov J K 2007 Phys.Chem. Chem. Phys. 9(25) 3241–3250
[19] Chan K and Nørskov J K 2015 The Journal ofPhysical Chemistry Letters 6 2663–2668 (Preprinthttps://doi.org/10.1021/acs.jpclett.5b01043)
[20] Kastlunger G, Lindgren P and Peterson A A 2018 TheJournal of Physical Chemistry C 122 12771–12781
[21] Lindgren P, Kastlunger G and Peterson A A 2020 ACSCatalysis 10 121–128
[22] Exner K S 2019 The Journal of Physical Chemistry C 12316921–16928
[23] Quaino P; Juarez F S E S W 2014 Beilstein J.Nanotechnol. 5 846–854
[24] Kuo D Y, Paik H, Kloppenburg J, Faeth B, Shen K M,Schlom D G, Hautier G and Suntivich J 2018 Journalof the American Chemical Society 140 17597–17605
[25] He Z D, Chen Y X, Santos E and Schmickler W 2018Angewandte Chemie International Edition 57 7948–7956
[26] Xiao H, Cheng T and Goddard W A 2017 Journal of theAmerican Chemical Society 139 130–136
[27] Xiao H, Cheng T, Goddard W A and Sundararaman R2016 Journal of the American Chemical Society 138483–486
[28] Zhang H, Goddard W A, Lu Q and Cheng M J 2018 Phys.Chem. Chem. Phys. 20(4) 2549–2557
[29] Huang Y, Nielsen R J and Goddard W A 0 Journal of theAmerican Chemical Society 0 null
[30] Holmberg N and Laasonen K 2015 The Journal of PhysicalChemistry C 119 16166–16178
[31] Akhade S A, Bernstein N J, Esopi M R, Regula M J andJanik M J 2017 Catalysis Today 288 63 – 73
[32] Tripkovic V, Bjorketun M E, Skulason E and Rossmeisl J2011 Phys. Rev. B 84(11) 115452
[33] Hammes-Schiffer S and Stuchebrukhov A A 2010 ChemicalReviews 110 6939–6960
[34] Goldsmith Z K, Lam Y C, Soudackov A V and Hammes-Schiffer S 2019 Journal of the American ChemicalSociety 141 1084–1090
[35] Lam Y C, Soudackov A V and Hammes-Schiffer S 2019The Journal of Physical Chemistry Letters 10 5312–5317
[36] Ignaczak A, Nazmutdinov R, Goduljan A, de Cam-pos Pinto L M, Juarez F, Quaino P, Santos E andSchmickler W 2016 Nano Energy 29 362 – 368
[37] Ignaczak A, Nazmutdinov R, Goduljan A, de Cam-pos Pinto L M, Juarez F, Quaino P, Belletti G, SantosE and Schmickler W 2017 Electrocatalysis 8 554–564
[38] Malko D and Kucernak A 2017 Electrochemistry Commu-nications 83 67 – 71
[39] Tse E C M, Varnell J A, Hoang T T H and GewirthA A 2016 The Journal of Physical Chemistry Letters7 3542–3547
[40] Bukas V J, Kim H W, Sengpiel R, Knudsen K, Voss J,McCloskey B D and Luntz A C 2018 ACS Catalysis 811940–11951
[41] Kim H W, Ross M B, Kornienko N, Zhang L, Guo J, YangP and McCloskey B D 2018 Nature Catalysis 1 282–290
[42] Gottle A J and Koper M T M 2017 Chem. Sci. 8(1) 458–465
[43] Verma S, Hamasaki Y, Kim C, Huang W, Lu S, JhongH R M, Gewirth A A, Fujigaya T, Nakashima N andKenis P J A 2018 ACS Energy Letters 3 193–198
[44] Koper M T M 2013 Chem. Sci. 4(7) 2710–2723[45] Exner K S and Over H 2019 ACS Catalysis 9 6755–6765[46] Sakaushi K, Lyalin A and Taketsugu T 2020 Current
Opinion in Electrochemistry 19 96 – 105[47] Sakaushi K, Lyalin A, Taketsugu T and Uosaki K 2018
Phys. Rev. Lett. 121(23) 236001[48] Dogonadze R and Kuznetsov A 1975 Progress in Surface
Science 6 1 – 41[49] Dogonadze R 1971 3. theory of molecular electrode kinetics
Reactions of Molecules at Electrodes ed Hush N (Wiley-Intersciences) pp 135–228
[50] Ignaczak A, Nazmutdinov R, Goduljan A, de Cam-pos Pinto L M, Juarez F, Quaino P, Belletti G, SantosE and Schmickler W 2017 Electrocatalysis 8 554–564
[51] Goduljan A, de Campos Pinto L M, Juarez F, Santos Eand Schmickler W 2016 ChemPhysChem 17 500–505
[52] Schmickler W, Santos E, Bronshtein M and NazmutdinovR 2017 ChemPhysChem 18 111–116
[53] Venkataraman C, Soudackov A V and Hammes-Schiffer S2008 The Journal of Physical Chemistry C 112 12386–12397
[54] Navrotskaya I, Soudackov A V and Hammes-Schiffer S2008 The Journal of Chemical Physics 128 244712
[55] Ghosh S, Soudackov A V and Hammes-Schiffer S 2016Journal of Chemical Theory and Computation 122917–2925
[56] Ghosh S, Horvath S, Soudackov A V and Hammes-SchifferS 2014 Journal of Chemical Theory and Computation10 2091–2102
[57] Borgis D and Hynes J T 1993 Chemical Physics 170 315– 346
[58] Borgis D and Hynes J T 1996 The Journal of PhysicalChemistry 100 1118–1128
[59] Cukier R I 1996 The Journal of Physical Chemistry 10015428–15443
[60] Soudackov A and Hammes-Schiffer S 2000 The Journal ofChemical Physics 113 2385–2396
[61] Soudackov A, Hatcher E and Hammes-Schiffer S 2005 TheJournal of Chemical Physics 122 014505
[62] Marcus R A 1965 The Journal of Chemical Physics 43679–701
[63] Schmickler W 1986 Journal of Electroanalytical Chemistryand Interfacial Electrochemistry 204 31 – 43 ISSN0022-0728
[64] Schmickler W 2017 Russian Journal of Electrochemistry53 1182–1188
[65] Soudackov A and Hammes-Schiffer S 1999 The Journal ofChemical Physics 111 4672–4687
[66] Santos E, Lundin A, Potting K, Quaino P and SchmicklerW 2009 Phys. Rev. B 79(23) 235436
[67] Lam Y C, Soudackov A V, Goldsmith Z K and Hammes-Schiffer S 2019 The Journal of Physical Chemistry C123 12335–12345
[68] Huang J and Chen S 2018 The Journal of Phys-ical Chemistry C 122 26910–26921 (Preprint
Grand canonical rate theory 17
https://doi.org/10.1021/acs.jpcc.8b07534) URLhttps://doi.org/10.1021/acs.jpcc.8b07534
[69] Nazmutdinov R R, Bronshtein M D and Santos E 2019 TheJournal of Physical Chemistry C 123 12346–12354
[70] Miller W H 1974 The Journal of Chemical Physics 611823–1834
[71] Miller W H, Schwartz S D and Tromp J W 1983 TheJournal of Chemical Physics 79 4889–4898
[72] Miller W H 1998 The Journal of Physical Chemistry A102 793–806
[73] Richardson J O 2012 Ring-polymer approaches toinstanton theory Ph.D. thesis University of Cambridge
[74] Kretchmer J S and Miller T F 2013 The Journal ofChemical Physics 138 134109
[75] Ananth N and III T F M 2012 Molecular Physics 1101009–1015
[76] Tuckerman M 2010 Statistical Mechanics: Theory andMolecular Simulations (Oxford University Press)
[77] Weiss U 2012 Quantum Dissipative Systems4th ed (WORLD SCIENTIFIC) (Preprinthttps://www.worldscientific.com/doi/pdf/10.1142/8334)URL https://www.worldscientific.com/doi/abs/10.1142/8334
[78] Delle-Site L and Praprotnik M 2017 Physics Reports693 1 – 56 ISSN 0370-1573 molecular systemswith open boundaries: Theory and Simulation URLhttp://www.sciencedirect.com/science/article/pii/S0370157317301539
[79] Agarwal A, Zhu J, Hartmann C, Wang H and SiteL D 2015 New Journal of Physics 17 083042 URLhttp://stacks.iop.org/1367-2630/17/i=8/a=083042
[80] Delle Site L 2016 Phys. Rev. E 93(2) 022130 URLhttps://link.aps.org/doi/10.1103/PhysRevE.93.022130
[81] Peters M H 1998 eprint arXiv:physics/9809039[82] Gianluca Stefanucci R v L 2012 Nonequilibrium Many-
Body Theory pf Quantum Systems: A ModernIntroduction (Cambridge University Press) chap 4, pp81–124
[83] Miller W H 1975 The Journal of Chemical Physics 621899–1906
[84] Agarwal A and Delle Site L 2015 The Journal of ChemicalPhysics 143 094102
[85] Habershon S, Manolopoulos D E, Markland T E and MillerT F 2013 Annual Review of Physical Chemistry 64 387–413
[86] Richardson J O and Thoss M 2014 The Journal ofChemical Physics 141 074106
[87] Richardson J O and Thoss M 2013 The Journal ofChemical Physics 139 031102
[88] Hele T J H and Althorpe S C 2016 The Journal ofChemical Physics 144 174107
[89] Althorpe S C and Hele T J H 2013 The Journal ofChemical Physics 139 084115
[90] Richardson J O and Althorpe S C 2009 The Journal ofChemical Physics 131 214106
[91] Richardson J O 2018 International Reviews in PhysicalChemistry 37 171–216
[92] Henriksen N E and Hansen F Y 2002 Phys. Chem. Chem.Phys. 4(24) 5995–6000
[93] Richardson J O, Bauer R and Thoss M 2015 The Journalof Chemical Physics 143 134115
[94] Chandler D 1978 The Journal of Chemical Physics 682959–2970
[95] Henkelman G, Uberuaga B P and Jonsson H 2000 TheJournal of Chemical Physics 113 9901–9904
[96] Warshel A and Weiss R M 1980 Journal of the AmericanChemical Society 102 6218–6226
[97] Kamerlin S C L and Warshel A 2011 Wiley Interdisci-plinary Reviews: Computational Molecular Science 130–45
[98] Schmitt U W and Voth G A 1998 The Journal of PhysicalChemistry B 102 5547–5551
[99] Vuilleumier R and Borgis D 1998 The Journal of PhysicalChemistry B 102 4261–4264
[100] Georgievskii Y and Stuchebrukhov A A 2000 The Journalof Chemical Physics 113 10438–10450
[101] Wu Q and Van Voorhis T 2005Phys. Rev. A 72(2) 024502 URLhttp://link.aps.org/doi/10.1103/PhysRevA.72.024502
[102] Wu Q and Van Voorhis T 2006 J. Chem. Phys. 125 164105[103] Kaduk B, Kowalczyk T and Voorhis T V 2012 Chemical
Review 112 321–370[104] Zusman L 1980 Chemical Physics 49 295 – 304 ISSN 0301-
0104[105] Warshel A 1982 The Journal of Physical Chemistry 86
2218–2224[106] Blumberger J 2015 Chemical Reviews 115 11191–11238[107] Nitzan A 2006 Chemical Dynamics in Condendsed Phases:
Relaxation, Transfer, and Reactions in CondensedMolecular Systems (Oxford University Press)
[108] Dogonadze R R, Kuznetsov A M and Chernenko A A 1965Russ. Chem. Rev. 34 759
[109] JOM Bockris S K 1979 Quantum Electrochemistry(Plenum Press)
[110] Zwanzig R W 1954 The Journal of Chemical Physics 221420–1426
[111] Tachiya M 1989 The Journal of Physical Chemistry 937050–7052
[112] Tateyama Y, Blumberger J, Sprik M and Tavernelli I 2005The Journal of Chemical Physics 122 234505
[113] Vuilleumier R, Tay K A, Jeanmairet G, Borgis D andBoutin A 2012 Journal of the American ChemicalSociety 134 2067–2074
[114] Rose D A and Benjamin I 1994 The Journal of ChemicalPhysics 100 3545–3555
[115] Zhou H X and Szabo A 1995 The Journal of ChemicalPhysics 103 3481–3494
[116] King G and Warshel A 1990 The Journal of ChemicalPhysics 93 8682–8692
[117] Small D W, Matyushov D V and Voth G A2003 Journal of the American Chemical Soci-ety 125 7470–7478 pMID: 12797822 (Preprinthttps://doi.org/10.1021/ja029595j) URLhttps://doi.org/10.1021/ja029595j
[118] Mattiat J and Richardson J O 2018 The Journal ofChemical Physics 148 102311
[119] Warshel A, Hwang J K and Aqvist J 1992 FaradayDiscuss. 93(0) 225–238
[120] Rosta E and Warshel A 2012 Journal of Chemical Theoryand Computation 8 3574–3585
[121] Trasatti S 1990 Electrochimica Acta 35 269 – 271[122] Trasatti S 1986 Journal of Electroanalytical Chemistry and
Interfacial Electrochemistry 209 417 – 428[123] Fletcher S 2010 Journal of Solid State Electrochemistry
14 705–739[124] Fletcher S 2009 Journal of Solid State Electrochemistry
13 537–549[125] Parsons R 1951 Trans. Faraday Soc. 47(0) 1332–1344[126] Nørskov J K, Rossmeisl J, Logadottir A, Lindqvist L,
Kitchin J R, Bligaard T and Jonsson H 2004 TheJournal of Physical Chemistry B 108 17886–17892
[127] Hormann N G, Andreussi O and Marzari N 2019 TheJournal of Chemical Physics 150 041730
[128] Dubouis N and Grimaud A 2019 Chem. Sci. 10(40) 9165–9181
[129] Enkovaara J, Rostgaard C, Mortensen J J, Chen J, Du lakM, Ferrighi L, Gavnholt J, Glinsvad C, Haikola V,Hansen H A, Kristoffersen H H, Kuisma M, LarsenA H, Lehtovaara L, Ljungberg M, Lopez-Acevedo O,Moses P G, Ojanen J, Olsen T, Petzold V, RomeroN A, Stausholm-Møller J, Strange M, Tritsaris G A,Vanin M, Walter M, Hammer B, Hakkinen H, Madsen
Grand canonical rate theory 18
G K H, Nieminen R M, Nørskov J K, Puska M, RantalaT T, Schiøtz J, Thygesen K S and Jacobsen K W 2010J. Phys. Condens. Matter 22 253202
[130] Held A and Walter M 2014 The Journal of ChemicalPhysics 141 174108
[131] Melander M, Jonsson E O, Mortensen J J, Vegge T andGarcıa Lastra J M 2016 Journal of Chemical Theoryand Computation 12 5367–5378
[132] Newton M D 1991 Chemical Reviews 91 767–792(Preprint https://doi.org/10.1021/cr00005a007)URL https://doi.org/10.1021/cr00005a007
[133] Calef D F and Wolynes P G 1983 The Journalof Physical Chemistry 87 3387–3400 (Preprinthttps://doi.org/10.1021/j100241a008) URLhttps://doi.org/10.1021/j100241a008
[134] Bakker H J 2008 Chemical Reviews 1081456–1473 pMID: 18361522 (Preprinthttps://doi.org/10.1021/cr0206622) URLhttps://doi.org/10.1021/cr0206622
[135] Huang J, Li P and Chen S 2019 The Journal ofPhysical Chemistry C 123 17325–17334 (Preprinthttps://doi.org/10.1021/acs.jpcc.9b03639) URLhttps://doi.org/10.1021/acs.jpcc.9b03639
[136] Santos E, Koper M and Schmickler W 2008 ChemicalPhysics 344 195 – 201
[137] Ignaczak A and Schmickler W 2003 Journal of Electroan-alytical Chemistry 554-555 201 – 209
[138] Park H, Kumar N, Melander M, Vegge T, Garcia LastraJ M and Siegel D J 2018 Chemistry of Materials 30915–928
[139] Holmberg N and Laasonen K 2018 The Journal ofChemical Physics 149 104702
[140] Wu Q, Kaduk B and Van Voorhis T 2009 The Journal ofChemical Physics 130 034109
[141] Wu Q, Cheng C L and Van Voorhis T 2007 The Journalof Chemical Physics 127 164119
[142] Bonnet N, Morishita T, Sugino O and Otani M 2012 Phys.Rev. Lett. 109(26) 266101
[143] Souza A M, Rungger I, Pemmaraju C D, Schwingen-schloegl U and Sanvito S 2013 Phys. Rev. B 88(16)165112
[144] Wu Q, Kaduk B and Van Voorhis T 2009 J. Chem. Phys.130 034109
[145] Rezac J, Levy B, Demachy I and de la Lande A 2012 J.Chem. Theory Comput. 8 418–427
[146] Ramos P and Pavanello M 2016 Phys. Chem. Chem. Phys.18 21172–21178
[147] Oberhofer H and Blumberger J 2009 J. Chem. Phys. 131064101
[148] Oberhofer H and Blumberger J 2010 J. Chem. Phys. 133244105
[149] Ghosh P and Gebauer R 2010 J. Chem. Phys. 132 104102[150] Sena A M P, Miyazaki T and Bowler D R 2011 J. Chem.
Theory Comput. 7 884–889[151] Ratcliff L E, Grisanti L, Genovese L, Deutsch T, Neumann
T, Danilov D, Wenzel W, Beljonne D and Cornil J 2015J. Chem. Theory Comput. 11 2077–2086
[152] Holmberg N and Laasonen K 2017 Journal of ChemicalTheory and Computation 13 587–601
[153] Rips I and Jortner J 1987 The Journal of ChemicalPhysics 87 2090–2104
[154] Mishra A K and Waldeck D H 2009 The Journal ofPhysical Chemistry C 113 17904–17914
[155] Hynes J T 1986 The Journal of Physical Chemistry 903701–3706
[156] Gladkikh V, Burshtein A I and Rips I 2005 The Journalof Physical Chemistry A 109 4983–4988
[157] Georgievskii Y, Hsu C P and Marcus R A 1999 TheJournal of Chemical Physics 110 5307–5317
[158] Matyushov D V 2018 The Journal of
Chemical Physics 148 154501 (Preprinthttps://doi.org/10.1063/1.5022709) URLhttps://doi.org/10.1063/1.5022709
[159] Goldsmith Z K, Soudackov A V and Hammes-SchifferS 2019 Faraday Discuss. 216(0) 363–378 URLhttp://dx.doi.org/10.1039/C8FD00240A
[160] BIXON M and JORTNER J 1999 Electron Transfer -From Isolated Molecules to Biomolecules, Part One.(Advances in Chemical Physics vol 106) (Wiley-Interscience) chap Electron Transfer—from IsolatedMolecules to Biomolecules
[161] Heller E R and Richardson J O 2020 The Jour-nal of Chemical Physics 152 034106 (Preprinthttps://doi.org/10.1063/1.5137823) URLhttps://doi.org/10.1063/1.5137823
[162] CHIDSEY C E D 1991 Science 251 919–922[163] Huang Z F, Wang J, Peng Y, Jung C Y, Fisher A and
Wang X 2017 Advanced Energy Materials 7 1700544–n/a 1700544
[164] Back S and Jung Y 2017 ChemCatChem 9 3173–3179ISSN 1867-3899
[165] Kulkarni A, Siahrostami S, Patel A and Nørskov J K 2018Chemical Reviews 118 2302–2312
download fileview on ChemRxivmain_article.pdf (1.12 MiB)
Supporting information for ”Grand canonical rate
theory for electrochemical and electrocatalytic
systems I: General formulation and proton-coupled
electron transfer reactions”
Marko M. Melander
Nanoscience Center, P.O. Box 35 (YN) FI-40014, Department of Chemistry,
University of Jyvaskyla, Finland
E-mail: [email protected]
10 June 2020
Supporting Information 2
1. Problem of combining ”orbital-based” rate theories and first principles
methods: Choosing localized empty and filled orbitals
1.1. Orbital based electron transfer rate theories
There are two commonly used orbital-based approaches for describing charge transfers
rate at electrode surfaces. The first one was developed by Dogonadze, Levich, and
Kutznetsov (DLK)[1], who assumed a weak interaction between the donor and the
acceptor. Their treatment yields an expression similar to Marcus theory and the model
goes by various names including Marcus-Hush-Chidsey[2], Gerischer[3], Marcus-DOS[3]
or just the density-of-states (DOS) model. In the case of a metallic electrode, the
molecular orbitals interact with a continuum of electronic states from the metal and
therefore one needs to integrate over all the metallic bands. An implicit assumption
in the DOS model is that charge transfer takes place between two one-electron orbitals
rather than two many-electron wave functions. Also, the effect of the electrode potential
E is assumed to linearly change the occupation of orbitals without otherwise changing
the one-electron levels. For electrochemical charge transfer reactions require information
only on the molecular orbital ε0 and its DOS ρ0(ε0) and (quasi-continuum) of electrode
bands ε with DOS ρ(ε). In this case the charge transfer is [1, 4] (see also 1)
kDOS(E) =
∫dεW (ε, ε0)f(ε− E)ρ(ε)ρ0(ε0) (1)
where W is the transition probability and f is the Fermi-Dirac distribution.
Originally, the DLK model was derived for the weak interaction limit and harmonic
energy surfaces in which case the result is well-known [1, 4]
kDOS(E) ≈√
β
4πλ
∫ ∞−∞
dε|Hab(ε, ε0)|2f(ε− E)
exp
[−β (λ+ e0(E0 − E)− ε0)2
4λ
] (2)
Hab(ε, ε0) = 〈ψε0a |H|ψεb〉 denotes the Hamiltonian matrix element between the
molecule and electrode orbitals corresponding to energy levels ε0 and ε, respectively,
in the initial a and final b diabatic states. E is the electrode potential and E0 is the
formal equilibrium potential. Depending on the model used for the reactant DOS, the
weakly interacting limit by Dogonadze, Gerischer’s model with a Gaussian dependency
or Schmickler’s model (see below) maybe obtained as shown in Ref. [5]
The other approach is due to Schmickler[6, 7] and has been dubbed as the potential
energy curve (PEC) method[5]. The PEC method applies a modified Newns-Anderson
(N-A) Hamiltonian for building the PECs
Supporting Information 3
Figure 1. Conventional (left) and GCE Marcus theory (right). The conventional
theory is based on transitions between single-electron orbitals while the current GCE
framework utilized general many-electron wave functions.
HN−A = ε0n0 +∑k
εknk +∑k
(vk0c†kcr + vrkc
†0ck)
+1
2
∑i
hωi(p2i + q2
i ) + (n0 − z)∑i
(hωigiqi)(3)
where the terms describe reactant orbital, orbitals of the electrode, electron
exchange terms using coupling matrix elements v, harmonic bath at frequencies ωi,
momenta pi and coordinate qi while the last term couples the reactant at charge state
z to the harmonic bath. Connecting the initial and final states of the redox reaction
along a charge transfer coordinate rq and using N-A Hamiltonian, the PEC is
u(rq, εF ) =r2q
4λ+ (ε0 + rq − εF )〈n(rq)〉
+∆
2πln[(εF − rq − ε0) + ∆2]
(4)
where 〈n(rq)〉 = 1/2 + 1/π tan−1((εF − rq − ε0)/∆) is the charge at rq and
∆(ε′) = π∑
k|v0,k|2δ(ε′−εk) is the effective coupling constant. Then, the charge transfer
barrier is u‡ = u(rq = rmax, εF )− u(rq = 0, εF ) and the rate is
kPEC = κ exp[−βu‡
](5)
At the weak interaction limit, both the DOS and PEC models are in their essence
formulations of Fermi’s Golden rule describing electron transfer between single electron
orbitals.[7]
Supporting Information 4
1.2. Orbital based Fermi Golden rule formulation
Here a modern and general Golden rule is revisited to discuss inherent limitations of the
orbital-based formulation. For this purpose the initial (final) state at the initial (final)
geometry can be approximated by a single diabatic electronic state Ψ(Rinitial) ≈ |ψI〉 |χj〉and using the Hamiltonian in Eq.(6). A similar equation can also be written for adiabatic
states as shown in Ref. [8, 9]. In the diabatic Fermi Golden rule formulation the
Hamiltonian is[10, 11]
Hel =∑i
EIi
∣∣I i⟩ ⟨I i∣∣+∑f
EFf
∣∣F f⟩ ⟨F f∣∣
+∑if
∆if (∣∣I i⟩ ⟨F f
∣∣+∣∣F f⟩ ⟨If∣∣) (6)
where∣∣Kk
⟩= |ψk〉 |χk〉 is a vibronic wave function consisting of |ψk〉, a one
electron orbital and |χj〉 a nuclear wave function. Reaction rates are computed
using the general flux formulation presented in the main article with the following
transition probability and flux[10, 11]: P (E) =1
2(2πh)2 Tr
[F δ(E − HN)F δ(E − HN)
]and F = 1/h∆[|0〉 〈1| + |1〉 〈0|], respectively. ∆if is a general diabatic coupling term,
which in the Franck-Condon approximation is
∆if =∣∣∣ 〈ψf |V |ψi〉∣∣∣2∑
kl
|〈χk|χl〉|2
= |Vif |2∑kl
|〈χk|χl〉|2(7)
Following standard thermalized Fermi-golden rule derivation[12, 10, 11] for a
transition between two electronic states gives
kI→F =2π
h
∑i∈I,f∈F
ki→f (8a)
ki→f =|Vif |2∑
l exp[−Ei,l]∑lk
exp[−Ei,l]|〈χk|χl〉|2 × δ(Ei − Ef + El − Ek)
= |Vif |2F (Ei − Ef )(8b)
where F (Ei −Ef ) is the thermally averaged Franck-Condon factor (see Section ??
of the SI) . If the nuclear wave functions are taken to be those of a harmonic oscillator,
the Marcus barrier and the rate constant can be obtained from the derivation in Section
2 of the SI. Note that transition between all one electron orbitals are considered here.
To obtain the DOS-model for electron transfer, only a subset of the transition rates
is considered. Intuitively, for an reduction of a molecule, transitions from the localized
Supporting Information 5
occupied metal orbitals to empty orbitals localized at the molecule should be considered.
This leads to
kred =2π
h
∑i∈filled
∑f∈empty
ki→f
=2π
h
∑f∈empty
∫dεiρ(ε)f(ε− εF )ki→f
(9)
where the second equation highlights the close correspondence with the DOS
method (Eq. (1)), ρ(ε) is the DOS and f the Fermi-Dirac distribution. Note also
that the PEC method uses a Hamiltonian similar diabatic Hamiltonian here. In PEC
the total transition probabilities from initial to final state are also computed using
orbital-to-orbital formulation described above.
1.3. Choosing the orbitals
Both DOS and PEC share a fundamental open question: how does one choose the
localized and empty/filled orbitals? This situation is faced in a typical first-principles
calculations, where (canonical) one electron orbitals are highly delocalized even when
charge-localized diabatic states are used, making the choice of active orbitals difficult.
An important result learned from orbital localization methods[13, 14, 15, 16] is that
the energy from a single determinant method such as DFT or Hartree-Fock methods
is invariant to orbital rotation within the occupied molecular orbitals. Thus, occupied
orbitals can be localized using a unitary rotation which leaves the energy, and the total
wave function unchanged ; during this process the spatial shape and spread of filled
one electron orbitals are drastically changed. Also, the empty, virtual orbitals can be
localized separately. However, the filled and empty are not allowed to mix during the
localization to avoid changes in occupation of numbers[17]. As mixing the unoccupied-
occupied subspaces is forbidden, orbital localization is performed separately for the
empty and filled orbitals and consequently two different unitary transforms are required.
A concrete example helps to understand why the orbital localization causes practical
difficulties. Consider for example an outer-sphere ET from an electrode to O2 forming
a superoxide species. Here the initial state wave functions |I〉 would be occupied
orbitals localized on the metal and the final state orbitals |F 〉 would be empty states
localized on O2. After a normal DFT calculation, one performs a unitary transform on
both the initial and final states separately such that the the orbitals are well localized
to the molecule and metal for both states: |I〉 = U∣∣∣IfilledDFT
⟩and |F 〉 = V
∣∣F emptyDFT
⟩, with
U U † = 1 and V V † = 1. Note that nuclear wave function remain unchanged as the
electronic energy is unaffected by the transformation. Thus, the unitary transformation
leaves the thermally averaged Franck-Condon weight unchanged.
However, the electronic coupling elements for a given Vif = 〈i|H|f〉 change
drastically as the electronic orbitals are rotated. This is easily seen from the close
Supporting Information 6
correspondence[18] between the coupling and overlap elements 〈i|V |f〉 ≈ v 〈i|f〉, where
v is a constant. Changing from the localized to delocalized states is written as
〈i|f〉 = 〈idft|U †V |fdft〉 6= 〈idft|fdft〉 when final (empty) and initial (filled) canonical
DFT orbitals are localized separately. Only when U = V is the overlap between the
localized and canonical orbitals the same; this would require of mixing of the filled and
empty canonical orbitals resulting in changes in the total energy and the total wave
function and is therefore discouraged.
As a concrete example consider Eq. (9) where only states below the Fermi-level
contribute to the reduction rate. As potential is changed, some orbitals become empty
or occupied changing the driving force of the reaction which remains unchanged under
the orbital localization. The rate is dictated by transition probability directly related to
the matrix elements in non-adiabatic reactions. Thus, the rate of non-adiabatic electron
transfer reaction depends on how the orbitals are chosen and localized. This can lead
to inconsistent and incorrect interpretation of the electrochemical rate as a function of
the potential if basic single determinant methods are used to parametrize the DOS or
PEC models. Great care is needed when the methods and as emphisized in Ref. [19]
”this is not a failure of the computational methods used but is a consequence of how the
rate constant is defined by the phenomenological equations. It is therefore important to
choose the approach which is equivalent to the experiment or thought experiment that
the theory is attempting to reproduce”.
Based on the above discussion, a very important conclusion is reached: the
rate obtained from the orbital-based Fermi golden rule, DOS or PEC using only
the ”active orbitals” depend on the way the orbitals are localized. Therefore, one
needs to acknowledge that orbital localization needed when the orbital-based models
are parametrized using canonical DFT methods, leads to arbitrary changes in the
rate constant depending on the localization or rotation used scheme. Hence, while
the energy, density, and the total wave function remain unchanged after a unitary
transformation of the orbitals, single orbitals and single orbital overlaps will necessarily
be affected. Therefore, a unitary transformation such as orbital localization will
unphysically affect the rate obtained from methods using one-electron orbitals and
orbital-to-orbital transitions to compute the transition probability. If one electron-based
the DOS or PEC methods are parametrized using first-principles approaches, methods
such as fragment orbital DFT[20] methods might be applicable.
Care is also required when using many-body wave functions for computing the
rates. While unitary transform does not change any observables of a single diabatic
wave function, the off-diagonal matrix elements might be sensitive to orbital rotations.
However, in approaches such constrained DFT employed in this work, the coupling
elements are functionals of the electron densities of the initial and final state[21] and as
such in principle unaffected by orbital localization.
Supporting Information 7
2. Adiabatic and non-adiabatic harmonic TST rates
Here classical harmonic TST (HTST) for adiabatic and non-adiabatic reactions within
GCE are derived.
2.1. GCE-TST with classical nuclei
For classical nuclei, the general rate equation in the GCE is written in terms of the time-
integral of the flux correlation function contains all the dynamic information[22, 11]:
Pr(p,q) = limt→∞ h[f(qt)] =∫∞
0dtd
dth[f(qt)] =
∫∞0dtCff (qt,pt). With this definition
the classical TST rate is
k(T, V, µ)ΞI =∞∑N=0
exp[βµN ]
∫dEPcl(E) exp[−βE]
=∞∑N=0
exp[βµN ]
∫dpdq exp[−βH(p,q)]F (p,q)Pr(p,q)
(2πh)N
=∞∑N=0
exp[βµN ]
∫dpdq exp[−βH(p,q)]F (p,q)
∫∞0dtCff (t)
(2πh)N
≈∞∑N=0
exp[βµN ]kBT
hQ‡∫dtδ(t) =
∑N
exp[βµN ]kBT
hQ†
≡ kBT
hΆ
(10)
where on the second last line making the short time approximation[22] to Cff (t→
0) =kBT
hQ‡δ(t) leads to the TST expression.
2.2. Adiabatic HTST
The general TST rate equation is shown in Eq. 5 of the main article. First, consider
a general case where potential the number of both nuclei and electrons is allowed to
fluctuate. Usually, for NN classical nuclei the Hamiltonian in mass-weighted coordinates
(xi) and momenta Pi is written as Hcl =∑
i∈NNP 2i + V (xi). V (xi) defines the (Born-
Oppenheimer) potential energy surface.
Then consider a system is open to electrons at a fixed electron chemical potential
while number of nuclei is fixed. Also, the system is assumed adiabatic meaning
that the number and distribution of electrons adjusts instantaneously to the nuclear
configuration. This is the common situation considered in first principles calculations
at fixed electrode potential calculations. For this case, the Kohn-Sham-Mermin theorem
guarantees that electronic energy and distribution are unique to a given electron
chemical potential and external potential (here provided by the nuclei). Hence, the
potential energy V is not only a parametric function of the nuclear positions but also
Supporting Information 8
the chemical potential of the electrons. Furthermore, as shown in Ref. [23], the grand
free energy of the electrons is given by Ωn(T, V,NN , µn; xi). As the nuclei move the
on the effective potential energy surface provided by the electrons, one recognizes that
V (xi, µn) = Ωn(T, V,NN , µn; xi) (see Ref. [23]). Then, for the open electronic system,
the classical Hamiltonian for the nuclei is
H(NN)cl =∑i∈NN
P 2i + V (xi, µn) ≡ Hcl (11)
where I and ‡ denote the initial and transition states. The TST rate is written as
[12]
kTST (NN , V, T )QI =
∫N
dP
∫N
dx exp[−H‡clβ
]δ(f(x))(∇f · PN)h(∇f · PN) (12)
where f is the N − 1 dimensional dividing surface between the reactants and
products, ∇f · PN = Pn‡ is the momentum normal to f identified as the reaction
coordinate, h(∇f · PN) = h(Pn‡) is a step function separating the reactant ad product
basins, and δ(f(xN)) restricts the geometries to lie on the dividing surface. With these
definitions the canonical HTST at fixed electron chemical potentials follows from:
kHTST (T, V,NN) =
∫N
dP
∫N−1
dxPn‡
exp[−H‡clβ
]ZI
=
∫NdP∫N−1
dxPn‡ exp[−β(
∑Ni=0 1/2P 2
i + V (xi, µn)‡)]
∫NdP∫Ndx exp
[−β(
∑Ni=0 1/2P 2
i + V (xi, µn)I)]
=1√2πβ
∫N−1
dx exp[−βV (xi, µn)‡
]∫Ndx exp[−βV (xi, µn)I ]
≈ vN√2π
∏N−1i vi∏N−1i v‡i
exp[−β(Ω‡n − ΩI
n)]
=vN2π
exp[−β(Ω‡N − ΩI
N)]
=vN2π
exp[−∆Ω‡Nβ
]
(13)
where at the second last row the effective potentials are Taylor expanded in terms
of normal mode coordinates with corresponding frequencies vi and vN is the frequency
along the reaction coordinate:V ‡/I = Ω‡/IN + 1/2
∑i viq
2i . The last equality follows from
setting the nuclear vibrational entropy SN = kB ln(∏N−1
i vi/∏N−1
i v‡i
)and setting the
total grand free energy to ΩN = Ωn − TSN . Here the subscript N reminds that the
number of nuclei was kept fixed above. Note that Eq. (13) would be used in
typical first principles calculations at fixed electrode potentials where the
electron chemical potential and number of nuclei are fixed.
The above treatment can also be extended to treat situations in which both the
number of electrons and nuclei are allowed to fluctuate. This is straight-forward and
Supporting Information 9
can be obtained by. Inserting Eq. (13) in Eq. 5 of the main paper and applying Eq. 6
of the main paper leads to
kHTST (T, V, µ) =〈vN〉µ
2πexp[−∆Ω‡β
](14)
where 〈vN〉µ is the effective frequency along the reaction coordinate computed using
effective fixed potential PESs.
2.3. Non-adiabatic HTST
Next, non-adiabatic harmonic transition state theory (NA-HTST) approximation to
the rate is developed. Unlike for the canonical case, only a fixed number of nuclei is
treated. NA-HTST also requires the calculation of matrix elements HAB = 〈ΨA|H|ΨB〉.These HABs are defined only when |ΨA〉 and |ΨB〉 have the same number of both
electron and nuclei. Also, the adiabatic approximation cannot be used and the
electrons do not instantaneously adapt to nuclear positions. Hence, unlike for the
adiabatic case, constant electron number V (x, n) rather that constant electron potential
V (x, µn) is used. The appropriate Hamiltonian is given by Eq. (6), in which Hcl =∑i∈NN
P 2i /2mi + Vi(xi).
Using this Hamiltonian, assuming a quadratic potential V and applying the Golden
rule form the basis for NA-HTST. This derivation can be found in e.g. Ref [24].
Another path, presented below, is to use the classical transitions state theory using
the Landau-Zener transition Pr probability[25, 12] and assuming that the potential
energies are quadratic. Then, the following identities are used: The reorganization
energy and vibrational frequency along the reaction coordinate are related as λ =
2v2N∆q2 = 2mv2
N∆x2, where ∆q and ∆x are the geometric differences of the initial
and final states in mass weighted and cartesian coordinates states, respectively. The
differences of forces can written as gradient of the two parabolas at the transition state
as shown in Ref. [24] to yield |∆F |‡ = λ/∆x. With these definitions, fixed number
(canonical) electronic/nuclear NA-HTST can be derived:
Supporting Information 10
knaHTST (T, V,NN , Nn) =
∫N
dP
∫N−1
dxPrPn‡
exp[−H‡clβ
]ZI
=
∫N
dP
∫N−1
dx
(1− exp
[− 2π|HIF |2
h|Pn‡∇n‡(VI − VF )|
])Pn‡
exp[−H‡clβ
]ZI
linearize exp≈
∫N
dP
∫N−1
dx2π|HIF |2
h|Pn‡∇n‡(VI − VF )|Pn‡
exp[−H‡clβ
]ZI
forces=
∫N
dP
∫N−1
dx2π|HIF |2
hPn‡|∆F |Pn‡
exp[−H‡clβ
]ZI
integrate P=
2π|HIF |2
h|∆F |
∫N−1
dx exp[−V ‡β
]∫Ndx exp[−VIβ]
harmonic TST≈
√2πβ|HIF |2
h|∆F |vN
∏N−1i vi∏N−1i v‡i
exp[−(E‡ − EI)β
]vib. entropy
=√
2πβ|HIF |2
h|∆F |vN exp
[−∆A‡β
]|∆F |‡=λ/∆x≈
√2πβ
√mvN∆x|HIF |2
hλexp[−∆A‡β
]λ=2mv2N∆x2
=
√πβ
h2λ|HIF |2 exp
[−∆A‡β
]Marcus barrier≈
√πβ
h2λ|HIF |2 exp
[−β (∆A0 + λ)2
4λ
](15)
The above rate is derived for fixed number of electrons and nuclei. As done for
the adiabatic case, this fixed particle rate needs to be turned to a fixed potential rate.
In particular, the electronic subsystem needs to be open in order to study kinetics
at a fixed electrode potential. However, generalization of the NA-HTST to GCE is
significantly more difficult compared the the adiabatic as discussed. The electronically
GCE NA-HTST can be accomplished the approach in Section 4 of the main paper.
To gain more insight, it is useful to compare the above derivation to the GCE-EVB
picture used for deriving the GCE equivalent of Marcus barriers and Landau-Zener
transmission probability in Sections 3 and 4 of the main paper. These considerations
directly lead to the hybrid NA-GCE-EVB rate constant
kNA−GCE−EV B(T, V, µn) ≈
⟨√πβ
h2Λ|ΩIF |2
⟩µn
exp
[−β (∆Ω + Λ)2
4Λ
](16)
where the ΩIF is the coupling matrix element, and the prefactor is computed for
either i) some particle number and assumed to independent of the electrode potential or
ii) various particle numbers and weighted according to the grand canonical distribution.
Supporting Information 11
3. Grand canonical diabatic states using density matrices
Here the theoretical and technical details of the GCE-EVB theory are presented. To
form the GCE diabatic states, the work of Reimers[8, 9] on canonical ensembles is
followed. As noted by Reimers, the density matrix ˆρ can be written using either
adiabatic or non-adiabatic states. Especially, when only two electronic states are used,
the connection of the Born-Huang expansion bears striking resemble to the commonly
used 2 × 2 diabatic Hamiltonians used for deriving electron transfer rate theory. In
the canonical ensemble, the diabatic states are φI and φF corresponding to the electron
localized on the initial (I) or final (F) state while the molecular electronic-vibrational
Hamiltonian is
Hdia(N, V, T ) =
[HII HIF
HFI HFF
](17)
with
HII(R) = 〈φI |Hel(R)|φI〉+ Tnuc = EI + Tnuc (18a)
H∗FI = HIF = 〈φI |Htot(R)|φF 〉 (18b)
HFF = 〈φF |Hel(R)|φF 〉+ Tnuc = EF + Tnuc (18c)
where Tnuc is the nuclear kinetic energy operator, H = Hel + Tnuc, and Hel includes
electron kinetic energy and Coulomb energies of the electron-nucleus system. The Born-
Huang, or vibronic, states are
Ψi(R) =∑j
[CIij |ψI〉 |χj〉+ CF
ij |ψF 〉 |χj〉]
=∑k=I,F
|ψk〉∑j
CIki,j |χj〉
(19)
where Ψ, ψ, and χ are the vibronic, electronic, and nuclear wave functions,
respectively. C is the weight of each state. Using these definitions the, the canonical
ensemble density matrix is
ρ(N, V, T ) =
[ρII ρIFρFI ρFF
](20)
with ρAB =∑
j CAjiC
Bji and the total density matrix has dimension (2×Ni)×(2×Ni).
Next the diabatic canonical Hamiltonian is generalized to the grand canonical
ensemble. To simplify the notation, it is assumed that the initial and final can
approximated as a single electronic state and a single vibrational state - extension to
include more vibrational state is straight-forward. Then, the total vibronic state is
written as Ψ(R) ≈ cI |ψI〉 |χI〉 + cF |ψF 〉 |χF 〉. In electron transfer theory the vibronic
Supporting Information 12
states are often assumed to be harmonic but here such an assumption is not needed.
Next, the total number of electrons is allowed to fluctuate while the electron Fermi-
level is fixed. These are effectively introduced by using the equilibrium reduced density
operator within the GCE [23]
ˆρred =∑N
pN∑ij
|ΨNi〉 〈ΨNj|
with |Ψi〉 = cI |ψI〉 |χI〉+ cF |ψF 〉 |χF 〉(21)
where pN is the GCE weight for a state with N electrons. The resulting density
matrix will have N -dimensional block-diagonal form with 2 × 2 blocks. Similarly
the Hamiltonian matrix is made of Eq.(17) HNdia blocks. Diagonalizing each block
separately will give canonical adiabatic states whereas Tr[
ˆρredH]
gives the adiabatic
grand canonical free energy. Because the trace is cyclic, both ˆρred and H can be
reorganized which keeps the (diabatic) free energy unchanged as long as diagonal
elements remain at the diagonal. This freedom is utilized can be utilized for reorganizing
the density matrix so that the upper part of ˆρred and H correspond to the initial state and
the lower part to the final diabatic states. Tracing the upper and lower parts separately,
diabatic GC free energies of initial and final states (ΩII and ΩFF ) are obtained. These
states and their effective couplings are used in the main paper to derive the GCE-EVB
model.
4. Grand canonical perturbation theory
Here it is shown how the needed GCE-EVB states rigorously defined and computed
with grand canonical perturbation theory. To keep the present work as general as
possible i.e. allowing both the number of electron and nuclear species to fluctuate, a
simple effective Hamiltonian cannot be specified. Instead, explicit sampling of the GCE
and number of electrons and nuclei is needed. In this case a novel extension of the
canonical thermodynamic perturbation theory[26] to GCE is utilized. Along these lines,
the canonical energy operator H = H0 + V is defined and partitioned to contributions
from the unperturbed H0 part and a perturbation V . The total GC partition function
Ξ and grand energy Ω are given as[23]
Ξ = Tr[H − TS − µN ] and exp[−βΩ] = Ξ (22)
Then, the total grand energy can be multiplied and divided by the unperturbed
grand energy
exp[−βΩ] =exp[−β(Ω− Ω0)]
exp[−βΩ0]=
exp[−β(ΩV )]
exp[−βΩ0]= 〈exp[−βV ]〉0 (23)
Supporting Information 13
where the last identity means that the perturbation part of the grand energy is
obtained with GCE sampling of the perturbation. For electron transfer reactions, the
total Hamiltonian can be written as[27]
H = K + U + Vx (24)
where K is the kinetic energy, U is the interaction energy and Vx is the perturbation
which depends on extent of the reaction: x = 0 and x = 1 correspond to initial and
final states, respectively. A linear switch from the initial to the final state is obtained
using a switching potential Vx = VI − x(VF − VI). This potential defines the initial and
final diabatic states and based on the energies of the initial and final states EI and EF .
Furthermore, one defines the instantaneous energy gap ∆E(R) = EF (R)− EI(R) = X
at geometry R. As noted by Zusman[28] and Warshel[29] (see also Ref. [30] for a
combined discussion), the energy gap coordinate is directly related to the (solvent/bath)
reorganization coordinate and both are often used in deriving electron transfer rates. It
was recently shown by Jeanmairet et.al.[27] that the energy gap coordinate is a valid
reaction coordinate also within GCE.
Combining the two-state GCE diabatic model for the initial I and final F states
with the general perturbation result leads to
exp[−β∆Ω] =〈exp[−βVF ]〉F〈exp[−βVI ]〉I
=
∑N e
βµN∫dPNdRNe−βVF∑
N eβµN
∫dPNdRNe−βVI
=ΞVF
ΞVI
(25)
which results in ∆Ω = −β−1 ln(ΞVF /Ξ
VI
). Next, the sampling is constrained to a
specific region of the energy gap. As recently shown in Ref. [27], a one-to-one mapping
exists between the vertical energy gap 〈∆E〉x, x, the potential Vx, and the probability
(px) of being in microstate sampled from the GCE: x↔ 〈∆E〉x ↔ Vx ↔ px. Introducing
the energy gap coordinate and noting that the energies of I and F are computed from
the same Hamiltonians except for the ”perturbation” part allows writing
∆Ω = −β−1 ln
(∑N e
βµN∫dPNdRNe−β(∆E+V I)∑
N eβµN
∫dPNdRNe−βV I
)=
− β−1 ln⟨e−β∆E
⟩I
= β−1 ln⟨eβ∆E
⟩F
(26)
where ∆E = VF − VI is used. One can also obtain a probability distribution for
the energy gap by performing constrained sampling[12] of the grand energy curves
Ξi(X) =∑N
eβµN∫dPNdRNe−βEiδ(∆E(R)−X) (27a)
Supporting Information 14
pi(X) =Ξi(X(R))
Ξi= 〈δ(∆E(R)−X)〉i =∑
N eβµN
∫dRNdPNδ(∆E(R)−X)e−βEi∑
N eβµN
∫dRNdPNe−βEi
(27b)
so that Ξi =∫dXΞi(X) ≡ e−βΩi and Ωi(X) = −β−1 ln(pi(X)) + Ωi. Above Ωi is
the diabatic grand energy and i = I or F . Using the last identity and observing that
integration over the probability is unity one obtains
Ωi = −β−1 ln
∫dXe−βΩi(X) (28)
An important identity linking the diabatic grand energies to the energy gap is
obtained by using the energy gap as the reaction coordinate. In this case the constrained
grand energy along the energy gap leads to
ΩI(∆E) = −β−1 ln(ΞI(∆E)
)=
− β−1 ln
(∑N
eβµN∫dPNdRNe−βEI(RN )δ(∆E(RN)−∆E)
)
= −β−1 ln
(∑N
eβµN∫dPNdRNe−β(EF (RN )−∆E(R))δ(∆E(RN)−∆E)
)
= −β−1 ln
(eβ∆E
∑N
eβµN∫dPNdRNe−β(EF (RN ))δ(∆E(RN)−∆E)
)
= −∆E − β−1 ln
(∑N
eβµN∫dPNdRNe−βEF (RN )δ(∆E(RN)−∆E)
)= −∆E + ΩF (∆E)
(29)
At this point all relevant free energy identities within the GCE corresponding to
the commonly used identities used for deriving the canonical Marcus theory have been
obtained.[31, 32, 33, 34, 35, 29, 36] Refs. [31, 32, 33, 34, 35, 29, 36] show various ways
to obtain the iconic canonical Marcus rate constant. To arrive at the corresponding rate
constant in the GCE, it is shown that detailed balance is satisfied. At the transition
state the initial and final diabatic grand energies are equal which results in
ΩI(∆E‡) = ΩF (∆E‡)
→ −β−1 ln(pI(∆E
‡))
+ ΩI = −β−1 ln(pF (∆E‡)
)+ ΩF
→ pI(∆E‡)
pF (∆E‡)= exp[−β(ΩF − ΩI)] = exp[−β∆ΩFI ]
(30)
which shows that detailed balance is satisfied. The diabatic grand energy surfaces
are computed from the energy gap distribution[35]
Supporting Information 15
gI(∆E) = −β−1 ln(pI(∆E)) and
gF (∆E) = −β−1 ln(pF (∆E)) + ∆ΩFI
(31)
The transition state can then be identified from the intersection of the relatice grand
energy curcves: gI(∆E‡) = gF (∆E‡). Computing the reaction rate using the standard
transition state theory expression gives
kIF = κexp[−βgI(∆E‡)
]∫d∆E exp[−βgI(∆E)]
= κpI(∆E‡) (32)
showing that the reaction rate is determined by the energy gap distribution function
pI(∆E) = 〈δ(∆E(R)−∆E)〉I from Eq. (27). Note, that microscopic reversibility is
satisfied by construction. To obtain the iconic Marcus rate within GCE, one may follow
the perturbation theory route[26, 35] and perform a cumulant expansion on the energy
gap distribution. It has been shown in several previous studies[35, 37, 32] that the second
order cumulant expansion results a Gaussian form for the energy gap distribution:
pI(∆E) =1√
2πσIexp
[−(∆E − 〈∆E〉I)2
2σ2I
](33)
where 〈∆E〉I is the energy gap expectation value in the initial state obtained from
Eq. (27) and σI = 〈(∆E)2〉I − (〈∆E〉I)2 = 〈(∆EI − 〈∆EI〉)2〉I is the gap variance. The
Marcus relation is then obtain after standard manipulations[32, 25] by inserting these
relations in Eq. (29) result in the GCE Marcus rate of Eq. 12 of the main paper.
5. Cumulant expansion
The cumulant expansion is a widely used and powerful statistical approach for
approximating the reaction barrier. To utilize the cumulant form, the ensemble average
of the energy gap is replaced by its time integral using the Fourier presentation of the
delta function
pI(∆E) =ΞI(∆E)
Ξi= 〈δ(∆E(R)−∆E)〉i =
β
2π
∫ ∞−∞
⟨e−iγβ(∆E(R)−∆E)
⟩Idγ (34)
As shown by e.g. Matyushov and collaborators[38, 39], the exponential form can be
expanded using cumulants to result in non-Gaussian gap distribution functions. In this
treatment, the gap function is written in terms using a cumulant generating function.
One obtains[39] ⟨e−iγβ(∆E(R)
⟩I
= eΩI(γ) (35)
where ΩI(γ) is written in terms of cumulants as
Supporting Information 16
〈(δE(R))n〉a =
(i
β
)n∂nΩI(γ)
∂γn
∣∣∣∣γ=0
(36)
and δE(R) = E(R)−〈E(R)〉 and 〈E(R)n=1〉I = 〈E(R)〉I . For a Gaussian variable,
〈E(R)n>2〉I = 0 which results in quadratic free energy curves along the energy gap
coordinate and the identities shown in the above Section.[38, 39, 40] For more complex
environments and distributions, higher order cumulants contribute and more complex
free energy curves emerge as shown in [38, 39, 40].
6. GCE-cDFT script
from ase import *
from ase.io import write, read
from ase.units import mol, kJ, kcal, Pascal, m, Bohr
from ase.data.vdw import vdw_radii
from gpaw import *
from gpaw.analyse.hirshfeld import HirshfeldPartitioning
from gpaw.analyse.vdwradii import vdWradii
from ase.calculators.vdwcorrection import vdWTkatchenko09prl
from gpaw.cdft.cdft import *
from gpaw.poisson import PoissonSolver
# Import solvation modules
from ase.units import mol, kJ, kcal, Pascal, m
import numpy as np
from gpaw.solvation import (
SolvationGPAW, # the solvation calculator
EffectivePotentialCavity, # cavity using an effective potential
Power12Potential, # a specific effective potential
LinearDielectric, # rule to construct permittivity function from the cavity
GradientSurface, # rule to calculate the surface area from the cavity
SurfaceInteraction # rule to calculate non-electrostatic interactions
)
from gpaw.poisson import PoissonSolver
import sys, os
from gpaw.solvation.sjm import *
# define system
atoms = read('Au_H3O_to_H2O_H_4.46.traj')
######## Potential
E_target_RHE = 0.0 #PZC of the pure gold surface
E_pzc_comp = 5.02 # absolute wrt continuum, computed
E_pzc_exp = 0.559 # vs RHE, Experimental
Supporting Information 17
E_ref_comp_RHE = E_pzc_comp - E_pzc_exp # computational reference potential wrt RHE
E_target = E_ref_comp_RHE + E_target_RHE
# Solvent parameters
u0 = 0.180 # eV
epsinf = 78.36 # Dielectric constant of water at 298 K
gamma = 18.4 * 1e-3 * Pascal * m
T = 298.15 # K
vdw_radii = vdw_radii.copy()
vdw_radii[79] = 2 #Au
atomic_radii = lambda atoms: [vdw_radii[n] for n in atoms.numbers]
# Define SJM calculator
calc = SJM(h=.18,
spinpol = True,
basis = 'dzp',
xc = 'PBE',
maxiter=500,
nbands = -40,
symmetry ='off',
eigensolver = Davidson(3),
kpts = (3,3,1),
poissonsolver='dipolelayer': 'xy',
occupations = FermiDirac(width = 0.05),
mixer = Mixer(beta = 0.05,nmaxold = 5,weight=90.0),
convergence = 'eigenstates': 1.0e-5, # eV^2 / electron
'energy': 1.0e-5, # eV / electron
'density': 1.0e-3,
txt = '1Au111_H3O_to_H2O_H_reorg_CDFT_TS09_%3.2f.txt'%(E_target_RHE),
potential=E_target, #define potential
dpot=0.025,
ne=-0.3,
doublelayer='upper_limit':atoms.get_cell()[2][2]-1,'start':'cavity_like',
cavity=EffectivePotentialCavity(
effective_potential=SJMPower12Potential(atomic_radii, u0,
unsolv_backside=False,
H2O_layer=False),
temperature=T,
surface_calculator=GradientSurface()),
dielectric=LinearDielectric(epsinf=epsinf),
interactions=[SurfaceInteraction(surface_tension=gamma)])
atoms.set_calculator(calc)
atoms.get_potential_energy()
# start cDFT loop with SJM calculator
cdft_a = CDFT(calc = atoms.calc,
Supporting Information 18
atoms=atoms,
charge_regions = [[45,46,47,48]], # atoms for H3O
charges = [0.], # charge of the diabatic state
charge_coefs = [-3.75], #initial guess for Vc
method = 'L-BFGS-B',
txt = '1Au111_H3O_to_H2O_Hreorg_TS09_constant_pot_%3.2f.cdft'%(E_target),
minimizer_options='gtol':0.05) # tolerance for charge constraint
atoms.set_calculator(cdft_a)
atoms.get_potential_energy()
# add vdw contributions
# van der Waals parameters
radii = vdWradii(atoms.get_chemical_symbols(), 'PBE')
cc = vdWTkatchenko09prl(HirshfeldPartitioning(cdft_a.calc),radii)
atoms.set_calculator(cc)
atoms.get_potential_energy()
Supporting Information 19
7. References
[1] Dogonadze R and Kuznetsov A 1975 Progress in Surface Science 6 1 – 41
[2] CHIDSEY C E D 1991 Science 251 919–922
[3] Memming R 2001 Semiconductor Electrochemistry (WILEY-VCH Verlag GmbH)
[4] Dogonadze R 1971 3. theory of molecular electrode kinetics Reactions of Molecules at Electrodes
ed Hush N (Wiley-Intersciences) pp 135–228
[5] Mishra A K and Waldeck D H 2011 The Journal of Physical Chemistry C 115 20662–20673
[6] Schmickler W 1986 Journal of Electroanalytical Chemistry and Interfacial Electrochemistry 204
31 – 43 ISSN 0022-0728
[7] Schmickler W 2017 Russian Journal of Electrochemistry 53 1182–1188
[8] Reimers J R, McKemmish L K, McKenzie R H and Hush N S 2015 Phys. Chem. Chem. Phys.
17(38) 24641–24665
[9] McKemmish L K, McKenzie R H, Hush N S and Reimers J R 2015 Phys. Chem. Chem. Phys.
17(38) 24666–24682
[10] Richardson J O, Bauer R and Thoss M 2015 The Journal of Chemical Physics 143 134115
[11] Miller W H, Schwartz S D and Tromp J W 1983 The Journal of Chemical Physics 79 4889–4898
[12] Nitzan A 2006 Chemical Dynamics in Condendsed Phases: Relaxation, Transfer, and Reactions
in Condensed Molecular Systems (Oxford University Press)
[13] Marzari N, Mostofi A A, Yates J R, Souza I and Vanderbilt D 2012 Rev. Mod. Phys. 84(4) 1419–
1475
[14] Lehtola S and Jonsson H 2013 Journal of Chemical Theory and Computation 9 5365–5372
[15] Knizia G 2013 Journal of Chemical Theory and Computation 9 4834–4843
[16] Edmiston C and Ruedenberg K 1963 Rev. Mod. Phys. 35(3) 457–464
[17] Thygesen K S, Hansen L B and Jacobsen K W 2005 Phys. Rev. B 72(12) 125119
[18] Gajdos F, Valner S, Hoffmann F, Spencer J, Breuer M, Kubas A, Dupuis M and Blumberger J
2014 Journal of Chemical Theory and Computation 10 4653–4660
[19] Richardson J O and Thoss M 2014 The Journal of Chemical Physics 141 074106
[20] Oberhofer H and Blumberger J 2012 Phys. Chem. Chem. Phys. 14(40) 13846–13852
[21] Wu Q and Van Voorhis T 2006 J. Chem. Phys. 125 164105
[22] Miller W H 1998 The Journal of Physical Chemistry A 102 793–806
[23] Melander M M, Kuisma M J, Christensen T E K and Honkala K 2019 The Journal of Chemical
Physics 150 041706
[24] Mattiat J and Richardson J O 2018 The Journal of Chemical Physics 148 102311
[25] Blumberger J 2015 Chemical Reviews 115 11191–11238
[26] Zwanzig R W 1954 The Journal of Chemical Physics 22 1420–1426
[27] Jeanmairet G, Rotenberg B, Levesque M, Borgis D and Salanne M 2019 Chem. Sci. 10(7) 2130–
2143
[28] Zusman L 1980 Chemical Physics 49 295 – 304 ISSN 0301-0104
[29] Warshel A 1982 The Journal of Physical Chemistry 86 2218–2224
[30] Raineri F O and Friedman H L 2007 Solvent Control of Electron Transfer Reactions (John Wiley
& Sons, Ltd) pp 81–189 ISBN 9780470141663
[31] Tachiya M 1989 The Journal of Physical Chemistry 93 7050–7052
[32] Tateyama Y, Blumberger J, Sprik M and Tavernelli I 2005 The Journal of Chemical Physics 122
234505
[33] Vuilleumier R, Tay K A, Jeanmairet G, Borgis D and Boutin A 2012 Journal of the American
Chemical Society 134 2067–2074
[34] Rose D A and Benjamin I 1994 The Journal of Chemical Physics 100 3545–3555
[35] Zhou H X and Szabo A 1995 The Journal of Chemical Physics 103 3481–3494
[36] King G and Warshel A 1990 The Journal of Chemical Physics 93 8682–8692
[37] Sun X and Geva E 2016 The Journal of Physical Chemistry A 120 2976–2990
Supporting Information 20
[38] Matyushov D V and Voth G A 2000 The Journal of Chemical Physics 113 5413–
5424 (Preprint https://aip.scitation.org/doi/pdf/10.1063/1.1289886) URL
https://aip.scitation.org/doi/abs/10.1063/1.1289886
[39] Matyushov D V 2018 The Journal of Chemical Physics 148 154501 (Preprint
https://doi.org/10.1063/1.5022709) URL https://doi.org/10.1063/1.5022709
[40] Matyushov D V and Newton M D 2018 The Journal of Physical Chemistry B
122 12302–12311 (Preprint https://doi.org/10.1021/acs.jpcb.8b08865) URL
https://doi.org/10.1021/acs.jpcb.8b08865
download fileview on ChemRxivsupporting_info.pdf (384.50 KiB)