Supplementary Information for

A bottom-up, multi-scale theory for transient mass transport of redox-active

species through porous electrodes beyond the pseudo-steady limit

Md Abdul Hamida and Kyle C. Smitha,b,c,d,*

a Department of Mechanical Science and Engineering,
University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA
b Department of Materials Science and Engineering,
University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA
c Computational Science and Engineering Program,
University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA
d Beckman Institute for Advanced Science and Technology,
University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA

*Corresponding author: [email protected]

A. Derivation of the criteria for the validity of uniform surface-concentration

boundary conditions.

Here we derive the criteria that must be satisfied to use a uniform surface
concentration boundary condition for active soluble species undergoing a coupled
heterogenous redox (s𝑅 𝑅 𝑧𝑅 + 𝑠𝑂 𝑂 𝑧𝑂 + 𝑒 − ⇌ 0), where the electromigration flux
contributions of such species are negligible in solution. The transient local concentration
of species 𝑖 is governed by the following mass conservation equation (MCE):
𝜕𝑐𝑖
+ ∇ · (𝒋𝑖 ) = 0 (A1)
𝜕𝑡

𝑍𝐹
𝑖
Here 𝒋𝑖 = −𝐷𝑖 ∇𝑐𝑖 − 𝑅𝑇 𝐷𝑖 𝑐𝑖 ∇𝜙 + 𝒖∎ 𝑐𝑖 is the total flux of species 𝑖. Note that 𝑖 appearing
as subscript denotes either species 𝑅 or 𝑂, not to be confused with 𝑖 appearing as
denotation of current density. The flux 𝒋𝑖 is also not to be confused with the imaginary
unit 𝑗 = √−1 appearing later. We assume that solid/solution interfaces are stationary and
impenetrable to bulk solution, resulting in 𝒖∎ ∙ 𝒏|𝑠 = 0. The coupled production and
consumption of species is enforced by the boundary condition (BC) as 𝒋𝑅 ∙ 𝒏 ̂ |𝑠 = −𝒋𝑂 ∙ 𝒏
̂ |𝑠
𝜕𝑐𝑅 𝜕𝑐𝑂
or 𝐷𝑅 | = −𝐷𝑂 | , since electromigration is assumed to be negligible and because
𝜕𝑛 𝑠 𝜕𝑛 𝑠
electrode surfaces are impermeable to solution. The current density 𝑖 at the solid/solution
interface is coupled to the flux of either species as a result of redox reaction stoichiometry:
𝐹 𝐹 𝜕𝑐𝑖
𝑖 = 𝑠 〈𝑗𝑖 〉𝑠 = − 𝑠 𝐷𝑖 | = 𝑓(𝜂) (A2)
𝑖 𝑖 𝜕𝑛 𝑠

Here, overpotential is expressed as 𝜂 = 𝜙𝑠 (𝑡) − 𝜙𝑒 (𝑡) − 𝜙𝑒𝑞 (𝑐𝑂,𝑠 , 𝑐𝑅,𝑠 ), and we assume
that the oxidation reaction produces a positive current, while reduction produces a
negative current. Redox reaction kinetics relate 𝑖 to overpotential 𝜂. Using Taylor series
expansion with respect to a certain current density 𝑖 ∗ that corresponds to a certain

1
 ∗ ∗
overpotential 𝜂 ∗ = 𝜙𝑠 (𝑡) − 𝜙𝑒 (𝑡) − 𝜙𝑒𝑞 (𝑐𝑂,𝑠 , 𝑐𝑅,𝑠 ), we obtain a linear expansion of current
density that will later be used to derive validity conditions on the basis of scaling analysis:

𝜕𝒊 𝜕𝒊 ∗ 𝜕𝒊
∗ 𝜕𝒊
𝑖 ≈ 𝑖 ∗ + 𝜕𝜙 | (𝜙𝑠 − 𝜙𝑠∗ ) + 𝜕𝜙 | (𝜙𝑒 − 𝜙𝑒∗ ) + 𝜕𝑐 | (𝑐𝑅,𝑠 − 𝑐𝑅,𝑠 ) + 𝜕𝑐 | (𝑐𝑂,𝑠 − 𝑐𝑂,𝑠 )
𝑠 ∗ 𝑒 ∗ 𝑅,𝑠 ∗ 𝑂,𝑠 ∗

𝜕𝒊 ∗ 𝜕𝒊 ∗ 𝜕𝒊
or, 𝑖 ≈ 𝑖 ∗ + 𝜕𝜂0 | (∆𝜙|𝑠 − ∆𝜙|∗𝑠 ) + 𝜕𝑐 | (𝑐𝑅,𝑠 − 𝑐𝑅,𝑠 ) + 𝜕𝑐 | (𝑐𝑂,𝑠 − 𝑐𝑂,𝑠 ) (A3)
∗ 𝑅,𝑠 ∗ 𝑂,𝑠 ∗

Here, ∆𝜙|𝑠 = 𝜙𝑠 − 𝜙𝑒 is a driving potential difference between the solid and the
electrolyte, and 𝜂0 = 𝜙𝑠 − 𝜙𝑒 − 𝜙𝑒𝑞
0
is the overpotential at standard conditions. Using the
law of mass action with rate constants for oxidation 𝑘𝑜𝑥 and reduction 𝑘𝑟𝑒𝑑 we find:

𝑖 = 𝑓(𝜂) = 𝐹(𝑘𝑓 𝑐𝑅,𝑠 − 𝑘𝑏 𝑐𝑂,𝑠 ) = 𝐹(𝑘𝑜𝑥 𝑐𝑅,𝑠 − 𝑘𝑟𝑒𝑑 𝑐𝑂,𝑠 ) (A4)

Using this kinetics formalism, the relevant partial derivatives of current density 𝑖 are:
𝜕𝑖 𝜕𝑘𝑜𝑥 𝜕𝑘𝑟𝑒𝑑
= 𝐹 (𝑐𝑅,𝑠 − 𝑐𝑂,𝑠 ) (A5a)
𝜕𝜂 0 𝜕𝜂 0 𝜕𝜂 0

𝜕𝑖 𝜕𝑘𝑜𝑥 𝜕𝜂 0 𝜕𝑘𝑟𝑒𝑑 𝜕𝜂 0
= 𝐹 (𝑐𝑅,𝑠 − 𝑐𝑂,𝑠 − 𝑘𝑟𝑒𝑑 ) (A5b)
𝜕𝑐𝑂,𝑠 𝜕𝜂 0 𝜕𝑐𝑂,𝑠 𝜕𝜂 0 𝜕𝑐𝑂 ,𝑠

𝜕𝑖 𝜕𝑘𝑜𝑥 𝜕𝜂 0 𝜕𝑘𝑟𝑒𝑑 𝜕𝜂 0
= 𝐹 (𝑘𝑜𝑥 + 𝑐𝑅,𝑠 − 𝑐𝑂,𝑠 ) (A5c)
𝜕𝑐𝑅,𝑠 𝜕𝜂 0 𝜕𝑐𝑅,𝑠 𝜕𝜂 0 𝜕𝑐𝑅,𝑠

Using Butler-Volmer kinetics with 𝑘𝑜𝑥 = 𝑘 0 𝑒𝑥𝑝[(1 − 𝛼)𝐹 𝜂0 /𝑅𝑇] and 𝑘𝑟𝑒𝑑 =
𝑘 𝑒𝑥𝑝[−𝛼𝐹 𝜂0 /𝑅𝑇], these derivatives are 𝜕𝑖⁄𝜕𝜂0 = 𝐹 2 ⁄𝑅𝑇 [(1 − 𝛼)𝑘𝑜𝑥 𝑐𝑅,𝑠 + 𝛼 𝑘𝑟𝑒𝑑 𝑐𝑂,𝑠 ],
0

𝜕𝑖⁄𝜕𝑐𝑅,𝑠 = 𝐹𝑘𝑜𝑥 and 𝜕𝑖⁄𝜕𝑐𝑂,𝑠 = −𝐹𝑘𝑟𝑒𝑑 . Using Eqs. A2 and A3, we obtain a linear, Robin-
type BC that couples the concentration of both species to the gradient of their
concentrations at the solid/solution interface (Eq. A6).

After rearrangement Eq. A6 reduces to the following expression:

𝐷𝑖 𝜕𝑐𝑖 ∗ ∗ ∗ ∗ 𝑖∗ ∗ ∗ ∗ 𝐹∗
| + 𝑘𝑜𝑥 𝑐𝑅,𝑠 − 𝑘𝑟𝑒𝑑 𝑐𝑂,𝑠 = − 𝐹 − 𝑘𝑜𝑥 𝑐𝑅,𝑠 [𝑅𝑇 (∆𝜙|𝑠 − ∆𝜙|∗𝑠 )] + (𝑘𝑜𝑥 𝑐𝑅,𝑠 − 𝑘𝑟𝑒𝑑 𝑐𝑂,𝑠 ) [1 +
𝑠𝑖 𝜕𝑛 𝑠
𝐹
𝛼 𝑅𝑇 (∆𝜙|𝑠 − ∆𝜙|∗𝑠 )] (A6)

Nondimensionalization of the terms in Eq. A6 enables us to perform scaling analysis. To

do so we choose a nondimensional normal coordinate 𝑛̃ = 𝑛⁄𝑏 with characteristic pore
dimension 𝑏, a nondimensional surface concentration 𝑐̃𝑖 |𝑠 = (𝑐𝑖 ⁄𝑐𝑖∗ )|𝑠 , and a
nondimensional potential difference Δ𝜙̃ = Δ𝜙𝐹 ⁄(𝑅𝑇):
∗
1 𝜕𝑐̃ ∗ 𝑏
𝑘𝑜𝑥 𝑐∗ ∗
𝑘𝑟𝑒𝑑 𝑏 𝑐𝑂,𝑠 𝑏 𝑖∗ ∗ 𝑏 𝑐∗
𝑘𝑜𝑥
( 𝜕𝑛̃𝑖 | ) + [ 𝑐𝑅,𝑠
∗ ] (𝑐̃𝑅,𝑠 ) − [ 𝑐 ∗ ] (𝑐̃𝑂,𝑠 ) = − 𝐹𝐷 𝑐 ∗ + 𝑅,𝑠
[ 𝑐 ∗ ] [1 − (1 − 𝛼) (∆𝜙̃|𝑠 −
𝑠𝑖 𝑠 𝐷𝑖 𝑖,𝑠 𝐷𝑖 𝑖,𝑠 𝑖 𝑖,𝑠 𝐷𝑖 𝑖,𝑠
∗ ∗
∗ 𝑘𝑟𝑒𝑑 𝑏 𝑐𝑂,𝑠 ∗
𝜙̃|𝑠 )] − [ 𝑐 ∗ ] [1 + 𝛼 (∆𝜙̃|𝑠 − ∆𝜙̃|𝑠 )] (A7)
𝐷𝑖 𝑖,𝑠

2
 To approach a Dirichlet condition, thus representing a boundary condition that
depends solely on the concentration of either species but not on its gradient, requires that
the gradient term in Eq. A7 (i.e., 𝜕𝑐̃𝑖 /𝜕𝑛̃|𝑠 ) vanish relative to the concentration-dependent
term. We now consider characteristic scales for the concentration of each species
∗ ∗
(𝑐𝑂,𝑠 ~𝑐𝑂,𝑠 and 𝑐𝑅,𝑠 ~𝑐𝑅,𝑠 ) and for length (𝑛~𝑏), which require that (𝜕𝑐̃𝑖 ⁄𝜕𝑛̃)|𝑠 ~1 , 𝑐̃𝑅,𝑠 ~1,
and 𝑐̃𝑂,𝑠 ~1). On this basis we argue that the gradient term in Eq. A7 vanishes relative to
the concentration-dependent terms when either of the two criteria are satisfied:
∗ ∗ ∗ 𝑏 𝑐∗
𝑘𝑟𝑒𝑑 𝑏 𝑐𝑂,𝑠 𝑘𝑜𝑥 𝑅,𝑠
1≪| ∗ | or 1≪| ∗ | (A8)
𝐷𝑖 𝑐𝑖,𝑠 𝐷𝑖 𝑐𝑖,𝑠

For oxidized and reduced species these conditions can be written as,
∗
𝑘𝑟𝑒𝑑 𝑏 ∗
𝑘𝑟𝑒𝑑 𝑏 ∗ 𝑐∗
𝑘𝑜𝑥 𝑅,𝑠
For 𝑂: 1 ≪ | |; or 1≪| ( ∗ ∗ )| (A9a)
𝐷𝑂 𝐷𝑂 𝑘𝑟𝑒𝑑 𝑐𝑂,𝑠

∗ 𝑏 ∗ ∗ ∗ 𝑏
𝑘𝑜𝑥 𝑘𝑟𝑒𝑑 𝑐𝑂,𝑠 𝑘𝑜𝑥
For 𝑅: 1 ≪ | ( ∗ ∗ )|; or 1≪| | (A9b)
𝐷𝑅 𝑘𝑜𝑥 𝑐𝑅,𝑠 𝐷𝑅

The terms inside the parentheses in Eq. A9 can be simplified by utilizing the de Donder
∗ ⁄ ∗ ∗ ⁄ ∗
relation (𝑘𝑜𝑥 𝑘𝑟𝑒𝑑 = 𝑒𝑥𝑝(𝐹𝜂0 ⁄𝑅𝑇)) and a logarithm identity (𝑐𝑂,𝑠 𝑐𝑅,𝑠 =
∗ ⁄ ∗ ∗ ∗ ∗ ∗
𝑒𝑥𝑝[𝑙𝑛(𝑐𝑂,𝑠 𝑐𝑅,𝑠 )]), by which we obtain (𝑘𝑜𝑥 𝑐𝑅,𝑠 )⁄(𝑘𝑟𝑒𝑑 𝑐𝑂,𝑠 ) = 𝑒𝑥𝑝[𝐹 ⁄𝑅𝑇 (𝜂0 −
𝑙𝑛(𝑐𝑂∗ ⁄𝑐𝑅∗ )|𝑠 )] = 𝑒𝑥𝑝(𝐹𝜂∗ ⁄𝑅𝑇). Introducing the definition of a Damköhler number for both
reduction and oxidation as 𝐷𝑎𝑟𝑒𝑑 = 𝑘𝑟𝑒𝑑 𝑏⁄𝐷𝑜 and 𝐷𝑎𝑜𝑥 = 𝑘𝑜𝑥 𝑏⁄𝐷𝑅 , the required criteria
to approach a uniform surface condition are given as:

𝐷𝑎𝑟𝑒𝑑 ≫ min[1, 𝑒𝑥𝑝(−𝐹𝜂 ∗ ⁄𝑅𝑇)] and 𝐷𝑎𝑜𝑥 ≫ min[1, 𝑒𝑥𝑝(𝐹𝜂∗ ⁄𝑅𝑇)] (A10)

Thus, the respective Damköhler numbers must be sufficiently large to assure that the
uniform surface concentration boundary is valid.

B. Validity criterion for periodic boundary conditions

Here we revisit the validity criteria for imposing periodic boundary conditions for
periodic microstructures – an assumption invoked ad hoc in section 2.1 – by performing
scaling analysis of electrolyte flow through an electrode, similar to our approach used in
the pseudo-steady limit (Ref. [1]) with the exceptions that (1) we consider a non-
recirculating cell configuration and (2) we perform our scaling analysis in the frequency
domain with the aid of the transfer functions introduced already. Following our previous
work, we analyze the time-dependent average concentration 〈𝑐〉𝑟 within the
electrode/reactor using an integral MCE assuming an electrolyte feed with certain uniform
concentration of redox-active species flowing in 𝑐 𝑖𝑛 with a volumetric flow rate 𝑉̇ :
𝑑(𝜀𝐴𝑐 𝐿𝑟 〈𝑐〉𝑟 ) 𝑎𝐿 𝐴 𝐷
̃ 〉(𝑐̅𝑠 − 〈𝑐̅〉𝑟 )} (B1)
= 𝑉̇ 𝑐 𝑖𝑛 − 𝐴𝑐 |𝒖𝑠 |[〈𝑐〉𝑟 − ℱ −1 {〈𝑁∥ 〉(𝑐̅𝑠 − 〈𝑐̅〉𝑟 )}] + 𝑟𝑏 𝑐 ℱ −1 {〈𝑆ℎ
𝑑𝑡

Here, the rightmost two terms, which contain the transfer functions 〈𝑁∥ 〉 and 〈𝑆ℎ ̃ 〉, are
respectively a result of the rate of solute advection through the outlet and the rate of solute

3
production by virtue of electrochemical reactions at the solid/solution interface. 𝐴𝑐 and 𝐿𝑟
are respectively the electrode’s cross sectional area normal to the superficial velocity 𝒖𝑠
and the electrode’s length in the streamwise direction. 𝑎 is the electrode’s surface area
per unit volume. Fourier transformation of this equation simplifies this expression to:
𝑎𝐿 𝐴 𝐷
̃ 〉(𝑐̅𝑠 − 〈𝑐̅〉𝑟 )
𝑗𝜔𝜀𝐴𝑐 𝐿𝑟 〈𝑐̅〉𝑟 = 𝑉̇ 𝑐̅𝑖𝑛 − 𝐴𝑐 |𝒖𝑠 |[〈𝑐̅〉𝑟 − 〈𝑁∥ 〉(𝑐̅𝑠 − 〈𝑐̅〉𝑟 )] + 𝑟𝑏 𝑐 〈𝑆ℎ (B2)

Similar to our previous work [1], we require the ratio of characteristic streamwise
(Δ𝑐)𝑚
ǁ and transverse (Δ𝑐)𝑚 ⊥ concentration differences to be sufficiently small at the
macroscopic scale in order to assure that streamwise concentration differences are
sufficiently small, ultimately requiring that the following nondimensional quotient Δ𝑐̂ satisfy
Δ𝑐̂ ≪ 1:

𝑏 〈𝑐̅〉 −𝑐̅𝑖𝑛 (Δ𝑐)𝑚

Δ𝑐̂ = 𝐿 ∙ | 〈𝑐̅〉𝑟 | ~ |(Δ𝑐)𝑚𝑖,ǁ | (B3)
𝑟 𝑟 −𝑐̅𝑠 𝑖,⊥

By solving the Fourier-transformed MCE for 〈𝑐̅〉𝑟 − 𝑐̅𝑖𝑛 we obtain the following criterion
after significant simplification to assure the validity of periodic boundary conditions:
1 2𝜋𝑗𝑓 ∗ 𝜀〈ℂ〉
Δ𝑐̂ = 𝑃𝑒 | ̃ 〉 − 𝑏 𝑃𝑒〈𝑁∥ 〉| ≪ 1
− 𝑎𝑏〈𝑆ℎ (B4)
1−〈ℂ〉 𝐿 𝑟

Hence, the validity of using periodic boundary conditions is, in a sense, a recursive
criterion because it depends not only on input nondimensional parameters (𝑃𝑒, 𝑓 ∗ , 𝑎𝑏,
̃ 〉 that are outputs
and 𝑏/𝐿𝑟 ) but also on the nondimensional transfer functions 〈𝑁∥ 〉 and 〈𝑆ℎ
produced based on those input parameters. In practice, 〈𝑁∥ 〉 and 〈𝑆ℎ ̃ 〉 would first be
predicted and Δ𝑐̂ would subsequently be calculated to verify the self-consistency of such
simulations performed using periodic boundary conditions. Further, Eq. B4 is a criterion
that is frequency- or mode-specific, and, hence, it is advisable to evaluate Δ𝑐̂ at
frequencies that are representative of the total response.

C. ̃ 〉 at vanishing frequency with

Equivalence of spectral Sherwood number 〈𝑺𝒉
Sherwood number 〈𝑺𝒉〉 in the pseudo-steady limit

Here we demonstrate the equivalence between the spectral Sherwood number

̃ 〉 in the limit of 𝜔 → 0 with the conventional Sherwood number obtained in the pseudo-
〈𝑆ℎ
steady limit. For the pseudo-steady limit, we consider the auxiliary problem that we
introduced previously in Ref. [1]. The auxiliary problem is described by a steady MCE in
terms of the time-independent contribution 𝑐ℎ (𝒓) to the time- and space- dependent
concentration field 𝑐(𝑡, 𝒓) as:

∇ · (𝒖∎ 𝑐ℎ − 𝐷∇𝑐ℎ ) = −𝜎; 𝑐ℎ |𝑠 = 0 (C1)

where 𝑐ℎ (𝒓) = 𝑐(𝒓, 𝑡) − 𝑐𝑠 (𝑡) and the source term 𝜎 represents the temporal rate of
change in surface concentration 𝜎 = 𝜕𝑐𝑠 ⁄𝜕𝑡. The conventional Sherwood number is
defined in terms of 𝑐ℎ (𝒓) using the film law of mass transport. In the present study, the
unsteady conjugate problem is replaced by a Fourier-transformed MCE (Eq. 4) where the

4
spectral Sherwood number 〈𝑆ℎ ̃ 〉 is defined by Eq. 6 from the main text. To demonstrate
the equivalence of the Fourier transformed MCE (Eq. 4) with the auxiliary problem’s MCE
(Eq. C1), we define 𝛿 = (𝑐̅ − 𝑐̅𝑠 )⁄𝑐̅𝑠 , and we substitute it back into Eq. 4 from the main
text to produce the following:
1
∇ · (𝒖∎ 𝛿 − 𝐷∇𝛿) = − (1 + 𝛿); 𝛿|𝑠 = 0 (C2)
𝑗𝜔

Scaling analysis of Eq. C2 with a characteristic length scale 𝑏 and a characteristic velocity
𝑈 reveals that its righthand side approaches unity for vanishing frequency (i.e.,
lim (1 + 𝛿) = 1), based either on characteristic scales for advection (𝛿~ 𝜔𝑏⁄𝑈) or for
𝜔→0
diffusion ( 𝛿~ 𝜔𝑏 2 ⁄𝐷𝑖 ). In this limit Eq. C2 approaches the following form:

∇ · (𝒖∎ 𝛿 − 𝐷∇𝛿) = − 𝑗𝜔; 𝛿|𝑠 = 0 (C3)

Since 𝛿 is a complex number, we expand it in terms of real and imaginary parts 𝛿 = 𝛿𝑅𝑒 +
𝑗 𝛿𝐼𝑚 to obtain a single PDE for the real part (∇ · (𝒖∎ 𝛿𝑅𝑒 − 𝐷∇𝛿𝑅𝑒 ) = 0 with 𝛿𝑅𝑒 |𝑠 = 0) and
a single PDE for the imaginary part (∇ · (𝒖∎ 𝛿𝐼𝑚 − 𝐷∇𝛿𝐼𝑚 ) = − 𝜔 with 𝛿𝐼𝑚 |𝑠 = 0). The only
solution possible for the real component is the trivial solution, i.e., 𝛿𝑅𝑒 = 0. Thus, we
deduce that 𝛿 = 𝑗𝛿𝐼𝑚 , and we solve only for the imaginary component:

∇ · (𝒖∎ 𝛿𝐼𝑚 − 𝐷∇𝛿𝐼𝑚 ) = − 𝜔; 𝛿𝐼𝑚 |𝑠 = 0 (C4)

′
To verify the equivalence of Eq. C1 and Eq. C4, we define 𝛿𝐼𝑚 = 𝛿𝐼𝑚 /𝜔 and 𝑐ℎ′ = 𝑐ℎ /𝜎
and substitute into the respective equations to obtain:
′ ′ ) ′ |
∇ · (𝒖∎ 𝛿𝐼𝑚 − 𝐷∇𝛿𝐼𝑚 = − 1; 𝛿𝐼𝑚 𝑠 =0

∇ · (𝒖∎ 𝑐ℎ′ − 𝐷∇𝑐ℎ′ ) = −1; 𝑐ℎ′ |𝑠 = 0

Inspection reveals that solutions to these PDEs are identical, such that 𝛿𝐼𝑚 ′
= 𝑐ℎ′ and
𝛿𝐼𝑚 𝑐
= 𝜎ℎ are satisfied exactly. By recognizing that surface fluxes are only due to diffusion
𝜔
(and not advection), the associated flux and concentration difference can be expressed
in terms of 𝛿 as 〈𝑗̅〉𝑠 = −𝐷〈𝜕𝑐̅⁄𝜕𝑛〉 = −𝐷〈𝜕𝛿𝐼𝑚 ⁄𝜕𝑛〉𝑐̅𝑠 and 𝑐̅𝑠 − 〈𝑐̅〉 = −〈𝛿𝐼𝑚 〉 𝑐̅𝑠 . Therefore,
̃ 〉 approaches the following expression in the limit of vanishing frequency:
〈𝑆ℎ
𝑑 〈𝑗̅〉 𝑑 𝜕𝛿𝐼𝑚 𝑑 𝜔 𝜕𝑐ℎ 𝑑 𝜕𝑐 𝑑 〈𝑗𝑠 〉
̃ 〉 = lim
lim 〈𝑆ℎ 𝑠
= 〈𝛿 〈 〉=𝜔 〈 〉= 〈 〉= = 〈𝑆ℎ〉𝑃𝑆𝐿 (C5)
𝜔→0 𝜔→0𝐷 𝑐̅ −〈𝑐̅〉
𝑠 𝐼𝑚 〉 𝜕𝑛 〈𝑐ℎ 〉 𝜎 𝜕𝑛 −(𝑐𝑠 −〈𝑐〉) 𝜕𝑛 𝐷𝑐 𝑠 −〈𝑐〉
𝜎

This analysis demonstrates that the value of 〈𝑆ℎ ̃ 〉 for 𝜔 → 0 must be equal to the value of
〈𝑆ℎ〉 obtained in the pseudo-steady limit. This analysis also suggests an alternative
̃ 〉, i.e., that it is essentially a frequency-dependent
interpretation of the transfer function 〈𝑆ℎ
version of the Sherwood number. This statement is further supported through comparison
between lim 〈𝑆ℎ̃ 〉 values obtained from present model and 〈𝑆ℎ〉𝑃𝑆𝐿 values obtained from
𝜔→0
pseudo-steady simulations [1], where the percentage deviation observed is found to be
within machine precession (~10−15 %).

5
D. Experimentation and model prediction comparison

To compare the predictions of our bottom-up transient model with experimentally

obtained polarization values, a redox flow-cell setup was implemented and tested
experimentally. To validate the present model with experimental data the flow cell was
prepared using a well-prescribed electrode geometry using redox chemistry with well-
known electrochemical properties. The reaction chemistry chosen was the ferro/ferri-
3−
cyanide (𝐹𝑒(𝐶𝑁)4−6 /𝐹𝑒(𝐶𝑁)6 ) couple dissolved in an aqueous solution along with 𝐾𝐶𝑙
as supporting salt.

The flow cell consisted of two 200 𝜇𝑚 thick graphite foil electrodes
(4.5 𝑐𝑚 × 4.5 𝑐𝑚) separated by a non-selective polymer separator (NP030 from
STERLITECH Corporation). The electrolyte solution was pumped at a rate of 5 𝑚𝐿/𝑚𝑖𝑛
through the triangular channel engraved electrodes (TCEE) to create flow along the
length of such channels, as shown in Figs. D2a and D2c. The parallel triangular channels
(see Fig. D1a) were engraved using a Trotec laser engraver (Speedy Flexx 400 fiber
laser) with 100 𝜇𝑚 spacing from valley to valley. A nominal 90 𝜇𝑚 deep and 75 𝜇𝑚 wide
isosceles channel cross-section (Fig. D1d) was determined from the mean values of the
Gaussian distribution of measured depth and width for model implementation, as
determined from 3D profilometry (Keyence). The nominal channel cross-section area
obtained was 3375 𝜇𝑚2 , which coincides with the mean values obtained from Gaussian
distribution of measured cross-sectional area for both electrodes (Figs. D1b and D1c).
The volume-specific reactive area and total flow area for the electrolyte were calculated
from this representative geometry. We note that the channel depth 𝑑𝑐ℎ = 90 𝜇𝑚 was
taken as the characteristic length to calculate non-dimensional parameters.

Fig. D1: (a) Schematic of a triangular channel engraved electrode (TCEE) produced by
laser engraving of graphite foil. (b, c) Histograms showing Gaussian distribution of the
measured cross-sectional area at different locations. Representative channel cross-
section approximated for pore-scale model implementation that produces a cross-
sectional area close to the average values of the distributions observed in Figs. D1b and
D1c.

The electrolyte solution was prepared by dissolving potassium

hexacyanoferrate(II) trihydrate, 𝐾4 𝐹𝑒(𝐶𝑁)6 ∙ 3𝐻2 𝑂, and potassium hexacyanoferrate(III),
𝐾3 𝐹𝑒(𝐶𝑁)6, in deionized water. All the chemicals were from Sigma-Aldrich and were used

6
as received to prepare 350 𝑚𝐿 of 15 𝑚𝑀 (individual species) solution. High salt
concentration (3 𝑀 𝐾𝐶𝑙) was used to increase the ionic conductivity of the electrolyte as
well as to decrease the electromigration of active species. For simplicity, the anodic and
cathodic kinetic processes were modeled as symmetric with equal Butler-Volmer transfer
coefficients (𝛼 = 0.5) with an identical diffusivity of 𝐷 = 7 × 10−6 𝑐𝑚2 /𝑠 and a rate
constant of 𝑘𝑠 = 0.1 𝑐𝑚/𝑠. The values of 𝐷 and 𝑘𝑠 were taken from published literature
[2–7]. From these values the calculated limiting current density 𝑖𝑙𝑖𝑚 , Peclet number 𝑃𝑒,
Damköhler number 𝐷𝑎, and nondimensional flow rate 𝛽 are listed in Table D1. To obtain
the transient response of the flow cell, applied current was varied as a square wave as
shown in Fig. D2b. During all experiments, each tank was prepared with 175 𝑚𝐿 of
electrolyte to ensure that the change in state-of-charge during any charging/discharging
step was less than 0.1%.

Fig. D2: (a) Experimental setup of flow cell, showing electrolyte flow circuits and
electrode/separator/electrode staking inside of the reactor. (b) Schematic of applied
current density variation with time for two steps of galvanostatic charging/discharging. (c)
Magnified 3D representation of triangular channel engraved electrodes (TCEEs) with flow
along channel length. (d) 2D pore scale model of TCEEs with different boundary
𝜕ℂ
conditions, reactive boundaries (ℂ|𝑠 = 1) and no flux boundary (𝜕𝑛| = 0). (e) 1D macro-
𝑠
scale model of the electrode with electrolyte flow direction. Frequency dependence and
fitting obtained from rational function approximation of different TFs calculated from
numerical solution of pore-scale model, (f) spectral Sherwood number 〈𝑆ℎ ̃ 〉, (g) parallel
̃
component of suppression vector 〈𝑁∥ 〉 and (h) 〈𝑍̃〉 = 1⁄〈𝑆ℎ〉.

7
Table D1: Values of different parameters associated with flow cell experiments using
triangular channel engraved electrodes (TCEE).
Parameter (unit) Value/range
Volume-specific active area, 𝑎 (𝑚2 /𝑚3 ) 5.778 × 104
Limiting current density, 𝑖𝑙𝑖𝑚 (𝑚𝐴/𝑐𝑚2 ) 10.1
Exchange current density, 𝑖0 (𝑚𝐴/𝑐𝑚2 ) 144.7
Mean residence time, 𝑡𝑟 (𝑠𝑒𝑐) 0.82
Characteristic diffusion time, 𝜏𝑑 (𝑠𝑒𝑐) 11.6
Damköhler number, 𝐷𝑎 (−) 128.5
Péclet number, 𝑃𝑒 (−) 7.05 × 103
Non-dimensional flow rate, 𝛽 (−) 2.4 ~ 0.3
Tank-to-reactor volume ratio 2.56 × 103

Fig. D3: Variation of 𝐼𝑅-corrected cell voltage with time due to current applied with
different 𝑖𝑟𝑝 and 𝜏 values.

For the model implementation, the local pore velocity (x-component 𝑢𝑥 ) field was
obtained from numerical solution of the steady form of the x-component of the Navier-
Stokes equations in the creeping flow limit:
∇𝑝
∇ ∙ (∇𝑢𝑥∎ ) = (D1)
𝜇

𝜕𝑢∎
with no-slip (𝑢𝑥∎ |𝑠 = 0) and impermeable ( 𝜕𝑛𝑥 | = 0) boundary condition at solid-solution
𝑠
interfaces, including the separator/solution interface. The local ℂ field was obtained from

8
numerical solution of Eq. 4 from the main text by using a constant surface concentration
𝜕ℂ
condition at reactive interfaces (ℂ|s = 1), along with a no-flux condition (𝜕𝑛| = 0) at the
s
separator/solution interface (see Fig. D2d). From the numerical solution of these
equations, the spectral Sherwood number 〈𝑆ℎ ̃ 〉 and the parallel component of the
suppression vector 〈𝑁∥ 〉 were calculated using Eq. 6 and Eq. 9. The variation of these TFs
with non-dimensional frequency and their fitting from rational function approximation are
shown in Fig. D2f-h. We note that, unlike arbitrary porous media, these TFs are
independent of 𝑃𝑒 because the velocity components along any direction perpendicular to
channel length is zero (i.e., 𝑢𝑦 = 𝑢𝑧 = 0). Finally, these TFs were embedded into the
macroscopic model following an approach identical to that discussed in Sec. G of the
supplementary information to obtain the transient response of the flow cell (see Figs. D2c-
e).

To study the transient response of this system, current was applied to the flow cell
as a square wave with certain current magnitude and half-period 𝜏 (Fig. D2b). Figure D3
shows the time variation of 𝐼𝑅-corrected voltage for different 𝑖𝑟𝑝 and 𝜏 values obtained
from experiment and from predictions of the bottom-up transient model (B-UTM) along
with conventional model (CM) predictions. A detailed discussion on Fig. D3 is presented
in Sec. 3 of the main text.

E. Discretization of pore-scale MCE, numerical implementation and verification

We model flow through a 2D microstructure’s pores as a Newtonian, isochoric

solution with constant dynamic viscosity and partial molar volume for each dissolved
species. The molar-volume averaged velocity field 𝒖∎ is governed by creeping flow,
assuming that the pore-scale Reynolds number 𝑅𝑒 = 𝑃𝑒/𝑆𝑐 is sufficiently small. These
assumptions enable us to simulate 𝒖∎ using the time-independent, inertia-free version of
the vorticity transport equation (VTE) ∇4 𝜓 = 0 [8], which is obtained by taking the curl of
the Navier-Stokes equations. Here, 𝜓 is a stream function from which the components
of velocity automatically enforce conservation of volumetric flow rates (∇ ∙ 𝒖∎ = 0): 𝑢𝑥∎ =
𝜕𝜓⁄𝜕𝑦 and 𝑢𝑦∎ = − 𝜕𝜓⁄𝜕𝑥 . The associated BCs at cylinder surfaces are enforced by the
no-slip (𝜕𝜓⁄𝜕𝑛|𝑠 = 0) and impermeability (𝜕𝜓⁄𝜕𝑠|𝑠 = 0) conditions. We also require that
no lift force is generated perpendicular to 𝒖𝑠 , such that the Kutta-Joukowski [9] theorem
requires fluid circulation 𝛤 to vanish around each isolated solid in the domain: 𝛤 = ∮𝑐 𝒖∎ ∙
𝑑𝒔 = − ∮𝑐 ∇2 𝜓 𝑑𝐴𝑠 = 0. Periodic jump/fall BCs are also applied at unit-cell boundaries,
𝜓(𝒓 + 𝑹) = 𝜓(𝒓) + (𝑹/|𝑹|) · Δ𝝍. Here Δ𝝍 = Δ𝜓𝑥 𝑖̂ + Δ𝜓𝑦 𝑗̂ is the periodic jump/fall in 𝜓
across the unit-cell that results in a certain volumetric flow rate per unit depth equal to
|Δ𝝍|. We used a second-order accurate finite difference method to discretize the VTE
subject to the BCs already described, and we solved for 𝜓 using a direct solver.

The equations that govern the transfer function ℂ(𝜔, 𝒓) (Eq. 4) were discretized
using the finite volume method [10] using central differencing of advection and diffusion
flux contributions at finite-volume cell faces subject to periodic BCs at unit-cell boundaries
(i.e., ℂ(𝒓 + 𝑹) = ℂ(𝒓)) and a unity Dirichlet condition at solid/solution interfaces. A

9
complex-number, direct solver was used in MATLAB to obtain ℂ(𝜔, 𝒓) at the locations of
finite-volume cell centroids. The finite-volume cell dimensions Δ𝑥 and Δ𝑦 were selected
to ensure that the corresponding grid-level Péclet numbers (𝑃𝑒𝑥 = 𝑢𝑥∎ Δ𝑥/𝐷 and 𝑃𝑒𝑦 =
𝑢𝑦∎ Δ𝑦/𝐷) remained less than 0.5 to guarantee numerical stability subjected to the
Courant–Friedrichs–Lewy condition [11].

Fig. E1. (a) Domain of the parallel reactive boundaries used for verification of the present
pore-scale numerical simulations. (b) and (c) respectively show the variation of analytical
̃ 〉 and 〈𝑁∥ 〉 with nondimensional angular frequency 𝜔∗ for the
and numerical values for 〈𝑆ℎ
parallel reactive boundaries.

Here we present verification of our numerical scheme by comparing the obtained

results with analytically derived results for a simple geometry, namely parallel reactive
boundaries (PRBs). Fig. E1a shows a unit cell with a single boundary in the middle that
repeats itself transversely to simulate an infinite array of pairs of parallel boundaries. The
surface concentration is assumed to be uniform and to vary with time periodically
following a sine function with angular frequency 𝜔 = 2𝜋𝑓. For such a simple geometry
analytical expression for the volume-averaged concentration transfer function 〈ℂ〉 =
〈𝑐̅〉/𝑐̅𝑠 and the spectral Sherwood number ⟨𝑆ℎ ̃ ⟩ = 𝑏〈𝑗̅〉𝑠 /[𝐷(𝑐̅𝑠 − 〈𝑐̅〉)] were obtained as:

1
〈ℂ〉 = tanh(√𝑗𝜔 ∗ ) (E1)
√𝑗𝜔 ∗

√𝑗𝜔 ∗ tanh(√𝑗𝜔 ∗ )
and, ̃⟩ =
⟨𝑆ℎ 1 (E2)
1− tanh(√𝑗𝜔 ∗ )
√𝑗𝜔∗

Here, frequency is nondimensionalized as 𝜔∗ = 𝜔𝑏 2 /𝐷, considering the half of the gap

between the boundaries as the characteristic dimension 𝑏. Further, the analytical
expression for the streamwise component of transfer function ⟨𝑁∥ ⟩ was obtained as:
3 1 3
( +1) tanh(√𝑗𝜔 ∗ )− ∗
𝑗𝜔∗ √𝑗𝜔∗ 𝑗𝜔
⟨𝑁∥ ⟩ = 1 (E3)
1− tanh(√𝑗𝜔 ∗ )
√𝑗𝜔∗

̃ ⟩ and ⟨𝑁∥ ⟩
The variations of both magnitude and phase in Figs. E1b and E1c for ⟨𝑆ℎ

10
respectively show excellent agreement between numerical simulation and the analytical
solution, though at high frequency the numerical results deviate from analytical values
because of the finite grid size used in numerical simulations. In practice we used a grid
size of 1000×1000 cells in most cases to minimize computational expense. For high 𝑃𝑒
cases more grid points were used to satisfy CFL criteria.

̃ 〉 and ⟨𝑁∥ ⟩ with 𝜔∗ that were obtained using different numbers of

Fig. E2. Variation of 〈𝑆ℎ
grid points applied to the periodic domain of cylinders depicted in Fig. 3c from the main
text.

̃ ⟩ and ⟨𝑁∥ ⟩ simulations were

To check the grid dependence of our predictions for ⟨𝑆ℎ
performed for cylinder arrays at 𝑃𝑒 = 100 using different grids, the results of which are
shown in Fig. E2. The most significant deviations among different grids were observed
in the limits of low and high frequency. However, a grid of 1000 × 1000 cells adequately
captures the fine-grid resolution while limiting computational expense.

F. Pore-scale concentration profiles

Here we analyze the shape of the concentration profiles at the midplane of the
cylinder within a periodic array and along the direction transverse to the streamwise
direction. To determine how frequency influences such profiles beyond that limit, the
magnitude and phase (𝜃ℂ = tan−1(Im[ℂ]⁄Re[ℂ])) of ℂ(𝜔, 𝒓) are shown in Fig. F1 on the
same surface, defined by the red-dashed line shown in Fig. 3c from the main text. All
profiles exhibit |ℂ| = 1 and 𝜃ℂ = 0 at the solid/solution interface, consistent with the
surface condition imposed in Eq. 4 from the main text. All profiles show a monotonic
decrease of |ℂ| toward zero as distance from the interface increases, while 𝜃ℂ increases
monotonically with distance from the interface for all profiles, except those simulated at
𝑃𝑒 ≫ 1 with either 𝑓 ∗ ≪ 1 or 𝑓 ∗ ≫ 200. At any given location of interest all profiles show
increased phase lag with increasing frequency, except for those simulated with 𝑓 ∗ ≫ 200
at 𝑃𝑒 ≫ 1. For 𝑓 ∗ ≫ 200 and 𝑃𝑒 ≫ 1 the non-monotonic variation of 𝜃ℂ with increasing
distance and frequency is caused by the high rate of solute advection under such
conditions. For 𝑓 ∗ ≪ 1 the lagless tracking of surface concentration causes 𝜃ℂ to vanish
uniformly.

11
Fig. F1. Spatial variation of |ℂ| and 𝜃ℂ in the transverse direction at the pore throat for (a)
𝜀 = 0.546 with 𝑃𝑒 = 0.019 (first row) and 𝑃𝑒 = 380 (second row), (b) 𝜀 = 0.874 with 𝑃𝑒 =
0.01 (first row) and 𝑃𝑒 = 300 (second). Spatial variation of |𝑐̂ | in the transverse direction
at the pore throat for (c) 𝜀 = 0.546 with 𝑃𝑒 = 0.019 (left) and 𝑃𝑒 = 380 (right), (d) 𝜀 =
0.874 with 𝑃𝑒 = 0.01 (left) and 𝑃𝑒 = 300 (right).

The effect of 𝑃𝑒 on system response in the low frequency limit (𝑓 ∗ ≪ 1) is not

apparent from Figs. F1a and F1b as a result of the vanishing degree of spatial variation
in ℂ that occurs in the limit of vanishing frequency. To compare the shape of
concentration profiles in this limit at finite frequency we also calculated a non-dimensional
concentration profile 𝑐̂ whose surface value is identically zero and whose volume-
averaged value is unity in the frequency domain: 𝑐̂ (𝜔, 𝒓) ≡ (𝑐̅ − 𝑐̅𝑠 )⁄(〈𝑐̅〉 − 𝑐̅𝑠 ), which
simplifies to 𝑐̂ (𝜔, 𝒓) = (ℂ − 1)⁄(〈ℂ〉 − 1). Consistent with our findings deduced from Figs.
F1a and F1b, the 𝑐̂ profiles in Fig. F1c and F1d reveal that the spatial extent of solution
over which concentration varies decreases with increasing frequency, producing
interfacial concentration gradients with magnitude that increases as frequency increases.
Such a profile is reminiscent of the diffusion boundary layers that are observed in
unbounded flows for sufficiently large frequencies. In addition, a negligible effect of Pe
on these profiles is observed at sufficiently high frequencies (𝑓 ∗ ≫ 200). In contrast,
Péclet number influences profile shape substantially at sufficiently low frequencies (𝑓 ∗ ≪
1), resembling the profiles obtained from simulations that we performed in the pseudo-
steady limit [1]. In summary, these results suggest that response in the low-frequency
limit is controlled primarily by the dominant transport mechanism in the solution’s bulk
(pseudo-steady diffusion and/or advection), while response in the high-frequency limit is

12
controlled by transient diffusion through boundary layers of solution that exhibit
frequency-specific thickness adjacent to interfaces.

G. Discretization of macro-scale MCEs and numerical implementation

We note that, while charging an RFB with known applied current variation 𝑖𝑎𝑝𝑝 (𝑡),
the instantaneous local current distribution 𝑖𝑙𝑜𝑐𝑎𝑙 (𝒓, 𝑡) inside the reactor is in general
unknown. However, 𝑖𝑙𝑜𝑐𝑎𝑙 (𝒓, 𝑡) must satisfy the current-overpotential relation at the
interface (Eq. 11 from the main text), and the mean of the distribution must be equal to
the instantaneous applied current: 〈𝑖𝑙𝑜𝑐𝑎𝑙 (𝒓, 𝑡)〉 = 𝑖𝑎𝑝𝑝 (𝑡). Since the solid phases are
electronically connected to the current collector, the solution obtained from Eqs. 10 and
13 in the main text must reproduce the equipotential condition, which requires that the
solid phase potential at every location inside the reactor is identical. We used this
equipotential condition to develop an iterative algorithm where at each time step the initial
distribution of local current was assumed to be uniform and equal to the instantaneous
applied current distribution 𝑖𝑎𝑝𝑝 (𝑡). First, Eqs. 10 and 13 from the main text were
discretized using the finite volume method to obtain a set of algebraic equations that are
simultaneously solved to obtain ⟨𝑐𝑖 ⟩ and ⟨𝑐𝑖 ⟩𝑇 using a direct solver, where the face
quantities were approximated using an up-wind scheme. Second, surface concentrations
of both species were calculated from the local concentration polarization, which was
obtained from the inverse Fourier transform of a rational transfer function approximation
of the reciprocal of the spectral Sherwood number, defined as 〈𝑍̃〉 = 1/〈𝑆ℎ ̃ 〉, to obtain a
state-space formulation in the time domain. Third, the local solid-phase potentials 𝜙𝑆 were
calculated (Eq. 12) from local overpotential 𝜂 values obtained by a Butler-Volmer (B-V)
kinetics expression (Eq. 11). Lastly, the equipotential condition was checked within a
specified nondimensional tolerance (𝑆𝑇𝐷(𝜙𝑆 )/〈𝜙𝑆 〉 ≤ 10−14 ). If not satisfied, the entire
process was repeated for a subsequent iteration starting with the non-uniform, scaled
local current distribution from last iteration. The scaling of current distribution was done
to ensure that the obtained non-uniform current distribution had a mean value equal to
the instantaneous applied current i.e., 〈𝑖𝑙𝑜𝑐𝑎𝑙 (𝒓, 𝑡)〉 = 𝑖𝑎𝑝𝑝 (𝑡). This iterative process was
repeated until the equipotential condition was satisfied. Then the algorithm proceeded to
the next time step. The iterative algorithm is shown as a block diagram in Fig. G1.

Below are described each of the components of the numerical methods used for the
macro-scale model.

• State space representation of transfer functions

Here we describe the step-by-step procedure to obtain the inverse Fourier

transforms of the frequency dependent TFs to solve Eq.10 in the time domain. Since the
associated PDEs and BCs are all linear, the system described here is a linear time-
invariant (LTI) system [12,13]. Further, 〈𝑆ℎ̃ 〉 is a scalar quantity, and for uniform flow
through symmetric (with respect to 𝒖𝑠 ) electrode microstructure, the transverse
component of 〈𝑵〉 is zero, i.e., ⟨𝑁⊥ ⟩ = 0, which allowed us to define a single-input/single-
output (SISO) state-space model (SSM) for both TFs separately. For such systems a
rational transfer function 〈𝑇𝐹〉 can be expressed using its state space representation as:

13
 〈𝑇𝐹〉 = 𝐶(𝑗𝜔𝐼 − 𝐴)−1 𝐵 + 𝐷 (G1)

Here, 𝐼 is an identity matrix, and 𝐴, 𝐵, 𝐶, and 𝐷 are the system matrices/vectors of the
corresponding state-space model. The system matrices/vectors were obtained from the
state-space representation that correlates the input 𝑢 to the output 𝑦 of the 〈𝑇𝐹〉 via a
state vector 𝑥 as governed by Eq. G2.
𝑑𝑥
State equation: =𝐴𝑥+𝐵𝑢 (G2a)
𝑑𝑡

Output equation: 𝑦 =𝐶𝑥+𝐷𝑢 (G2b)

The state matrix 𝐴 consists of the poles 𝑝𝑖 as non-zero elements along its diagonal. The
input vector 𝐵 is a column vector of ones. The output vector 𝐶 is a row vector with residues
𝑟𝑖 as its elements. 𝐷 is zero for a proper transfer function, of which 〈𝑍̃〉 and 〈𝑵〉 are specific
forms of. Since 〈𝑆ℎ ̃ 〉 is improper 〈𝑆ℎ̃ 〉(𝜔 → ∞) → ∞ (Fig. 4) we define its reciprocal as
〈𝑍̃〉 = 1⁄〈𝑆ℎ̃ 〉, which is a proper TF. We note that 〈𝑍̃〉 represents the diffusional impedance
of the system [14,15]. The input 𝑢 and output 𝑦 were taken from the definition of the
corresponding TFs as shown in Table G1.

Fig. G1. Flow chart showing the steps in the self-consistent iteration algorithm used at
each time step to solve macro-scale mass conservation equations (Eqs. 10 and 13 from
the main text), while the equipotential condition is checked at each iteration using Eqs.
11 and 12 from the main text.

14
Table G1: List of input 𝑢 and output 𝑦 for various transfer functions to implement their
corresponding state space model.
Transfer Frequency domain Time domain
Type
function Input 𝑢 Output 𝑦 Input 𝑢 Output 𝑦
〈𝑍̃〉 proper 𝑏 〈𝑗̅𝑖 〉𝑠 ⁄𝐷 𝑐̅𝑠,𝑖 − 〈𝑐̅𝑖 〉 𝑏 〈𝑗𝑖 〉𝑠 ⁄𝐷 𝑐𝑠,𝑖 − 〈𝑐𝑖 〉
⟨𝑁∥ ⟩ proper 𝑐̅𝑠,𝑖 − 〈𝑐̅𝑖 〉 𝜀 〈𝑊 ̅𝑖 〉⁄|𝑢𝑠 | 𝑐𝑠,𝑖 − 〈𝑐𝑖 〉 𝜀 〈𝑊𝑖 〉⁄|𝑢𝑠 |

The poles 𝑝𝑖 and the residues 𝑟𝑖 of the TFs 〈𝑍̃〉 and ⟨𝑁∥ ⟩ were obtained from a
rational function approximation with a finite number of poles (details in the following sub-
section). Equation F2 was discretized using an implicit scheme and solved at each time
step for each finite volume cell by using a direct solver to obtain their corresponding state
variables 𝑥 and the outputs 𝑦 in time domain, as shown in Table G1. First, the
concentration polarization 𝑐𝑠,𝑖 − 〈𝑐𝑖 〉 was obtained from 〈𝑍̃〉 using instantaneous local
current density (i.e., 𝑏 〈𝑗𝑖 〉𝑠 ⁄𝐷) as input. Lastly, the covariance of velocity and
concentration field 𝜀 〈𝑊𝑖 〉⁄|𝑢𝑠 | was obtained from ⟨𝑁∥ ⟩ considering 𝑐𝑠,𝑖 − 〈𝑐𝑖 〉 as input.

• Rational function approximation and vector fitting of transfer functions

To obtain the poles 𝑝𝑖 and residues 𝑟𝑖 , the TFs 〈𝑍̃〉 and ⟨𝑁∥ ⟩ were approximated as
rational transfer functions with finite numbers of poles (Eq. G3).
𝑟𝑖
〈𝑇𝐹〉 = ∑𝑁
𝑖=1 (G3)
𝑗𝜔−𝑝𝑖

Here, 𝑁 is the number of poles approximated, and 〈𝑇𝐹〉 = 〈𝑍̃〉 or 〈𝑁∥ 〉. To obtain 𝑝𝑖 and 𝑟𝑖
a vector fitting algorithm [16–18] was used. This algorithm assumes an arbitrary initial
guess for poles 𝑝𝑖 , that makes Eq. G3 linear with 𝑟𝑖 as unknowns. A set of linear algebraic
equations for 𝑟𝑖 was obtained by expressing Eq. G3 at each frequency sampled to yield
an overdetermined linear system of equations 𝐴𝑥 = 𝑏, where 𝑥 is a column vector of
unknown residues. This overdetermined problem was solved using a least-squares fitting
technique. To improve the accuracy of the fitting these steps were repeated using an
iterative scheme, where the initial guess of the poles in the subsequent iteration was
updated by a new set of poles obtained from Eq. G3 using the residues from the previous
iteration. Weighting functions were used for the least-squares fitting, based on the
magnitude of data sampled at each frequency, so as TF values with small magnitude
were given greater weight. Stable poles were ensured by changing the sign of the real
part of the poles whenever necessary (i.e., pole shifting). To check the accuracy of the
vector fitting algorithm, we reproduced the TFs using poles and zeros obtained from the
vector fitting algorithm and compared with the original TFs values in Fig. G2. Vector fitting
with 𝑁 = 30 poles/zeros resulted in excellent accuracy, where the reproduced TFs
showed maximum deviation among all sampled frequencies less than 10−2 % for all
cases.

15
 ̃ 〉 and ⟨𝑁∥ ⟩ for 𝜀 = 0.546 (left panel)
Fig. G2. Bode plot of the transfer functions 〈𝑍̃〉 = 1⁄〈𝑆ℎ
and 𝜀 = 0.874 (right panel) porosity electrodes. Pore-scale simulation data are shown as
solid lines and the fitted data are shown in circles.

• Selection of time step size and number of finite volumes and verification of
numerical implementation of the macroscopic model

Here, we present the dependence of the results on the time step size ∆𝑡 and the
number of finite volumes 𝑁𝐹𝑉 used in the one-dimensional macroscale model. Further,
we used our TF embedded model to investigate the response for a finite diffusional space
between two reactive parallel boundaries, for which analytical results were obtained by
solving the associated MCEs in the time domain. The transient response of such a simple
geometry obtained from the transient model introduced here is compared with the
analytical results.

• Selection of time step size and number of finite volumes

Figure G3a shows the variation of the numerically calculated volume-averaged

concentration polarization with time step size ∆𝑡 used in the numerical implementation.
The absolute values of the volume-averaged concentration polarization are normalized
by their arithmetic mean (as the figure is designated to show the variation in data only).
The figure shows that for ∆𝑡 ≤ 0.1 𝑠, the solution approaches an asymptotic value
showing negligible dependence on ∆𝑡. In our calculations different values of ∆𝑡 have been
used (10−6 𝑠 ≤ ∆𝑡 ≤ 10−1 𝑠) to minimize computation time, where smaller and larger ∆𝑡
were used for short and large duration impulse simulations, respectively.
16
 The finite volume discretization of associated MCEs ensures the conservation of
mass for a finite-sized representative control volume. Thus, volume-averaged quantities
did depend on the number of finite volumes 𝑁𝐹𝑉 used in the calculation. Thus, we show
the dependency of the maximum polarization produced inside the reactor with 𝑁𝐹𝑉 in Fig.
F3b. The data are normalized by their arithmetic mean to emphasize their variations. The
results shows a negligible dependence of maximum polarization on 𝑁𝐹𝑉 for 𝑁𝐹𝑉 ≥ 50
(𝑁𝐹𝑉 = 50 produces deviation that is less than 10−1 % compared to 𝑁𝐹𝑉 = 200).
Consequently, we used 𝑁𝐹𝑉 = 50 in all simulations to minimize the computation cost while
capturing the local maximum values with acceptable accuracy.

Fig. G3. (a) Time step dependence of numerical results for the macro-scale model. (b)
Dependence of numerical results on number of finite volumes 𝑁𝐹𝑉 used in numerical
simulations. (c) Schematic of the model used for model verification. (d) Variation of
surface concentration 𝑐𝑠 with time in the model shown in (c). (e) Comparison of volume
averaged concentration 〈𝑐〉 variation with Fourier number 𝐹𝑜 = 𝐷𝑡/𝐿2 obtained from
analytical solution and the bottom-up transient model (B-UTM). Comparison of time
variation of the surface flux 〈𝑗〉𝑠 obtained from analytical solution and B-UTM (f) in linear
scale and (g) in logarithmic scale.

• Verification of numerical implementation of the macroscopic model

In Fig. G3c we show the geometry of the simplistic problem used for model
verification. Figure G3c shows a stationary finite diffusional space between two parallel
reactive boundaries separated by distance 2𝐿 (length along 𝑦 direction is considered
infinite), where surface concentrations were varied identically by a step jump at time 𝑡 =
0, as shown in Fig. F3d [i.e., 𝑐|𝑥=0 = 𝑐|𝑥=2𝐿 = 𝑐𝑠 (𝑡)]. Thus, the finite diffusional space
enclosed can be considered as symmetric with respect to the midplane at 𝑥 = 𝐿 (i.e.,
𝜕𝑐⁄𝜕𝑥|𝑥=𝐿 = 0). For such a problem the analytical solution can be obtained using

17
separation of variables. The analytical solution obtained is shown in Eq. F4 using
nondimensional parameters.
4 (2𝑛+1)𝜋 (2𝑛+1)2 𝜋 2
𝑐 ∗ = 1 − ∑∞
𝑛=0 (2𝑛+1)𝜋 𝑠𝑖𝑛 ( 𝑥 ∗ ) 𝑒𝑥𝑝 (− 𝐹𝑜) (G4)
2 4

Here, 𝑐 ∗ = 𝑐/𝑐𝑠 , 𝑥 ∗ = 𝑥/𝐿 and the Fourier number 𝐹𝑜 = 𝐷𝑡/𝐿2 . The macroscopic
response of the system can be observed from the time variation of non-dimensional
1 𝐿 〈𝑗〉
volume-averaged concentration 〈𝑐 ∗ 〉 = 𝐿 ∫0 𝑐 ∗ 𝑑𝑥 and the surface flux 〈𝑗〉∗𝑠 = 𝐷𝑐 ⁄𝑠𝐿 =
𝑠
−𝜕𝑐 ∗ ⁄𝜕𝑥 ∗ |𝑥 ∗=0 as shown by Eqs. G5 and G6.
8 (2𝑛+1)2 𝜋2
⟨𝑐⟩∗ = 1 − ∑∞
𝑛=0 (2𝑛+1)2 𝜋 2 𝑒𝑥𝑝 (− 𝐹𝑜) (G5)
4

(2𝑛+1)2 𝜋 2
〈𝑗〉∗𝑠 = 2 ∑∞
𝑛=0 𝑒𝑥𝑝 (− 𝐹𝑜) (G6)
4

For such a geometry, the analytical expressions for transfer functions 〈ℂ〉 and 〈𝑆ℎ ̃〉
are shown in Eqs. E1 and E2, which were derived from the Fourier transform of the
corresponding MCEs. Because 〈𝑆ℎ ̃ 〉 is an improper transfer function, we defined its
reciprocal as 〈𝑍̃〉 = 1/〈𝑆ℎ̃ 〉 and obtained the macro-scale response of the system using
the TF embedded transient model following identical approaches discussed earlier in this
section (i.e., identical discretization, self-consistent iterative algorithm, state space
representation, rational function approximation, and vector fitting). The results for ⟨𝑐⟩∗
from the analytical solution and the B-UTM model are shown in Fig. G3e. The results
show that the B-UTM matches the analytical solution for 10−5 ≤ 𝐹𝑜 ≤ 10−2 with maximum
deviation of 10−2 %. Meanwhile, for 𝐹𝑜 ≥ 5 the difference between the surface
concentration 𝑐𝑠 and volume-averaged concentration 〈𝑐〉 becomes negligible with an
infinitesimal flux at the interface. Thus, the variation of 〈𝑗〉∗𝑠 obtained from the analytical
solution and the B-UTM model are shown for 10−5 ≤ 𝐹𝑜 ≤ 5 on a linear scale (Fig. G3f)
and on a logarithmic scale (Fig. G3g) to compare them in the low and high 𝐹𝑜 limits.
These curves show excellent agreement with maximum deviation of 2 × 10−2 % from the
analytical results.

H. Multiple long duration impulses

When acceleration/suppression of reactive solute is accounted for as with our

present bottom-up transient model (B-UTM), an initial step in current from open-circuit
conditions causes the electrode to contain less reactant and more product in the bulk
solution (Fig. 6 from the main text). Since the initial step lasts for the entire
charging/discharging duration studied, the corresponding acceleration/suppression effect
is not recovered. Rather, this pre-existing effect gets superimposed with the
acceleration/suppression effect introduced later at 𝑡/𝜏 = 0. A schematic of such
superposition is depicted in Fig. H1 for a double-step transient in the applied current. For
this illustration we assumed that the corresponding impulse time for both the impulses
was large enough to approach the pseudo-steady limit (PSL).

18
Fig. H1. Superposition of solute acceleration/suppression effect for multiple-step variation
of input current.

Figure H1 shows the prediction from both the conventional model (CM) and the B-
UTM model, where CM shows continuous change in reactant concentration with time.
Meanwhile, B-UTM shows an abrupt change in 〈〈𝑐𝑟𝑒𝑎𝑐 〉〉 at the beginning of both the
current steps, and during 𝑖𝑎𝑝𝑝 = 𝑖2 the observed effect has contributions from both
transients. When current is returned to 𝑖1 the acceleration/suppression, the partial
contribution of the second transient vanishes immediately, while the partial contribution
of the initial transient persists.

Supplementary References

[1] M.A. Hamid, K.C. Smith, Modeling the Transient Effects of Pore-Scale Convection
and Redox Reactions in the Pseudo-Steady Limit, J. Electrochem. Soc. 167 (2020)
013521. https://doi.org/10.1149/2.0212001jes.
[2] H. Sarac, M.A. Patrick, A.A. Wragg, Physical properties of the ternary electrolyte
potassium ferri-ferrocyanide in aqueous sodium hydroxide solution in the range 10-
90°C, J. Appl. Electrochem. 23 (1993) 51–55.
https://doi.org/https://doi.org/10.1007/BF00241575.
[3] J.C. Bazán, A.J. Arvia, The diffusion of ferro- and ferricyanide ions in aqueous
solutions of sodium hydroxide, Electrochim. Acta. 10 (1965) 1025–1032.

19
 https://doi.org/10.1016/0013-4686(65)80014-7.
[4] N.G. Tsierkezos, A. Knauer, U. Ritter, Thermodynamic investigation of
ferrocyanide/ferricyanide redox system on nitrogen-doped multi-walled carbon
nanotubes decorated with gold nanoparticles, Thermochim. Acta. 576 (2014) 1–8.
https://doi.org/10.1016/J.TCA.2013.11.020.
[5] K. Gong, F. Xu, J.B. Grunewald, X. Ma, Y. Zhao, S. Gu, Y. Yan, All-Soluble All-Iron
Aqueous Redox-Flow Battery, ACS Energy Lett. 1 (2016) 89–93.
https://doi.org/10.1021/acsenergylett.6b00049.
[6] P. Kulesza, T. Jedral, Z. Galus, The electrode kinetics of the
Fe(CN)63−/Fe(CN)64− system on the platinum rotating electrode in the presence
of different anions, J. Electroanal. Chem. Interfacial Electrochem. 109 (1980) 141–
149. https://doi.org/https://doi.org/10.1021/ja00357a007.
[7] N. Frenzel, J. Hartley, G. Frisch, Voltammetric and spectroscopic study of ferrocene
and hexacyanoferrate and the suitability of their redox couples as internal standards
in ionic liquids †, Phys. Chem. Chem. Phys. 19 (2017) 28841–28852.
https://doi.org/10.1039/c7cp05483a.
[8] Y. Jaluria, K. Torrance, Computational heat transfer, Hemisphere Publishing
Corporation, 1986.
[9] J. Anderson Jr, Fundamentals of Aerodynamics, 2001.
https://doi.org/10.1036/0072373350.
[10] S. Patankar, Numerical heat transfer and fluid flow, CRC Press, 1980.
[11] L. R., Courant, K., O., Friedrichs, H., On the partial difference equations of
mathematical physics, IBM J. Res. 11 (1967) 215–234.
https://doi.org/10.1109/TED.2022.3218267.
[12] X. Hu, S. Stanton, L. Cai, R.E. White, A linear time-invariant model for solid-phase
diffusion in physics-based lithium ion cell models, J. Power Sources. 214 (2012)
40–50. https://doi.org/10.1016/j.jpowsour.2012.04.040.
[13] X. Hu, S. Stanton, L. Cai, R.E. White, Model order reduction for solid-phase
diffusion in physics-based lithium ion cell models, J. Power Sources. 218 (2012)
212–220. https://doi.org/10.1016/j.jpowsour.2012.07.007.
[14] T. Jacobsen, K. West, Diffusion impedance in planar, cylindrical and spherical
symmetry, Electrochim. Acta. 40 (1995) 255–262. https://doi.org/10.1016/0013-
4686(94)E0192-3.
[15] A. Caprani, C. Deslouis, S. Robin, B. Tribollet, Transient mass transfer at partially
blocked electrodes: a way to characterize topography, J. Electroanal. Chem. 238
(1987) 67–91. https://doi.org/10.1016/0022-0728(87)85166-5.
[16] B. Gustavsen, A. Semlyen, Rational approximation of frequency domain responses
by vector fitting, IEEE Trans. Power Deliv. 14 (1999) 1052–1059.
https://doi.org/10.1109/61.772353.
[17] B. Gustavsen, Improving the pole relocating properties of vector fitting, IEEE Trans.
Power Deliv. 21 (2006) 1587–1592. https://doi.org/10.1109/TPWRD.2005.860281.
[18] D. Deschrijver, M. Mrozowski, T. Dhaene, D. De Zutter, Macromodeling of multiport
systems using a fast implementation of the vector fitting method, IEEE Microw.
Wirel. Components Lett. 18 (2008) 383–385.
https://doi.org/10.1109/LMWC.2008.922585.

20