Processing math: 35%
Research article Special Issues

Evaluating Community Preferences for Waste-to-Energy Development in Jakarta: An Analysis Using the Choice Experiment Method

  • Received: 12 September 2023 Revised: 18 November 2023 Accepted: 27 November 2023 Published: 05 December 2023
  • The Indonesian Presidential Regulation No. 35 of 2018 heralds a transformative agenda for producing electrical energy from biomass in an environmentally conscious manner. Jakarta emerges as a pivotal area in this transformation, with its strategic adoption of Waste to Energy (WtE) systems. In this study, we probe into the multiple layers of community preferences and the acceptance of WtE developments in Jakarta, factoring in an array of local concerns and policy-driven directives. Through a methodologically structured choice experiment, participants weighed in on various scenarios delineating shifts from the status quo to innovative WtE technological adoptions. we scrutinize a spectrum of attributes, each with defined status quo levels and proposed advancements: From enhancing awareness of landfill impacts (P1), escalating local policy commitments (K1) and integrating waste treatment facilities (F1) to diversifying waste processing outputs into liquid (PP1), solid (PP2), gas (PP3) and electricity (PP4). We also consider the transition from unmanaged landfills to controlled applications of landfill gas (PA1) and thermal treatment (PA2), as well as the initiation of emission and pollutant monitoring (M1).Our findings illuminate a significant public inclination to move beyond the current paradigms towards embracing WtE conversions, with particular willingness to support socialization of new waste processing technologies (P1), generation of energy in various forms especially liquid (PP1) and electricity (PP4) and implementation of environmental monitoring measures (M1). These attributes were marked by a notable willingness to accept (WTA) the proposed changes, signaling a readiness for policy and infrastructural advancements in waste management.

    Citation: Aarce Tehupeiory, Iva Yenis Septiariva, I Wayan Koko Suryawan. Evaluating Community Preferences for Waste-to-Energy Development in Jakarta: An Analysis Using the Choice Experiment Method[J]. AIMS Environmental Science, 2023, 10(6): 809-831. doi: 10.3934/environsci.2023044

    Related Papers:

    [1] Zheng Dai, I.G. Rosen, Chuming Wang, Nancy Barnett, Susan E. Luczak . Using drinking data and pharmacokinetic modeling to calibrate transport model and blind deconvolution based data analysis software for transdermal alcohol biosensors. Mathematical Biosciences and Engineering, 2016, 13(5): 911-934. doi: 10.3934/mbe.2016023
    [2] Lernik Asserian, Susan E. Luczak, I. G. Rosen . Computation of nonparametric, mixed effects, maximum likelihood, biosensor data based-estimators for the distributions of random parameters in an abstract parabolic model for the transdermal transport of alcohol. Mathematical Biosciences and Engineering, 2023, 20(11): 20345-20377. doi: 10.3934/mbe.2023900
    [3] Marcella Noorman, Richard Allen, Cynthia J. Musante, H. Thomas Banks . Analysis of compartments-in-series models of liver metabolism as partial differential equations: the effect of dispersion and number of compartments. Mathematical Biosciences and Engineering, 2019, 16(3): 1082-1114. doi: 10.3934/mbe.2019052
    [4] Shuang-Hong Ma, Hai-Feng Huo . Global dynamics for a multi-group alcoholism model with public health education and alcoholism age. Mathematical Biosciences and Engineering, 2019, 16(3): 1683-1708. doi: 10.3934/mbe.2019080
    [5] Hai-Feng Huo, Shuang-Lin Jing, Xun-Yang Wang, Hong Xiang . Modelling and analysis of an alcoholism model with treatment and effect of Twitter. Mathematical Biosciences and Engineering, 2019, 16(5): 3561-3622. doi: 10.3934/mbe.2019179
    [6] Salih Djillali, Soufiane Bentout, Tarik Mohammed Touaoula, Abdessamad Tridane . Global dynamics of alcoholism epidemic model with distributed delays. Mathematical Biosciences and Engineering, 2021, 18(6): 8245-8256. doi: 10.3934/mbe.2021409
    [7] Shuang Hong Ma, Hai Feng Huo, Hong Xiang, Shuang Lin Jing . Global dynamics of a delayed alcoholism model with the effect of health education. Mathematical Biosciences and Engineering, 2021, 18(1): 904-932. doi: 10.3934/mbe.2021048
    [8] Ridouan Bani, Rasheed Hameed, Steve Szymanowski, Priscilla Greenwood, Christopher M. Kribs-Zaleta, Anuj Mubayi . Influence of environmental factors on college alcohol drinking patterns. Mathematical Biosciences and Engineering, 2013, 10(5&6): 1281-1300. doi: 10.3934/mbe.2013.10.1281
    [9] José Daniel Padilla-de la-Rosa, Mario Alberto García-Ramírez, Anne Christine Gschaedler-Mathis, Abril Ivette Gómez-Guzmán, Josué R. Solís-Pacheco, Orfil González-Reynoso . Estimation of metabolic fluxes distribution in Saccharomyces cerevisiae during the production of volatile compounds of Tequila. Mathematical Biosciences and Engineering, 2021, 18(5): 5094-5113. doi: 10.3934/mbe.2021259
    [10] Hugo Flores-Arguedas, Marcos A. Capistrán . Bayesian analysis of Glucose dynamics during the Oral Glucose Tolerance Test (OGTT). Mathematical Biosciences and Engineering, 2021, 18(4): 4628-4647. doi: 10.3934/mbe.2021235
  • The Indonesian Presidential Regulation No. 35 of 2018 heralds a transformative agenda for producing electrical energy from biomass in an environmentally conscious manner. Jakarta emerges as a pivotal area in this transformation, with its strategic adoption of Waste to Energy (WtE) systems. In this study, we probe into the multiple layers of community preferences and the acceptance of WtE developments in Jakarta, factoring in an array of local concerns and policy-driven directives. Through a methodologically structured choice experiment, participants weighed in on various scenarios delineating shifts from the status quo to innovative WtE technological adoptions. we scrutinize a spectrum of attributes, each with defined status quo levels and proposed advancements: From enhancing awareness of landfill impacts (P1), escalating local policy commitments (K1) and integrating waste treatment facilities (F1) to diversifying waste processing outputs into liquid (PP1), solid (PP2), gas (PP3) and electricity (PP4). We also consider the transition from unmanaged landfills to controlled applications of landfill gas (PA1) and thermal treatment (PA2), as well as the initiation of emission and pollutant monitoring (M1).Our findings illuminate a significant public inclination to move beyond the current paradigms towards embracing WtE conversions, with particular willingness to support socialization of new waste processing technologies (P1), generation of energy in various forms especially liquid (PP1) and electricity (PP4) and implementation of environmental monitoring measures (M1). These attributes were marked by a notable willingness to accept (WTA) the proposed changes, signaling a readiness for policy and infrastructural advancements in waste management.



    Historically, researchers and clinicians interested in tracking alcohol consumption and metabolism in the field would require data from either a drinker's self-report or from having them use a breath alcohol analyzer. Because both methods require active participation by the subject, the data they produce are often plagued by inaccuracies. Self-report often leads to misrepresentation as (1) subjects may deviate from naturalistic behaviors due to the reporting requirement seeming unnatural, and (2) alcohol directly impairs subjects' ability to be an active participant [1]. Using a breath alcohol analyzer correctly requires specialized training and can produce erroneous measurements due to mouth alcohol and/or a reading based on a shallow breath by the subject. Dating back to the 1930's, ethanol, the type of alcohol in alcoholic beverages, has been known to be excreted from the human body through the skin [2,3,4,5]. This is due to the fact that water and ethanol are highly miscible [6] and the ethanol finds its way into all of the water in the body. More recently, this observation paved the way for the development of a device to measure the amount of alcohol excreted transdermally through the skin [8,9]. The benefits derived from such a device include the availability of near continuous measurements and the ability to collect them passively (i.e., without the active participation of the subject). This gives researchers and clinicians the potential to continuously observe naturalistic drinking behavior and patterns. There is also the possibility of making these devices available on the consumer market (e.g., wearable body system monitoring technology like Fitbits, Apple watches, etc.). In addition, the ideas we discuss here may also be applicable to the monitoring of other substances once the appropriate sensor hardware has been developed.

    The challenge in using transdermal alcohol sensors is that they provide transdermal alcohol concentration (TAC), whereas alcohol researchers and clinicians have always based their studies and treatments on measurements of breath alcohol concentration (BrAC) and blood alcohol concentration (BAC). Thus, a means to reliably and accurately convert TAC to BrAC or BAC would be desirable. At levels up to approximately 0.08 (see, for example, [10,11]) BrAC correlates well with BAC via a simple linear relationship based on an empirical relationship known as Henry's law [12,13]: BAC=ρB/Br×BrAC, where the constant ρB/Br is known as the partition coefficient of ethanol in blood and breath.

    More generally, according to Henry's law, when a liquid is in contact with a gas, the concentrations, CL and CG, of a compound present in both the liquid and the gas will come to equilibrium according to the linear relationship CL=ρL/GCG, where the empirical determined constant ρL/G is known as the partition coefficient for the that compound in that liquid and gas. Not surprisingly, the partition coefficient, ρL/G, is temperature dependent and of course its actual value will vary depending on the choice of units for CL and CG. It has been shown (see, for example, [14]) that at 34C, the partition coefficient for ethanol in blood and air is ρB/A=2157±9.6 for men and ρB/A=2195±10.9 for women, at 37C, the partition coefficient for ethanol in blood and air is ρB/A=1783±8.1 for men and ρB/A=1830±7.8 for women. Using a regression model, Jones [14] found that at 37C the partition coefficient for ethanol in water and air is ρW/A=2133, in blood and air is ρB/A=1756, and between plasma (all of the components of blood with the exception of the oxygen carrying red blood cells) and air is ρP/A=2022. All of these values are for the case when the concentration of ethanol in air is given in units of grams per liter, and in water, blood, or plasma in units of grams per deciliter. We note that it is generally accepted that a BrAC reading of 0.08 percent alcohol corresponds to.008 grams of ethanol per 210 liters of breath and a BAC of 0.08 grams of ethanol per 100 milliliters (equal to 1 deciliter (dL) or 0.1 liters (L)) of blood.

    Unfortunately, however, the correlation between TAC and BrAC/BAC, on the other hand, can vary due to a number of confounding factors. These factors include, but are not limited to, stable features of the skin like its thickness, tortuosity, and porosity, particularly as they apply to the epidermal layer of the skin, which does not have an active blood supply. Environmental factors such as ambient temperature and humidity can also affect both perspiration and vasodilation, and can thus alter skin conductance, blood flow, the amount of alcohol passing below the skin in the blood, and the amount and rate of alcohol diffusing through the skin. One would also expect there to be manufacturing and operational variations among different TAC sensors.

    Earlier attempts to investigate the relationship between TAC and BrAC/BAC have used deterministic models [15,16,17,18,19,20,21]. Some utilized regression-based models [16], whereas others utilized first principles physics-based models that on occasion included modeling the transport of alcohol all the way from ingestion to excretion through the skin [22,23]. In our group's initial efforts, we modeled the transport of alcohol from the blood in the dermal layer through the epidermal layer and its eventual measurement by the sensor using a one-dimensional diffusion equation [15,21]. The parameters in the diffusion equation model then had to be fit or tuned (i.e., calibrated) to each individual subject, the environmental conditions, and the device through the use of simultaneous BrAC/TAC training data collected in the laboratory or clinic through a procedure known as an alcohol challenge. Once the model was fit, it could then be used to deconvolve BrAC from TAC collected in the field. This two-pass approach and the related studies were relatively successful [15,19,20,21,24,25]. However, this calibration procedure is quite burdensome to researchers, clinicians, subjects and patients, and because the models were fit to a single uni-modal drinking episode, unaccounted for variation and uncertainty in the relationship between BrAC and TAC frequently arose, making it difficult to accurately convert TAC collected in the field to BrAC [26,27].

    More recently, to eliminate the need for calibration, deconvolution of BrAC from TAC was effected using population models fit to BrAC/TAC training data from drinking episodes across a cohort of subjects, devices, and environmental conditions [24,25,28]. These population models took the form of the deterministic transport models but now the parameters appearing in the model equations were considered to be random. Then in fitting the models, instead of estimating the actual values of these parameters, it was their joint distributions that were estimated. Once the models were fit, they could be used to deconvolve an estimate of the BrAC input, and by making use of the distribution of the population parameters, conservative error bands could also be generated which quantified the uncertainty in the estimated BrAC [24,25]. The results in these studies were based on a naive pooled data statistical model and a non-linear least squares estimator.

    In this paper, we seek to build on the approach described in the previous paragraph by now using a Bayesian approach to account for the underlying uncertainty and variation in the alcohol diffusion and measurement process. We obtain posterior distributions for the transport model parameters conditioned on the training BrAC/TAC data and regularized by prior distributions based on deterministic fits. Being Bayesian based, our approach yields credible sets for the estimated parameters and what we shall refer to as conservative credible or error bands for the deconvolved estimated BrAC. What is meant by the term conservative credible band will be made precise later.

    An outline of the remainder of the paper is as follows. In the next section of the paper we provide a description of our method including a derivation of a new abstract parabolic hybrid PDE/ODE model for the transdermal transport of alcohol through he epidermal layer of the skin and its capture in the vapor collection bay of the sensor. Then using linear semigroup theory we obtain an input/output model in the form of a discrete time convolution. A discussion of finite dimensional approximation and convergence issues related to the use of our model to carry out the requisite computations is also included. Then in the results section of the paper we first construct our Bayesian estimator and present two theoretical results related to it: convergence of the finite dimensional approximation and consistency. We then show how our population model based on Bayesian estimates for the random parameters can be used as part of a deconvolution scheme that yields estimated BrAC curves and conservative credible or error bands from a biosensor provided TAC signal. In this section we also present and discuss a sample of our numerical findings demonstrating the efficacy of our approach. Our numerical studies were based on human subject data collected in the Luczak laboratory in the Department of Psychology at USC. A final section contains some discussion of our theoretical and numerical results along with a few concluding remarks and avenues for possible future research.

    As in [21] and [24], and making use of an idea recently introduced in [28], we model the alcohol biosensor problem described in Section 1 using a one dimensional diffusion equation to describe alcohol transport through the epidermal layer of the skin coupled with an inflow/outflow compartment model to describe the perspiration vapor collection chamber of the TAC biosensor.

    The epidermal layer of the skin sits atop the dermal layer. The dermal layer has an active blood supply while the epidermal layer does not. The latter consists of both dead (the stratum corneum layer which is closest to the surface) and living (the deeper layers closer to the dermal layer) cells surrounded by interstitial fluid. Not having an active blood supply, the cells in the epidermal layer obtain nourishment primarily from O2 that diffuses in from the environment beyond the skin.

    The SCRAM TAC biosensor (see 1 in section 3.4 below) has a perspiration vapor collection chamber on the bottom of the sensor that sits atop, and is in direct contact with, the stratum corneum layer of the skin's epidermal layer. Perspiration in vapor form collects in the chamber. A small pump extracts a sample of the vapor from the collection chamber approximately once every 30 minutes. This sample is then electro-chemically analyzed based on an oxidation-reduction (redox) reaction in much the same way that a fuel cell produces a current (and heat and water) from hydrogen and oxygen. In the TAC sensor, ethanol molecules in the sample are oxidized producing electrons in the form of an electrical current. This current is converted into the TAC measurement based on an a priori bench calibration.

    To make this more precise, we let Λ denote the thickness of the epidermal layer (units: cm) of the skin at the location of the sensor and let η denote the depth in the epidermal layer (units: cm), 0ηΛ, η=0 denoting the skin surface and η=Λ denoting the boundary between the epidermal and dermal layers. Let t denote time (units: hrs) and let x(t,η) denote the concentration of ethanol at time t and depth η in the epidermal layer (units: grams per milliliter of interstitial fluid). Let ˜w(t) denote the concentration of ethanol in the TAC sensor collection chamber at time t (units: grams per milliliter of air), and let u(t) denote the BrAC at time t (units: grams per milliliter of air). Let y(t) denote the TAC at time t (units: grams per milliliter of air), and let ˜w0 (units: grams per milliliter of air) and φ0 (units: grams per milliliter of interstitial fluid) denote the initial conditions for ˜w and x, respectively. We will assume that there is no ethanol in either the epidermal layer or the TAC biosensor collection chamber at time t=0 so ˜w0=0 and φ0=0. Let T denote the duration of the drinking episode (units: hrs). Then, With these definitions, our model takes the form

    xt(t,η)=α2xη2(t,η),0<η<Λ,0<t<T,d˜wdt(t)=γαρP/Axη(t,0)δ˜w(t),0<t<T,x(t,0)=ρP/A˜w(t),0<t<Tαxη(t,Λ)=βρP/Au(t),0<t<T,˜w(0)=˜w0,x(0,η)=φ0(η),0<η<Λ,y(t)=θ˜w(t),0<t<T, (2.1)

    where α>0 denotes the effective diffusivity of ethanol in the interstitial fluid in the epidermal layer (units: cm2/hr), β>0 denotes the effective linear flow rate at which capillary blood plasma from the dermal layer replenishes the interstitial fluid in the epidermal layer (units: cm/hr), and ρP/A denotes the partition coefficient for ethanol in plasma and air with respect to the concentration units of grams per milliliter of plasma and grams per milliliter of air at 37C (normal body temperature).

    In modeling the TAC collection chamber, we assume that the inflow of ethanol is proportional to the flux out of (i.e., from right to left) the epidermal layer at the surface of the skin (i.e., at η=0), αxη(t,0), with constant of proportionality γ (units: cm1), and the outflow is simply proportional to the concentration of ethanol in the collection chamber (i.e., a simple linear model) with constant of proportionality δ (units: hr1. Finally, the output gain, θ, represents the bench calibration factor for the TAC sensor that converts the concentration or ethanol in the collection chamber into a TAC (units: dimensionless).

    Since the thickness of the epidermal layer, Λ, is in general difficult to measure and can be mathematically difficult to estimate computationally due to it determining the spatial domain of the diffusion equation, it is desirable to transform the system eq. (2.1) to a domain of fixed length, Λ=1. We make the standard change of variable ηηΛ thus rendering η dimensionless. For t0, We also set w(t)=ρP/A˜w(t). Then, recalling our assumption of zero initial conditions, the following hybrid ordinary-partial differential equation input/output system results

    xt(t,η)=q12xη2(t,η),0<η<1,t>0,dwdt(t)=q3xη(t,0)q4w(t),t>0,x(t,0)=w(t),t>0q1xη(t,1)=q2u(t),t>0,w(0)=0,x(0,η)=0,0<η<1,y(t)=w(t),t0, (2.2)

    where q1=αΛ2, q2=θβρP/AΛ, q3=γαΛ, and q4=δ. We note that since the only observable and observed quantities are BrAC, u, and TAC, y, the physiological interpretations of the variables and parameters in between that define our model in the form of an input/output map from BrAC to TAC are of little interest to us. Although we have relied on first principles modeling to derive the system of equations given in eq. (2.2), our motivation was not to gain a deeper understanding of the transdermal transport of ethanol. Rather, it was to be able to keep the dimension of the space of unknown parameters as low as possible by capturing the underlying physics and physiology of the transport process, albeit in a greatly simplified form. Indeed, our primary objective here is to first fit the parameters (or, more precisely, their distributions) in the model to observed input/output BrAC/TAC training pairs and to then use the resulting population model to obtain an estimate for the BrAC and associated error bars corresponding to a given TAC signal collected in the field from a member of the cohort or population that provided the data which was used to train the model.

    Let q=[q1,q2]T denote the unknown, un-measurable, and, in general, subject-dependent physiological parameters. The parameters q3 and q4 are device (i.e., hardware) dependent parameters and as such, we consider them to be bench-measurable empirically in the lab. We do note however, with simple changes of variable, the theory and methods we develop below apply, and their distributions could also be estimated along with those of q1 and q2 with the same techniques we use here to estimate the distributions of q1 and q2. In addition, q3 and q4 could also be estimated using a deterministic scheme such as a regularized nonlinear least squares approach. For clarity and ease of exposition, we will focus our attention here on the development of a population model for a cohort of subjects by estimating the distribution of the un-measurable physiological parameters q1 and q2.

    We consider this system on a finite-time horizon, 0T<, and we assume zero-order hold input, u(t)=uk, t[kτ,(k+1)τ), k=0,1,2,..., where τ denotes the sampling time of the biosensor. We set xk=xk(η)=x(kτ,η), wk=w(kτ), and yk=y(kτ), k=0,1,,K, where we assume T=Kτ. For k=0,1,2,... we consider the system eq. (2.2) on the interval [kτ,(k+1)τ] and make the change of variable: v(t,η)=x(t,η)ξ(η)uk where ξ(η)=q2q1η. It is then easily verified that w and v satisfy the following hybrid system

    vt(t,η)=q12vη2(t,η),0<η<1,kτ<t<(k+1)τ,dwdt(t)=q3vη(t,0)q4w(t)+q3q2q1uk,kτ<t<(k+1)τ,v(t,0)=w(t),kτ<t<(k+1)τ,q1vη(t,1)=0,kτ<t<(k+1)τ, (2.3)

    with initial conditions v(kτ,)=x(kτ,)ξ()uk=xkξuk and w(kτ)=wk on [0,1].

    We then use linear semigroup theory to rewrite the system eq. (2.3) in state space form in an appropriately chosen Hilbert space and subsequently obtain a discrete time evolution system for (wk,xk), k=0,1,2,....K which is equivalent to eq. (2.2). Let Q denote a compact subset of the positive orthant of R4, and for q=[q1,q2,q3,q4]TQ we define the Hilbert spaces

    Hq=R×L2(0,1)andV={(θ,ψ)Hq:ψH1(0,1),θ=ψ(0)} (2.4)

    with respective corresponding inner products (θ1,ψ1),(θ2,ψ2)q=(q1/q3)θ1θ2+10ψ1(η)ψ2(η)dη and (ψ1(0),ψ1),(ψ2(0),ψ2)V=ψ1(0)ψ2(0)+10ψ1(η)ψ2(η)dη, and respective norms ||q, and V. Note that the Sobolev Embedding Theorem [29] yields that the norm induced by the V inner product is equivalent to the standard H1 norm on V. It is not difficult to show that V is densely and continuously embedded in Hq and that we have the Gelfand triple of dense and continuous embeddings VHqV.

    Then based on the weak formulation of the system eq. (2.3), for each qQ define the bilinear form a(q,,):V×VR by a(q,ˆφ,ˆψ)=(q1q4/q3)φ(0)ψ(0)+q110φ(η)ψ(η)dη for ˆφ,ˆψV, where ˆφ=(φ(0),φ), ˆψ=(ψ(0),ψ), and q=[q1,q2,q3,q4]TQ. Standard arguments can be used to argue that the form a(q,,) satisfies the following three properties.

    1. Boundedness There exists a constant α0>0 such that |a(q,ˆψ1,ˆψ2)|α0ˆψ1Vˆψ2V, ˆψ1,ˆψ2V,

    2. Coercivity There exists constants λ0R and μ0>0 such that a(q,ˆψ,ˆψ)+λ0|ˆψ|2qμ0ˆψ2V,ˆψV.

    3. Continuity For all ˆψ1,ˆψ2V, we have that qa(q,ˆψ1,ˆψ2) is a continuous mapping from Q into R.

    Note that in (1)–(3) above, Q compact implies that the constants α0, λ0, and μ0 may all be chosen independent of qQ. Furthermore, properties (1)–(3) immediately yield that the form a(q,,) defines a bounded, elliptic (i.e., λ0=0) operator A(q)L(V,V) given by A(q)ˆφ,ˆψ=A(q)(φ(0),φ),(ψ(0),ψ)=a(q,φ,ψ) for ˆφ=(φ(0),φ),ˆψ=(ψ(0),ψ)V. If we define the set D={ˆφV:A(q)ˆφHq}={(φ(0),φ)V:φH2(0,1),φ(1)=0} which is independent of q for qQ, we obtain the closed, densely defined linear operator A(q):DHqHq given by A(q)ˆφ=A(q)(φ(0),φ)=(q3φ(0)q4φ(0),q1φ"),ˆφ=(φ(0),φ)D. The operator A(q):DHqHq is regularly dissipative and (see, for example, [30]) is the infinitesimal generator of a holomorphic semigroup of bounded linear operators {eA(q)t:t0} on Hq and V. Moreover, the system eq. (2.3) then has a state space form where dˆvdt(t)=A(q)ˆv(t)+(q3q2q1,0)uk and ˆv(kτ)=(wk,xkξuk) for kτ<t<(k+1)τ. Then for time step τ>0 and k=0,1,,K, letting ˆxk=(wk,xk), ˆA(q)=eA(q)τL(Hq,Hq), and ˆB(q)=(IˆA(q)){(0,ξ)A(q)1(q3q2q1,0)}L(R,Hq), it follows that

    ˆxk+1=(wk+1,xk+1)=(w((k+1)τ),x((k+1)τ,))=ˆv((k+1)τ)+(0,ξ)uk=eA(q)τ(wk,xkξuk)+τ0eA(q)s(q3q2q1,0)dsuk+(0,ξ)uk=ˆA(q)ˆxk+(IˆA(q))(0,ξ)uk+A(q)1(ˆA(q)I)(q3q2q1,0)uk=ˆA(q)ˆxk+ˆB(q)uk (2.5)

    where for ˆB(q) we have used the fact that the operator A(q)1 commutes with the semigroup generated by A(q). Note that the operator ˆB(q) is in fact an element in Hq and that (0,ξ)V, but that (q3q2q1,0) is only an element in Hq. From eq. (2.5) the state space form of our discrete time model is

    ˆxk+1=ˆA(q)ˆxk+ˆB(q)uk,k=0,1,2,,ˆx0=(w0,φ0),yk=ˆCˆxk,k=0,1,2,, (2.6)

    where ˆxk=(wk,xk), k=0,1,2, and the operator ˆCL(Hq,R) is given by ˆC(θ,ψ)=θ, where (θ,ψ)Hq. Note that the ellipticity of A(q) guarantees the existence of A(q)1.

    From eq. (2.6) it is immediately clear that if we assume zero initial conditions, ˆx=(w0,φ0)=(0,0), the output y can be written as a discrete time convolution of the input, u, with a filter, h(q), as

    yk=k1j=0ˆCˆA(q)kj1ˆB(q)uj=k1j=0hkj1(q)uj,k=0,1,2,, (2.7)

    where for qQ, hi(q)=ˆCˆA(q)iˆB(q), i=0,1,2,. Using the Trotter-Kato semigroup approximation theorem (see, for example, [31]), the following result can be shown (see, for proof, [32]).

    Lemma 2.1. For Q a compact subset of the positive orthant or R4, K=Kτ for constant time step τ>0, and hi as defined in eq. (2.7), we have that the mapping qhi(q) from Q into R is continuous, uniformly in q and i, for qQ and i{0,1,2,,K}.

    Now although the input/output model given in eq. (2.7) is a standard convolution in R, the filter, {hk(q)} involves the semigroup {eA(q)t:t0} which is defined on the infinite dimensional Hilbert space Hq. Consequently, finite dimensional approximation is required. For n=1,2, let {φni(η)}ni=0 denote the standard linear B-splines on the interval [0,1] defined with respect to the uniform mesh {0,1n,2n,,n1n,1}, φni(η)=(nηi+1)1[i1n,in]+(1nη+i)1[in,i+1n]. Set

    Vn=span{ˆφni}=span{(φni(0),φni)}ni (2.8)

    and let Pnq:HqVn denote the orthogonal projection of Hq on to Vn along (Vn). Standard arguments from the theory of splines (see, for example, [33]) can be used to argue that |Pnq(θ,ψ)(θ,ψ)|q0, as n, for all (θ,ψ)Hq, and that PnqˆφˆφV0, as n, for all ˆφV with the convergence uniform in q for qQ.

    For n=1,2,... and k=0,1,2,... we set ˆxnk(η)=ni=0Xn,kiˆφni(η), and we approximate the operator A(q) using a Galerkin approach. That is, we define the operator An(q)L(Vn,Vn) by restricting the form a(q,,) to Vn×Vn. We then set

    ˆAn(q)=eAn(q)τ,andˆBn(q)=(InˆAn(q)){(0,ξ)An(q)1Pnq(q3q2q1,0)}, (2.9)

    where ˆBn(q)L(R,Vn)=Vn. The matrix representations for these operators with respect to the basis {ˆφni}ni=0 are then given by [An(q)]=[Mn(q)]1Kn(q), [ˆAn(q)]=e[Mn(q)]1Kn(q)τ, and [ˆBn(q)]=(I[ˆAn(q)]){Ξn[An(q)]1[Mn(q)]1[q2,0,0,,0]T}=(I[ˆAn(q)]){Ξn+[Kn(q)]1[q2,0,0,,0]T}, where [M]ni,j=0=(φni(0),φni),(φnj(0),φnj)q, [K]ni,j=0=a(q,φni(0),φni),(φnj(0),φnj)), and Ξn=q2q1[0,1n,2n,,n1n,1]T. Letting [ˆCn]=[1,0,0,,0]R1×(n+1), we consider the discrete time dynamical system in Vn given by

    ˆxnk+1=ˆAn(q)ˆxnk+ˆBn(q)uk,k=0,1,2,,K1ynk=ˆCˆxnk,k=0,1,2,,K,ˆxn0=(0,0)Vn (2.10)

    or equivalently in Rn+1 given by the system where Xn,k+1=[ˆAn(q)]Xn,k+[ˆBn(q)]uk, ynk=[ˆCn]Xn,k, and Xn,0=[0,0,,0]TRn+1, we obtain that

    ynk=k1j=0ˆCnˆAn(q)kj1ˆBn(q)uj=k1j=0[ˆCn][ˆAn(q)]kj1[ˆBn(q)]uj=k1j=0hnkj1(q)uj, (2.11)

    for k=0,1,2,,K where hni(q)=[ˆCn][ˆAn(q)]i[ˆBn(q)]R, i=0,1,2,,K1.

    Using linear semigroup theory (see, for example, [21,34,35]) and in particular the Trotter-Kato semigroup approximation theorem (see, for example, [36] and [31]) the following results can be established (for proof, see [32]).

    Theorem 2.1. For Q a compact subset of the positive orthant of R4, n=1,2,, {φni(η)}ni=0 the standard linear B-splines on the interval [0,1] defined with respect to the uniform mesh {0,1n,2n,,n1n,1}, Vn=span{(φni(0),φni)}ni, Pnq:HqVn the orthogonal projection of Hq on to Vn along (Vn), and ˆAn(q) and ˆBn(q) defined as in eq. (2.9), we have that |ˆAn(q)Pnq(θ,ψ)ˆA(q)(θ,ψ)|q0, as n, for all (θ,ψ)Hq, that ˆAn(q)PnqˆφˆA(q)ˆφV0, as n, for all ˆφV, and that ˆBn(q)ˆB(q)V0, as n, with the convergence in all cases uniform in q for qQ.

    Theorem 2.2. Under the same hypotheses as Theorem 2.1, we have that ˆxnk(q)ˆxk(q)V0, that |ˆxnk(q)ˆxk(q)|q0, that |ynkyk|0, and that |hnk(q)hk(q)|0 as n uniformly in k for k{0,1,2,...,K} and uniformly in q for qQ.

    Finally, we will assume that we have training data, {{uik}Kk=0,{yik}Kk=0}Ri=1, from R participants or subjects where without loss of generality (i.e., by padding with zeros) we have assumed that all training input/output datasets have the same number, K, of observations. In this case, for i=1,,R we have,

    yik=k1j=0hkj1(q)uij,andyn,ik=k1j=0hnkj1(q)uij,k=0,1,,K, (2.12)

    where hj(q)=ˆCˆA(q)jˆB(q)R and hnj(q)=[ˆCn][ˆAn(q)]j[ˆBn(q)]R, for j=0,1,,K1. This formulation facilitates the estimation of the population parameters q. If one wishes to find the parameters q for a specific individual, the methods outlined in Section 2.2 can still be applied by letting the indices i=1,,R refer to different measured BrAC/TAC events each with k=0,,K denoting the measurement times for the desired individual subject.

    In this section we develop a Bayesian framework to estimate the unknown parameters q=[q1,q2]T in the system eq. (2.2). To illustrate our approach, for simplicity but without loss of generality, we have assumed that the sensor parameters q3 and q4 have been bench-measured and are therefore known and concentrate our effort on estimating the physiological subject-dependent parameters q1 and q2. All of what follows below can easily be extended to estimating all four of the parameters in the model. Our underlying statistical model incorporating noise is based on the observation of {yij} as in eq. (2.12) and is given by

    Vij=yij+εij=j1=0hj1ui+εij (2.13)

    where Vij are our measured TAC values, and εij are the i.i.d. noise terms corresponding to person i at time jτ with σ>0,τ>0. Commonly, as we will assume in Section 3.2 and beyond, εijN(0,σ2). In order to be able to carry out the requisite computations, we consider the approximating statistical model based on eq. (2.12)

    Vij=yn,ij+εij=j1=0hnj1ui+εij, (2.14)

    where once again the Vij's are assumed to be the measured TAC values. We consider q to be a random vector on some probability space {Ω,Σ,P} with support Q and assume that the prior distribution of q is given by the push forward measure π0. That is for AQ, P({qA})=P(q1(A))=q1(A)dP=Adπ0(q).

    We assume independence across both i (individuals) and j (sampling times for each individual), for each i and j we have Vijyij=εij (commonly distributed N(0,σ2)) and similarly, Vijyn,ij=εij (again commonly distributed N(0,σ2)). Letting φ denote the density of εij's, for qQ the likelihood and the approximating likelihood functions are given respectively by (see, for example, [37,38,39,40])

    L(q|{Vij})=Ri=1Kj=1φ(Vijyij),andLn(q|{Vij})=Ri=1Kj=1φ(Vijyn,ij).

    An application of Bayes' Theorem (see, for example, Theorem 1.31 in [41]) yields that the posterior distribution of q or the conditional distribution of q conditioned on the data, {Vij}, is a push forward measure π=π(|{Vij}) that is absolutely continuous with respect to π0 and whose Radon-Nikodym derivative, or density, for qQ is given by

    dπdπ0(q|{Vij})=L(q|{Vij})QL(q|{Vij})dπ0(q)=1ZRi=1Kj=1φ(Vijyij(q)), where (2.15)
    Z=QL(q|{Vij})dπ0(q)=QRi=1Kj=1φ(Vijyij(q))dπ0(q). (2.16)

    In this way, for AQ, we have P(qA|{Vij})=Adπ(q)=Adπ(q|{Vij}). If, in addition, we have π0λ with density dπ0dλ=f0 where λ denotes Lebesgue measure on Q, then πλ with conditional density f given by

    f(q)=dπdλ(q|{Vij})=L(q|{Vij})f0(q)QL(q|{Vij})f0(q)dλ(q)=1ZRi=1Kj=1φ(Vijyij(q))f0(q), (2.17)
    Z=QL(q|{Vij})f0(q)dλ(q)=QRi=1Kj=1φ(Vijyij(q))f0(q)dλ(q),and (2.18)
    P({qA}|{Vij})=Af(q)dλ(q)=Af(q|{Vij})dλ(q). (2.19)

    Analogously, in the case of the approximating likelihood eq. (2.15), eq. (2.16), eq. (2.17), eq. (2.18), and eq. (2.19) respectively become

    dπndπ0(q|{Vij})=Ln(q|{Vij})QLn(q|{Vij})dπ0(q)=1ZRi=1Kj=1φ(Vijyn,ij(q)), (2.20)
    Zn=QLn(q|{Vij})dπ0(q)=QRi=1Kj=1φ(Vijyn,ij(q))dπ0(q), (2.21)
    fn(q)=Ln(q|{Vij})f0(q)QLn(q|{Vij})f0(q)dλ(q)=1ZnRi=1Kj=1φ(Vijyn,ij(q))f0(q), (2.22)

    Zn=QLn(q|{Vij})f0(q)dλ(q)=QRi=1Kj=1φ(Vijyn,ij(q))f0(q)dλ(q), and P(qA|{Vij})=Afn(q)dλ(q)=Afn(q|{Vij})dλ(q).

    Consider the random variable q with posterior distribution described by the measure π given in eq. (2.15) and eq. (2.16), and let qn denote the random variable q but with posterior distribution πn given by eq. (2.20) and eq. (2.21). In this section we establish that qnDistq as n; that is that qn converges in distribution to q. Recall that due to the physical constraints based on our model for the alcohol biosensor problem, eq. (2.2), we require that the parameters q lie in the interior of the positive orthant of R2.

    Theorem 3.1. For Q a compact set in the interior of the positive orthant of R2, a prior π0 with compact support Q and a density that is continuous on Q, and a noise distribution with bounded density function φ and support on R, qn, the random variable with posterior distribution πn given by eq. (2.20) and eq. (2.21) converges in distribution to the random variable q with posterior distribution π given by eq. (2.15) and eq. (2.16).

    Proof. For S a subset Q, the triangle inequality yields

    |P(qnS|{Vij})P(qS|{Vij})|=|1ZnSLn(q|{Vij})dπ0(q)1ZSL(q|{Vij})dπ0(q)||1Z1Zn|(SL(q|{Vij})dπ0(q))+(1Zn)S|L(q|{Vij})Ln(q|{Vij})|dπ0(q), (3.1)

    where φ is the normal density describing the distribution of the noise term in eq. (2.13), and Z and Zn are as in eq. (2.16) and eq. (2.21), respectively.

    Focusing first on the limit of |1/Z1/Zn| as n, by Lemma 2.1 we have that the yji(q) are continuous in q for qQ, i{0,1,,R}, and j{0,1,,K}. Since Q is compact, the {yij(q)} are bounded and thus 0<Z<. By Theorem 2.2, since yn,ij(q)yij(q) uniformly in q for qQ as n, 0<Zn< for n large enough. Again by Theorem 2.2 it follows from eq. (2.16), eq. (2.21), and the Bounded Convergence Theorem that ZnZ as n and therefore that |1/Z1/Zn|0 as n. Then, essentially the same arguments yield that S|L(q|{Vij})Ln(q|{Vij})|dπ0(q)0, from which it then immediately follows that (1Zn)S|L(q|{Vij})Ln(q|{Vij})|dπ0(q)0, and therefore from eq. (3.1) that qnDistq as n and the theorem has been proved.

    For the push forward measures π and πn from eq. (2.15) and eq. (2.20), respectively, we are commonly interested in their respective expected values and the convergences there within. Since Q is compact, a direct invocation of the Portmanteau theorem (see, [41]) establishes the following corollary which guarantees the convergence of all moments described by πn to those of π.

    Corollary 3.1. Under the hypotheses of Theorem 3.1, and for any continuous function g:QR we have that Eπn[g(q)]=Qg(q)dπn(q)Qg(q)dπ(q)=Eπ[g(q)] as n.

    In this section we demonstrate the strong consistency of the posterior distribution with respect to the parameters, q, by imposing stronger assumptions on the distribution of the noise terms εij in eq. (2.13) and on the prior, π0, by restricting Q to a rectangle in the positive orthant of R2, and by applying the framework summarized in [42].

    As in [42], we show consistency of the posterior distribution π as in eq. (2.17), rather than consistency of a point estimator based on the posterior distribution. As such, for prior π0 over Q, posterior π(|{Vij}) as in eq. (2.15), and i.i.d. noise εijN(0,σ2) for σ>0, we also consider that for qQ assumed known we have random variables VijN(yij(q),σ2) as determined by eq. (2.13) for i=1,2,,R and j=1,2,,K. Further, we have that {Vij}i,j are independent in i and j, but are non-identically distributed (i.n.i.d).

    For clarity and brevity we consider the random vector Vi=(Vi0,Vi1,,ViK)T with values in RK+1 and independent entries derived from the matrix equivalent to eq. (2.13) and eq. (2.14), namely Vi=yi+εi=Hui+εi and Vi=yn,i+εi=Hnui+εi, with noise vectors εi=(εi0,εi1,,εiK)T, observed TAC vectors Vi=(Vi0,Vi1,,ViK)T, theoretical TAC vectors yi=(yi0,ji1,,yiK)T and analogous yn,i, BrAC data vectors ui=(ui0,ui1,,uiK)T and analogous un,i, and kernel matrices H and Hn with entries [H]i,j=hij1ji and [Hn]i,j=hnij1ji, respectively. In this way, for every i{1,2,,R}, by independence in j we have the family of joint distributions, {fi,q()=Kj=0φ(jyij(q)):qQ}, representing the possible densities of Vi for φ the noise density and yij as in eq. (2.12). We are interested in the scenario where the number of subjects R.

    With our reframing, for AQ by independence in i we may rewrite eq. (2.15) as

    π(A|{Vi})=ARi=1fi,q(Vi)dπ0(q)QRi=1fi,q(Vi)dπ0(q)=AL(q|{Vi})dπ0(q)QL(q|{Vi})dπ0(q)=AL(q|{Vi})L(q0|{Vi})dπ0(q)QL(q|{Vi})L(q0|{Vi})dπ0(q)=JA({Vi})J({Vi}), (3.2)

    where for all i we have data vectors Vi, and for our purposes we will be interested in the equivalent form on the right-hand side of the equation above where \mathit{\boldsymbol{q}}_0\in Q is the true value of our parameters [q_1, q_2]^T .

    We first formalize the results discussed in Section 7 of [42] that handle the i.n.i.d case of posterior consistency. As such, we say that our posterior distributions \{ \pi(\cdot \vert \{ \mathit{\boldsymbol{V}}^i \}) \} as in eq. (3.2) are strongly consistent at \mathit{\boldsymbol{q}}_0 if \{ \pi(U \vert \{ \mathit{\boldsymbol{V}}^i \}) \} \rightarrow 1 a.s P_{q_0}^{\infty} for every neighborhood U of \mathit{\boldsymbol{q}}_0 as R\rightarrow \infty , where P_{ \mathit{\boldsymbol{q}}_0}^{\infty} = \prod_{i = 1}^{\infty}P_{i, \mathit{\boldsymbol{q}}_0} with P_{i, \mathit{\boldsymbol{q}}_0} the probability distribution generated by f_{i, \mathit{\boldsymbol{q}}_0} with data samples \{ \mathit{\boldsymbol{V}}^i \} .

    For this we show that for sets A\subset Q with \mathit{\boldsymbol{q}}_0 \notin A , J_A(\{ V_k \})\rightarrow 0 and J(\{ \mathit{\boldsymbol{V}}^i \})\rightarrow \infty as R\rightarrow \infty in some appropriate manner to be made precise below. For J_A(\{ \mathit{\boldsymbol{V}}^i \})\rightarrow 0 we take the same approach as expressed in [42] and thus state the following definition below without motivation, where we note that for any two densities f , g on some space \mathcal{X} their affinity, denoted \text{Aff}(f, g) , is given by \text{Aff}(f, g) = \int_{\mathcal{X}} \sqrt{f(x)g(x)} dx .

    Definition 3.1. Let A\subset Q and \delta > 0 . The set A and \mathit{\boldsymbol{q}}_0 are strongly \delta separated if for every probability measure \nu on A , \text{Aff}(f_{1, \mathit{\boldsymbol{q}}_0}, \mathcal{v}_{\nu}^1) < \delta where f_{1, \mathit{\boldsymbol{q}}_0}(\mathit{\boldsymbol{x}}) = \prod_{j = 0}^K \varphi(x_j - y^1_j(\mathit{\boldsymbol{q}}_0)) for \mathit{\boldsymbol{x}}\in \mathbb{R}^{K+1} as in the work surrounding eq. (3.2), and \mathcal{v}_{\nu}^R is the marginal density of \{ \mathit{\boldsymbol{\mathcal{V}}}^i \}_{i = 1}^R given by \mathcal{v}_{\nu}^R (\{ \mathit{\boldsymbol{V}}^i \}_{i = 1}^R) = \int_A L(\mathit{\boldsymbol{q}} | \{ \mathit{\boldsymbol{V}}^i \}_{i = 1}^R) d\nu(\mathit{\boldsymbol{q}}) for any R = 1, 2, \ldots . We will say that A and \mathit{\boldsymbol{q}}_0 are strongly separated if they are strongly \delta separated for some \delta > 0 .

    From these definitions we provide the following theorem without proof. For proof see Sections 3 and 7 of [42] (or, for examples, [43])

    Theorem 3.2. Let \pi_0 be a prior over parameter space Q , \{ \mathit{\boldsymbol{V}}^i \}_{i = 1}^{\infty} be independent but not identically distributed data with distribution generated by f_{i, \mathit{\boldsymbol{q}}} for \mathit{\boldsymbol{q}} \in Q , \mathit{\boldsymbol{q}}_0 \in Q the true value of the parameters [q_1, q_2]^T , and A \subset Q with \mathit{\boldsymbol{q}}_0 \notin A . If A = \bigcup\limits_{i \geq 1} A_i such that

    1. For some \delta > 0 , all A_i 's are strongly \delta separated from \mathit{\boldsymbol{q}}_0 for the model \mathit{\boldsymbol{q}} \mapsto f_{i, \mathit{\boldsymbol{q}}} , and

    2. \sum\limits_{i \geq 1} \sqrt{\pi_0(A_i)} < \infty ,

    then for some \beta_0 > 0 , e^{R\beta_0}J_A(\{ \mathit{\boldsymbol{V}}^i \}_{i = 1}^R) \rightarrow 0 \quad \mathit{\text{a.s.}} P_{ \mathit{\boldsymbol{q}}_0}^{\infty} as R \rightarrow \infty for J_A(\{ \mathit{\boldsymbol{V}}^i \}) as in eq. (3.2).

    To show J(\{ \mathit{\boldsymbol{V}}^i \})\rightarrow \infty as R\rightarrow \infty we utilize the approach as outlined in the proof Theorem 1 of Appendix A.2 in [44] (specifically the proof of (8) in Appendix A.2). For a direct proof see [32]. For a similar approach see [45].

    Theorem 3.3. Let \pi_0 be a prior over parameter space Q , \{ \mathit{\boldsymbol{V}}^i \}_{i = 1}^{\infty} be independent but not identically distributed data with distribution generated by f_{i, \mathit{\boldsymbol{q}}} for \mathit{\boldsymbol{q}} \in Q , and \mathit{\boldsymbol{q}}_0 \in Q the true value of the parameters [q_1, q_2]^T . For \mathit{\boldsymbol{q}}\in Q define \Lambda_i(\mathit{\boldsymbol{q}}_0, \mathit{\boldsymbol{q}}) = \log\dfrac{f_{i, \mathit{\boldsymbol{q}}_0}(\mathit{\boldsymbol{V}}^i)}{f_{i, \mathit{\boldsymbol{q}}}(\mathit{\boldsymbol{V}}^i)} , \mathcal{K}_i(\mathit{\boldsymbol{q}}_0, \mathit{\boldsymbol{q}}) = \mathbb{E}_{ \mathit{\boldsymbol{q}}_0}[\Lambda_i(\mathit{\boldsymbol{q}}_0, \mathit{\boldsymbol{q}})] , and S_i(\mathit{\boldsymbol{q}}_0, \mathit{\boldsymbol{q}}) = \mathrm{Var}_{ \mathit{\boldsymbol{q}}_0}[\Lambda_i(\mathit{\boldsymbol{q}}_0, \mathit{\boldsymbol{q}})] . If there exists a set B \subseteq Q with \pi_0(B) > 0 such that

    1. \sum\limits_{i \geq 1} \dfrac{S_i(\mathit{\boldsymbol{q}}_0, \mathit{\boldsymbol{q}})}{i^2} < \infty \quad\forall \mathit{\boldsymbol{q}} \in B , and

    2. For every \varepsilon > 0 , \pi_0(B \cap \{ \mathit{\boldsymbol{q}} : \mathcal{K}_i(\mathit{\boldsymbol{q}}_0, \mathit{\boldsymbol{q}}) < \varepsilon \quad \forall i \}) > 0 .

    Then \forall \beta > 0 , e^{R\beta}J(\{ \mathit{\boldsymbol{V}}^i \}_{i = 1}^R) \rightarrow \infty \quad \mathit{\text{a.s.}} P_{ \mathit{\boldsymbol{q}}_0}^{\infty} as R\rightarrow \infty for J(\{ \mathit{\boldsymbol{V}}^i \}) as in eq. (3.2).

    Before moving on to our main theorem we apply the following theorem and subsequent corollary to prove a lemma that will be of use to us later. For proof of the following see Theorem 5.3 of [46].

    Theorem 3.4. For parameters \mathit{\boldsymbol{q}} \in Q a subset of the positive orthant of \mathbb{R}^n , and \mathit{\boldsymbol{q}} -dependent semigroup \{ T(\cdot; \mathit{\boldsymbol{q}}) : t > 0 \} with infinitesimal generator A(\mathit{\boldsymbol{q}}) defined in terms of a \mathit{\boldsymbol{q}} -dependent sesquilinear form \sigma(\mathit{\boldsymbol{q}}): V\times V \rightarrow \mathbb{C} on a Hilbert space V satisfying items 1 and 2 in Section 2.1, assume

    1. (Affine) The map \mathit{\boldsymbol{q}} \mapsto \sigma(\mathit{\boldsymbol{q}}) is affine, in the sense that for any u, v \in V , \sigma(\mathit{\boldsymbol{q}}, u, v) = \sigma_0(u, v) + \sigma_1(\mathit{\boldsymbol{q}}, u, v) where \sigma_0 is independent of \mathit{\boldsymbol{q}} and the map \mathit{\boldsymbol{q}} \mapsto \sigma_1(\mathit{\boldsymbol{q}}, \cdot, \cdot) is linear, and

    2. (Continuous) For any \mathit{\boldsymbol{q}}, \bar{ \mathit{\boldsymbol{q}}} \in Q with metric d_Q(\cdot, \cdot) we have the bound \vert\sigma(\mathit{\boldsymbol{q}}, u, v) - \sigma(\bar{ \mathit{\boldsymbol{q}}}, u, v)\vert \leq d_Q(\mathit{\boldsymbol{q}}, \bar{ \mathit{\boldsymbol{q}}}) \lVert u \rVert_V\lVert v \rVert_V , for all u, v \in V .

    Then, the semigroup T(\cdot; \mathit{\boldsymbol{q}}) is (Fréchet) differentiable in \mathit{\boldsymbol{q}} in the interior of Q , where for t > 0 , \bar{ \mathit{\boldsymbol{q}}} \in Q , and acting on \delta \mathit{\boldsymbol{q}} \in Q the derivative is given by

    \begin{align} T_{ \mathit{\boldsymbol{q}}}(t, \bar{ \mathit{\boldsymbol{q}}})\delta \mathit{\boldsymbol{q}} = \dfrac{1}{2\pi i}\int_{\partial \Sigma_{\gamma}} e^{\lambda t} R(\lambda, \mathit{\boldsymbol{q}})A(\delta \mathit{\boldsymbol{q}})R(\lambda, \mathit{\boldsymbol{q}}) d\lambda \end{align} (3.3)

    for R(\lambda, \mathit{\boldsymbol{q}}) = (A(\mathit{\boldsymbol{q}})-\lambda I)^{-1} the resolvent of A(\mathit{\boldsymbol{q}}) , and the obtuse sector \Sigma_{\gamma} = \{ \lambda \in \mathbb{C} : \arg(\lambda - \lambda_0) \leq \frac{\pi}{2} + \arctan((1+\alpha_0\mu_0)(1-\gamma)) \} with \gamma \in (0, 1) , \alpha_0 as in item 1, and \lambda_0 , \mu_0 as in item 2.

    The following corollary is an immediate consequence of the work in [6]. Specifically that the map \mathit{\boldsymbol{q}} \mapsto R(\lambda, \mathit{\boldsymbol{q}}) is analytic as a map from Q to \mathscr{L}(V^{\ast}, V) , and from eq. (3.3) the map \bar{ \mathit{\boldsymbol{q}}} \mapsto T_{ \mathit{\boldsymbol{q}}}(t, \bar{ \mathit{\boldsymbol{q}}}) depends continuously on R(\lambda, \bar{ \mathit{\boldsymbol{q}}}) . Note that \Sigma_{\gamma} is independent of \mathit{\boldsymbol{q}} as the constants \alpha_0 and \lambda_0 , \mu_0 from items 1 and 2, respectively, of Section 2.1 are independent of \mathit{\boldsymbol{q}} .

    Corollary 3.2. Under the same hypotheses as Theorem 3.4, we have that the map \bar{ \mathit{\boldsymbol{q}}} \mapsto T_{ \mathit{\boldsymbol{q}}}(t, \bar{ \mathit{\boldsymbol{q}}}) is continuous in the operator norm on \mathscr{L}(Q, \mathscr{L}(V^{\ast}, V)) for \bar{ \mathit{\boldsymbol{q}}} in the interior of Q .

    Lemma 3.1. For Q a rectangle in the positive orthant of \mathbb{R}^2 , Hilbert spaces H_q and V as in eq. (2.4), bilinear form a(\mathit{\boldsymbol{q}}, \cdot, \cdot): V \times V \rightarrow \mathbb{R} as in Section 2.1 with q_3 and q_4 assumed known, and induced infinitesimal generator A(\mathit{\boldsymbol{q}}) as in Section 2.1, then the generated holomorphic semigroup of bounded linear operators \{ e^{A(\mathit{\boldsymbol{q}})t} : t \geq 0 \} on H_{ \mathit{\boldsymbol{q}}} and V^\ast is (Fréchet) differentiable and Lipschitz in \mathit{\boldsymbol{q}} in the interior of Q . Further, for i = 1, \ldots, R and j = 1, \ldots, K , y_j^i and y_j^{n, i} as in eq. (2.12) and eq. (2.12) are (Fréchet) differentiable and Lipschitz in \mathit{\boldsymbol{q}} with Lipschitz constants independent of i and j .

    Proof. First, by item 3 of Section 2.1 we have that for \mathit{\boldsymbol{q}} \in Q , \mathit{\boldsymbol{q}} \mapsto a(\mathit{\boldsymbol{q}}, \cdot, \cdot) is continuous in \mathit{\boldsymbol{q}} . Second, notice that for \hat{\psi}, \hat{\varphi} \in V with \hat{\psi} = (\psi(0), \psi) and \hat{\varphi} = (\varphi(0), \varphi) , we have a(\mathit{\boldsymbol{q}}, \hat{\psi}, \hat{\varphi}) = \dfrac{q_1 q_4}{q_3}\psi(0)\varphi(0) + q_1 \int\limits_0^1 \psi'(x)\varphi'(x)dx = a_0(\mathit{\boldsymbol{q}}, \psi, \varphi) + a_1(\mathit{\boldsymbol{q}}, \psi, \varphi) , where \mathit{\boldsymbol{q}} \mapsto a_0(\mathit{\boldsymbol{q}}, \psi, \varphi) and \mathit{\boldsymbol{q}} \mapsto a_1(\mathit{\boldsymbol{q}}, \psi, \varphi) are clearly linear in \mathit{\boldsymbol{q}} for \mathit{\boldsymbol{q}} \in Q . Hence the bilinear form a(\mathit{\boldsymbol{q}}, \cdot, \cdot) is affine and continuous in \mathit{\boldsymbol{q}} Thus, by Theorem 3.4, we have that the semigroup generated by a(\mathit{\boldsymbol{q}}, \cdot, \cdot) , \{ e^{A(\mathit{\boldsymbol{q}})t} : t \geq 0 \} = \{ T(t, \mathit{\boldsymbol{q}}) : t \geq 0 \} , is (Fréchet) differentiable in \mathit{\boldsymbol{q}} for \mathit{\boldsymbol{q}}\in Q . Denote the derivative in \mathit{\boldsymbol{q}} and \bar{ \mathit{\boldsymbol{q}}} acting on \delta \mathit{\boldsymbol{q}} \in Q by T_{ \mathit{\boldsymbol{q}}}(t, \bar{ \mathit{\boldsymbol{q}}})\delta \mathit{\boldsymbol{q}} .

    Moreover for t \in (0, T] we have for \mathit{\boldsymbol{q}}_1, \mathit{\boldsymbol{q}}_2 \in Q and line segment S = \{ s \mathit{\boldsymbol{q}}_1 + (1-s) \mathit{\boldsymbol{q}}_2 : 0 \leq s \leq 1 \} ,

    \begin{align} \lVert T(t, \mathit{\boldsymbol{q}}_1) - T(t, \mathit{\boldsymbol{q}}_2) \rVert_{\mathscr{L}( V^{\ast}, V)} &\leq \lVert \mathit{\boldsymbol{q}}_1 - \mathit{\boldsymbol{q}}_2 \rVert_1 \sup\limits_{\mathbf{x}\in Q}\lVert T_{ \mathit{\boldsymbol{q}}}(t, \mathbf{x}) \rVert_{\mathscr{L}(Q, \mathscr{L}( V^{\ast}, V))} \leq \tilde{C} \lVert \mathit{\boldsymbol{q}}_1 - \mathit{\boldsymbol{q}}_2 \rVert_1 \end{align} (3.4)

    with \lVert\cdot\rVert_{\mathscr{L}(V^{\ast}, V)} the operator norm, where the first inequality follows from the Mean Value Theorem on Banach spaces (see Theorem 4 of Section 3.2 in [47]), and the second and third inequalities follow from the compactness of Q , Corollary 3.2, and the continuity of the map T_{\mathit{\boldsymbol{q}}}(t, \bar{\mathit{\boldsymbol{q}}})) \mapsto \lVert T_{\mathit{\boldsymbol{q}}}(t, \bar{\mathit{\boldsymbol{q}}})) \rVert_{\mathscr{L}(Q, \mathscr{L}(V^{\ast}, V))} .

    Further, under zero-order hold the differentiability and Lipschitz properties of \hat{A}(\mathit{\boldsymbol{q}}) = e^{A(\mathit{\boldsymbol{q}})\tau} = T(\tau, \mathit{\boldsymbol{q}}) remain. Considering \hat{B}(\mathit{\boldsymbol{q}}) from Section 2.1, \hat{B}(\mathit{\boldsymbol{q}}) = (I - \hat{A}(\mathit{\boldsymbol{q}}))\big\{(0, \xi) - A(\mathit{\boldsymbol{q}})^{-1}(\dfrac{q_3q_2}{q_1}, 0)\big\} , we find that it is a sum and product of \mathit{\boldsymbol{q}} -differentiable and \mathit{\boldsymbol{q}} -Lipschitz terms and thus is differentiable and Lipschitz in \mathit{\boldsymbol{q}} . Since h and y as in eq. (2.12) are a composition and sum of \mathit{\boldsymbol{q}} -differentiable terms they remain differentiable. Further, using eq. (3.4) we have the following Lipschitz bound for all j \in \{0, 1, \ldots, K\} and \mathit{\boldsymbol{q}}, \bar{ \mathit{\boldsymbol{q}}} in the interior of Q ,

    \begin{equation} \begin{aligned} \vert h_j( \mathit{\boldsymbol{q}}) - h_j(\bar{ \mathit{\boldsymbol{q}}}) \vert &\leq \lVert \hat{C} \rVert_{\mathscr{L}(H_{ \mathit{\boldsymbol{q}}}, \mathbb{R})} \Big[ \lVert \hat{A}( \mathit{\boldsymbol{q}})^j - \hat{A}(\bar{ \mathit{\boldsymbol{q}}})^j \rVert_{\mathscr{L}(H_{ \mathit{\boldsymbol{q}}})} \lVert \hat{B}( \mathit{\boldsymbol{q}}) \rVert_{\mathscr{L}(\mathbb{R}, H_{ \mathit{\boldsymbol{q}}})} + \\ & \qquad\qquad\qquad\qquad\qquad \lVert \hat{A}(\bar{ \mathit{\boldsymbol{q}}})^j \rVert_{\mathscr{L}(H_{ \mathit{\boldsymbol{q}}})} \lVert \hat{B}( \mathit{\boldsymbol{q}}) - \hat{B}(\bar{ \mathit{\boldsymbol{q}}}) \rVert_{\mathscr{L}(\mathbb{R}, H_{ \mathit{\boldsymbol{q}}})} \Big] \\ & \leq C_1 \left[ \lVert \hat{A}( \mathit{\boldsymbol{q}})^j - \hat{A}(\bar{ \mathit{\boldsymbol{q}}})^j \rVert_{\mathscr{L}(H_{ \mathit{\boldsymbol{q}}})} + \lVert \hat{B}( \mathit{\boldsymbol{q}}) - \hat{B}(\bar{ \mathit{\boldsymbol{q}}}) \rVert_{\mathscr{L}(\mathbb{R}, H_{ \mathit{\boldsymbol{q}}})} \right] \\ &\leq C_1 (\tilde{C}_{\hat{A}} + \tilde{C}_{\hat{B}}) \lVert \mathit{\boldsymbol{q}}_1 - \mathit{\boldsymbol{q}}_2 \rVert_1 \end{aligned} \end{equation} (3.5)

    for C_1 the max of the operator norms for \hat{A}(\mathit{\boldsymbol{q}}) , \hat{B}(\mathit{\boldsymbol{q}}) , \hat{C} over Q , and \tilde{C}_{\hat{A}} , \tilde{B}_{\hat{A}} the max Lipschitz constants of \hat{A}(\mathit{\boldsymbol{q}}) and \hat{B}(\mathit{\boldsymbol{q}}) over all k and Q . The final inequality above follows from eq. (3.4) by noticing that for all \hat{\varphi} \in H_{ \mathit{\boldsymbol{q}}} , \lVert \hat{A}(\mathit{\boldsymbol{q}})\hat{\varphi} - \hat{A}(\bar{ \mathit{\boldsymbol{q}}})\hat{\varphi} \rVert_{H_{ \mathit{\boldsymbol{q}}}} \leq C_V \lVert \hat{A}(\mathit{\boldsymbol{q}})\hat{\varphi} - \hat{A}(\bar{ \mathit{\boldsymbol{q}}})\hat{\varphi} \rVert_{V} and that by identification the supremum over V^{\ast} is larger than that over H_{ \mathit{\boldsymbol{q}}} . For \hat{A}^n(\mathit{\boldsymbol{q}}) and \hat{B}^n(\mathit{\boldsymbol{q}}) as in eq. (2.9) by a repetition of the above arguments we maintain differentiability in \mathit{\boldsymbol{q}} , and thus h^n and y^n as in eq. (2.12) are differentiable and Lipschitz in \mathit{\boldsymbol{q}} .

    Lastly, for all i = 1, \ldots, R , j = 1, \ldots, K we have that for \mathit{\boldsymbol{q}}, \bar{ \mathit{\boldsymbol{q}}} \in Q

    \begin{align} \vert y_j^i( \mathit{\boldsymbol{q}}) - y_j^i(\bar{ \mathit{\boldsymbol{q}}}) \vert \leq \sum\limits_{\ell = 0}^{j - 1} \vert h_{j - \ell - 1}( \mathit{\boldsymbol{q}}) - h_{j - \ell - 1}(\bar{ \mathit{\boldsymbol{q}}}) \vert u_j^i &\leq \tilde{M}^{(K+1)} (K+1) \lVert \mathit{\boldsymbol{q}} - \bar{ \mathit{\boldsymbol{q}}} \rVert_1 \end{align} (3.6)

    where \{u_j^i\} are BrAC values bounded by definition to be in [0, 1] , \tilde{M} is the Lipschitz constant from eq. (3.5), and K+1 is the fixed upper bound on the number of temporal observations, j . Hence, the Lipschitz constant for y_j^i is independent of (i, j) . By noticing that the previous statement holds for y^{i, n}_j with a repetition of the work leading to eq. (3.6), our lemma has been proved.

    A direct consequence of Lemma 3.1 is that for all i = 1, 2, \ldots, R with \mathcal{K}_i(\bar{ \mathit{\boldsymbol{q}}}, \mathit{\boldsymbol{q}}) and \Lambda_i(\bar{ \mathit{\boldsymbol{q}}}, \mathit{\boldsymbol{q}}) as in the statement of Theorem 3.3, we have

    \begin{equation} \begin{aligned} \vert \mathcal{K}_i(\bar{ \mathit{\boldsymbol{q}}}, \mathit{\boldsymbol{q}}) \vert & = \left\vert \mathbb{E}_{\bar{ \mathit{\boldsymbol{q}}}}\left[\Lambda_i(\bar{ \mathit{\boldsymbol{q}}}, \mathit{\boldsymbol{q}}) \right] \right\vert \leq \sum\limits_{j = 0}^K \left\vert \dfrac{1}{2\sigma^2} \left[ y_j^i(\bar{ \mathit{\boldsymbol{q}}})^2 - y_j^i( \mathit{\boldsymbol{q}}) + 2y_j^i(\bar{ \mathit{\boldsymbol{q}}})(y_j^i( \mathit{\boldsymbol{q}}) - y_j^i(\bar{ \mathit{\boldsymbol{q}}})) \right] \right\vert \\ & = \dfrac{1}{2\sigma^2}\sum\limits_{j = 0}^K \left\vert (y_j^i(\bar{ \mathit{\boldsymbol{q}}}) - y_j^i( \mathit{\boldsymbol{q}}))^2 \right\vert \leq \dfrac{K\bar{M}^2}{2\sigma^2}\lVert \bar{ \mathit{\boldsymbol{q}}} - \mathit{\boldsymbol{q}} \rVert_1 = \tilde{\ell} \lVert \bar{ \mathit{\boldsymbol{q}}} - \mathit{\boldsymbol{q}} \rVert_1 \end{aligned} \end{equation} (3.7)

    where \sigma > 0 is the standard deviation of the N(0, \sigma^2) noise density, and \bar{M} is the Lipschitz constant from eq. (3.6) that is independent of i (and j ). Thus, for any \delta^{\ast} > 0 , i \in \{1, \ldots, R\} , and \bar{ \mathit{\boldsymbol{q}}}, \mathit{\boldsymbol{q}} \in \{ \mathit{\boldsymbol{q}} \in Q : \lVert \mathit{\boldsymbol{q}}_0 - \mathit{\boldsymbol{q}} \rVert_1 < \delta^{\ast} \} , we have that \lVert f_{i, \bar{ \mathit{\boldsymbol{q}}}} - f_{i, \mathit{\boldsymbol{q}}} \rVert_{L_1} \leq (2 \vert \mathcal{K}_i(\bar{ \mathit{\boldsymbol{q}}}, \mathit{\boldsymbol{q}}) \vert)^{\frac{1}{2}} \leq (2 \tilde{\ell} \lVert \bar{ \mathit{\boldsymbol{q}}} - \mathit{\boldsymbol{q}} \rVert_1)^{\frac{1}{2}} < 2(\tilde{\ell}\delta^{\ast})^{\frac{1}{2}} by the relationship between total variation and Kullback-Leibler distances, for \tilde{\ell} as in eq. (3.7). If we let \mathit{\boldsymbol{q}}^{\ast} \in Q be such that \lVert f_{i, \mathit{\boldsymbol{q}}_0} - f_{i, \mathit{\boldsymbol{q}}^{\ast}} \rVert_{L_1} > \delta^{\ast} and consider the set \mathcal{G} = \{ \mathit{\boldsymbol{q}} \in Q : \lVert \mathit{\boldsymbol{q}}^{\ast} - \mathit{\boldsymbol{q}} \rVert_1 < (\delta^{\ast})^2/(16\tilde{\ell}) \} , then \mathcal{G} is strongly separated from \mathit{\boldsymbol{q}}_0 (see Definition 3.1). This follows from the relationship between Affinity and total variation distance (via the Hellinger distance) as well as by noticing that for any density \nu on \mathcal{G} , the marginal density of \mathit{\boldsymbol{V}}^1 satisfies \lVert f_{i, \mathit{\boldsymbol{q}}^{\ast}} - \mathcal{v}_{\nu}^i(\mathit{\boldsymbol{V}}^i) \rVert_{L_1} \leq \delta^{\ast}/2 (for full example, see [32] or Example 3.5 in [42]). If this holds for all i then \mathcal{G} and \mathit{\boldsymbol{q}}_0 are strongly \bar{\delta} separated with \bar{\delta} independent of i .

    With this example in mind we note that items 1 and 2 of Theorem 3.2 are satisfied if the following special condition is met:

    1. For every \delta^{\ast} > 0 , there exist sets A_1, A_2, \ldots with L_1 diameter less than \delta^{\ast} , \text{diam}(A_i) < \delta^{\ast} , \bigcup\limits_{i \geq 1}A_i = Q , and \sum\limits_{i \geq 1} \sqrt{\pi_0(A_i)} < \infty for the mappings \mathit{\boldsymbol{q}} \mapsto f_{i, \mathit{\boldsymbol{q}}}

    where \pi_0 is the prior over Q . This follows from the fact that if special item 1 holds then we may take an \varepsilon^{\ast} -neighborhood of \mathit{\boldsymbol{q}}_0 , U = \{ \mathit{\boldsymbol{q}} \in Q : \lVert f_{i, \mathit{\boldsymbol{q}}_0} - f_{i, \mathit{\boldsymbol{q}}} \rVert_{L_1} < \varepsilon^{\ast} \quad \forall i\} . As discussed above, since \lVert f_{i, \bar{ \mathit{\boldsymbol{q}}}} - f_{i, \mathit{\boldsymbol{q}}} \rVert_{L_1} is independent of i , U is non-empty and contains the set \{ \mathit{\boldsymbol{q}} \in Q : \lVert \mathit{\boldsymbol{q}}_0 - \mathit{\boldsymbol{q}} \rVert_1 < (\varepsilon^{\ast})^2/(4\tilde{\ell}) \} . Now set \delta^{\ast} = (\varepsilon^{\ast})^2 / (16\tilde{\ell}) , and by compactness cover Q with a finite number of disjoint sets A_i determined by the balls \{ \mathit{\boldsymbol{q}} \in Q : \lVert \bar{ \mathit{\boldsymbol{q}}}_i - \mathit{\boldsymbol{q}} \rVert_1 < \delta^{\ast} \} with model \mathit{\boldsymbol{q}} \mapsto f_{i, \mathit{\boldsymbol{q}}} , where \{\bar{ \mathit{\boldsymbol{q}}}_i\}_{i = 1}^{\gamma} represents a finite set of points in Q chosen so that \bigcup\limits_{i \geq 1}A_i = Q . From these A_i 's we have that the finite subset that intersect with U^{ \mathit{c}} must cover U^{ \mathit{c}} . This finite subset of A_i 's subsequently satisfies the assumptions of Theorem 3.2. Specifically, the strong separation condition is satisfied as per the discussion leading up to special item 1 by noticing that on each A_i we have \lVert f_{i, \mathit{\boldsymbol{q}}_0} - f_{i, \bar{ \mathit{\boldsymbol{q}}}_i} \rVert_1 > \varepsilon^{\ast} , and the convergent sum condition is satisfied by the fact that the A_i 's can be considered (made) mutually exclusive with union contained in Q . We now state and prove our main theorem.

    Theorem 3.5. For Q a rectangle in the interior of the positive orthant of \mathbb{R}^2 , a prior \pi_0 with compact support Q and a density that is continuous on Q , i.i.d noise distributed as N(0, \sigma^2) for \sigma > 0 , data \{ \mathit{\boldsymbol{V}}^i \}_{i = 1}^{R} drawn from independent but not identically distributed distributions generated by f_{i, \mathit{\boldsymbol{q}}} as in eq. (3.2), Hilbert spaces H_q and V as in eq. (2.4), bilinear form a(\mathit{\boldsymbol{q}}, \cdot, \cdot): V \times V \rightarrow \mathbb{R} as in Section 2.1 with q_3 and q_4 assumed known, induced infinitesimal generator A(\mathit{\boldsymbol{q}}) as in Section 2.1, and true parameter \mathit{\boldsymbol{q}}_0 \in Q , we have that our posterior \pi(\cdot \vert \{ \mathit{\boldsymbol{V}}^i\}) as in eq. (3.2) is consistent for \mathit{\boldsymbol{q}}_0 as R \rightarrow \infty .

    Proof. For any set A \subset Q with \mathit{\boldsymbol{q}}_0 \notin A we will use the form of \pi(A \vert \{ \mathit{\boldsymbol{V}}^i\}) as in eq. (3.2) and handle the numerator, J_A and denominator, J separately.

    First, as Q is compact, for any \delta > 0 we may cover Q by a finite number of sets A_i , i = 1, 2, \ldots, \gamma where each A_i is a subset of an L_1 ball in Q . That is, for every i and \mathit{\boldsymbol{q}}, \bar{ \mathit{\boldsymbol{q}}} \in A_i we have that \lVert \mathit{\boldsymbol{q}} - \bar{ \mathit{\boldsymbol{q}}} \rVert_1 < \delta . For R large enough, if on each A_i we consider the model \mathit{\boldsymbol{q}} \mapsto f_{i, \mathit{\boldsymbol{q}}} for i \in \{ 1, \ldots, R \} and f_{i, \mathit{\boldsymbol{q}}} the density of the random variable \mathit{\boldsymbol{\mathcal{V}}}^i with \mathit{\boldsymbol{q}} assumed known, then special item 1 is satisfied for prior \pi_0 . Hence, by Theorem 3.2 we have that for some \beta_0 > 0 , e^{R\beta_0} J_A(\{ \mathit{\boldsymbol{V}}^i \}) \rightarrow 0 a.s. P_{ \mathit{\boldsymbol{q}}_0}^{\infty} as R\rightarrow \infty .

    Now for \Lambda_i , \mathcal{K}_i , and S_i as in the statement of Theorem 3.3, for i = 1, 2, \ldots, R and \mathit{\boldsymbol{q}} \in Q , we have \vert \mathcal{K}_i(\mathit{\boldsymbol{q}}_0, \mathit{\boldsymbol{q}}) \vert \leq \tilde{\ell} \lVert \mathit{\boldsymbol{q}}_0 - \mathit{\boldsymbol{q}} \rVert_1 and

    \begin{align} S_i( \mathit{\boldsymbol{q}}_0, \mathit{\boldsymbol{q}}) & = \sum\limits_{j = 0}^K \mathrm{Var}_{ \mathit{\boldsymbol{q}}_0}\left[ \dfrac{1}{2\sigma^2}\left( y_j^i( \mathit{\boldsymbol{q}}_0)^2 - y_j^i( \mathit{\boldsymbol{q}})^2 + 2 \mathcal{V}_k(y_j^i( \mathit{\boldsymbol{q}}_0) - y_j^i( \mathit{\boldsymbol{q}})) \right) \right] \end{align}
    \begin{align} & = \sum\limits_{j = 0}^K \dfrac{1}{4\sigma^4} \mathrm{Var}_{ \mathit{\boldsymbol{q}}_0}\left[ 2 \mathcal{V}_j^i(y_j^i( \mathit{\boldsymbol{q}}_0) - y_j^i( \mathit{\boldsymbol{q}})) \right] = \sum\limits_{j = 0}^K \dfrac{4\sigma^2}{4\sigma^4} (y_j^i( \mathit{\boldsymbol{q}}) - y_j^i( \mathit{\boldsymbol{q}}_0))^2 \end{align}
    \begin{align} &\leq \dfrac{\tilde{\ell}^2}{4} \lVert \mathit{\boldsymbol{q}}_0 - \mathit{\boldsymbol{q}} \rVert^2_1 \end{align}

    for \tilde{\ell} as determined by eq. (3.7).

    Thus, for \mathit{\boldsymbol{q}}_0, \mathit{\boldsymbol{q}} \in Q we find that \sum\limits_{i \geq 1} \dfrac{S_i(\mathit{\boldsymbol{q}}_0, \mathit{\boldsymbol{q}})}{i^2} \leq \left(\dfrac{\tilde{\ell}^2}{4} \lVert \mathit{\boldsymbol{q}}_0 - \mathit{\boldsymbol{q}} \rVert^2_1 \right) \sum\limits_{i \geq 1} \dfrac{1}{i^2} < \infty . Further, by the bounds above we have that for every \varepsilon > 0 and i , \{ \mathit{\boldsymbol{q}} : \vert \mathcal{K}_i(\mathit{\boldsymbol{q}}_0, \mathit{\boldsymbol{q}})\vert < \varepsilon \} is non-empty and our choice in such \mathit{\boldsymbol{q}} does not depend on i . Hence the set \{ \mathit{\boldsymbol{q}} : \vert \mathcal{K}_i(\mathit{\boldsymbol{q}}_0, \mathit{\boldsymbol{q}})\vert < \varepsilon \quad \forall i\} is non-empty. Thus, for B = Q we satisfy the assumptions of Theorem 3.3 and therefore find that \forall \beta > 0 , e^{R\beta}J(\{ \mathit{\boldsymbol{V}}^i \}) \rightarrow \infty a.s. P_{ \mathit{\boldsymbol{q}}_0}^{\infty} as k\rightarrow \infty .

    So for any set A \subset Q with \mathit{\boldsymbol{q}}_0 \notin A , from eq. (3.2) we have that \pi(A \vert \{ \mathit{\boldsymbol{V}}^i\}) \rightarrow 0 a.s. P_{ \mathit{\boldsymbol{q}}_0}^{\infty} as R\rightarrow \infty and thus the theorem has been proved.

    From Lemma 3.1 we find that we maintain the differentiability and Lipschitz properties of the finite-dimensional semigroup as in eq. (2.9) and respective kernel as in eq. (2.12). Thus, with a straightforward rewriting of eq. (3.2) and eq. (3.2) in terms of the finite-dimensional posterior eq. (2.20), and repetition of the work following Lemma 3.1 through the proof of Theorem 3.5 we have the following corollary.

    Corollary 3.3. Under the same hypotheses as Theorem 3.5, for fixed positive integer n we have that our finite-dimensional posterior \pi^n(\cdot \vert \{ \mathit{\boldsymbol{V}}^i\}) as in eq. (2.20) is consistent at \mathit{\boldsymbol{q}}_0 as R \rightarrow \infty .

    In this section we consider the problem of using the biosensor measured TAC signal to estimate BrAC. We do this by deconvolving it; to wit we invert the convolution given in eq. (2.12) subject to a positivity constraint and regularization to mitigate the inherent ill-posedness of the inversion. Recall that the convolution given in eq. (2.12) was found by solving the finite-dimensional discrete time system eq. (2.10) derived from eq. (2.2). We employ the method originally described in [25], wherein the problem is formulated as a constrained, regularized, optimization problem (see, for example, [48]).

    We first briefly summarize the treatment in [25] and then follow by showing how our work is able to make direct use of this theory. Let \tilde{V} and \tilde{H} be Hilbert spaces forming a Gelfand Triple, \tilde{V} \hookrightarrow \tilde{H} \hookrightarrow \tilde{V}^{\ast} . For an admissible set Q , a compact subset of the positive orthant of \mathbb{R}^{2} , with \mathit{\boldsymbol{q}} \in Q , let A(\mathit{\boldsymbol{q}}) be an abstract parabolic operator defined by a sesquilinear form a(\mathit{\boldsymbol{q}}, \cdot, \cdot): V\times V \rightarrow \mathbb{R} (i.e., one that satisfies items 1 to 3 in Section 2.1) that when restricted to \{ \varphi \in \tilde{V} : A(\mathit{\boldsymbol{q}})\varphi \in \tilde{H} \} generates a holomorphic semigroup on \tilde{H} , \{ e^{A(\mathit{\boldsymbol{q}})t} : t\geq 0\} . For bounded operators B(\mathit{\boldsymbol{q}}) \in \mathscr{L}(\mathbb{R}, \tilde{H}) and C(\mathit{\boldsymbol{q}}) \in \mathscr{L}(\tilde{H}, \mathbb{R}) consider the input/output system where \dot{x}(t) = A(\mathit{\boldsymbol{q}})x(t) + B(\mathit{\boldsymbol{q}})u(t) , x(0) = x_0 \in \tilde{H} , and y(t) = C(\mathit{\boldsymbol{q}})x(t) where on the interval [0, T] , u\in L_2(0, T) is the input, y the output, and x is the state variable.

    For sampling interval \tau > 0 and zero-order hold input u(t) = u_j , t \in [j\tau, (j+1)\tau) , j = 0, 1, 2, ... the corresponding sampled-time system is given by

    \begin{align} \begin{aligned} x_{j+1} & = \hat{A}( \mathit{\boldsymbol{q}})x_j + \hat{B}( \mathit{\boldsymbol{q}})u_j \quad x_0 \in \tilde{H}, \\ y_j & = \hat{C}( \mathit{\boldsymbol{q}})x_j \end{aligned} \end{align} (3.8)

    where \hat{A}(\mathit{\boldsymbol{q}}) = e^{A(\mathit{\boldsymbol{q}})\tau} \in \mathscr{L}(\tilde{H}, \tilde{H}) , \hat{B}(\mathit{\boldsymbol{q}}) = \int_0^{\tau} e^{A(\mathit{\boldsymbol{q}})s}B(\mathit{\boldsymbol{q}}) ds \in \mathscr{L}(\mathbb{R}, \tilde{H}) , and \hat{C}(\mathit{\boldsymbol{q}}) = C(\mathit{\boldsymbol{q}}) \in \mathscr{L}(\tilde{H}, \mathbb{R}) , x_j = x(j\tau) \in \tilde{H} , y_j = y(j\tau) \in \mathbb{R} j = 0, 1, 2, \ldots , and for all j , \{u_j\} \subset \mathbb{R} are zero-order hold input values.

    Now let \mathit{\boldsymbol{\mathcal{q}}} be a random variable with support the parameter space Q . For \tilde{\pi} the probability measure of \mathit{\boldsymbol{\mathcal{q}}} , define the Bochner spaces \mathcal{V} = L^2_{\tilde{\pi}}(Q; \tilde{V}) , \mathcal{H} = L^2_{\tilde{\pi}}(Q; \tilde{H}) , and \mathcal{U} = L^2_{\tilde{\pi}}(Q; \mathbb{R}) . It is easily shown that the spaces \mathcal{V} and \mathcal{H} form a Gelfand triple \mathcal{V} \hookrightarrow \mathcal{H} \hookrightarrow \mathcal{V}^{\ast} . Define \mathcal{a}(\cdot, \cdot): \mathcal{V} \times \mathcal{V}\rightarrow \mathbb{R} by \mathcal{a}(\varphi, \psi) = \mathbb{E}_{\tilde{\pi}}[a(\mathit{\boldsymbol{\mathcal{q}}}, \varphi(\mathit{\boldsymbol{\mathcal{q}}}), \psi(\mathit{\boldsymbol{\mathcal{q}}}))] = \int_{Q} a(\mathit{\boldsymbol{q}}, \varphi(\mathit{\boldsymbol{q}}), \psi(\mathit{\boldsymbol{q}})) d\tilde{\pi}(\mathit{\boldsymbol{q}}) for \varphi, \psi \in \mathcal{V} . Then, as in Section 2.1, the form -\mathcal{a}(\cdot, \cdot) satisfies items 1 to 3 and therefore defines a linear map \mathcal{A} that when restricted to \{ \varphi \in \mathcal{V} : \mathcal{A}\varphi \in \mathcal{H} \} , generates an analytic semigroup on \mathcal{H} , \{ e^{\mathcal{A}t} : t\geq 0\} .

    Assume further that the map \mathit{\boldsymbol{q}} \mapsto B(\mathit{\boldsymbol{q}}) is in L_\infty(Q, \mathscr{L}(\mathbb{R}, \tilde{H})) and that the map \mathit{\boldsymbol{q}} \mapsto C(\mathit{\boldsymbol{q}}) is in L_2(Q, \mathscr{L}(\tilde{H}, \mathbb{R})) with respect to the measure \tilde{\pi} . (Note that since the domain space in \mathscr{L}(\mathbb{R}, \tilde{H}) and the co-domain space in \mathscr{L}(\tilde{H}, \mathbb{R}) are both \mathbb{R} , it follows that in fact the mapping \mathit{\boldsymbol{q}} \mapsto B(\mathit{\boldsymbol{q}}) \in L_\infty(Q, \tilde{H}) , and by the Riesz Representation Theorem, that effectively the mapping \mathit{\boldsymbol{q}} \mapsto C(\mathit{\boldsymbol{q}}) \in L_2(Q, \tilde{H}) = \mathcal{H} ). Then define bounded linear operators \mathcal{B} \in \mathscr{L}(\mathcal{U}, \mathcal{H}) and \mathcal{C} \in \mathscr{L}(\mathcal{H}, \mathbb{R}) by \langle \mathcal{B}\mathcal{u}, \psi \rangle_{\mathcal{H}} = \mathbb{E}_{\tilde{\pi}}[\langle B(\mathit{\boldsymbol{\mathcal{q}}})\mathcal{u}(\mathit{\boldsymbol{\mathcal{q}}}), \psi(\mathit{\boldsymbol{\mathcal{q}}}) \rangle_{\tilde{H}}] = \int_{Q} \langle B(\mathit{\boldsymbol{q}})\mathcal{u}(\mathit{\boldsymbol{q}}), \psi(\mathit{\boldsymbol{q}}) \rangle_{\tilde{H}} d\tilde{\pi}(\mathit{\boldsymbol{q}}) and \mathcal{C}\psi = \mathbb{E}_{\tilde{\pi}}[C(\mathit{\boldsymbol{\mathcal{q}}})\psi(\mathit{\boldsymbol{\mathcal{q}}})] = \int_{Q} C(\mathit{\boldsymbol{q}})\psi(\mathit{\boldsymbol{q}}) d\tilde{\pi}(\mathit{\boldsymbol{q}}) , respectively, for \mathcal{u} \in \mathcal{U} and \psi \in \mathcal{H} . It can then be shown [49,50] that for \mathcal{u} \in L_2([0, T], \mathcal{U}) the solution to the input/output system from above with u(t) = u(t, \mathit{\boldsymbol{q}}) = \mathcal{u}(t) agrees with the solution to the system where \dot{ \mathcal{x}}(t) = \mathcal{A} \mathcal{x}(t) + \mathcal{B}\mathcal{u}(t) , \mathcal{x}(0) = \mathcal{x}_0 \in \mathcal{H} , and \mathcal{y}(t) = \mathcal{C} \mathcal{x}(t) for \tilde{\pi}- almost every \mathit{\boldsymbol{q}} \in Q . Then with sampling interval \tau > 0 as in eq. (3.8), and zero-order hold input \mathcal{u}(t) = \mathcal{u}_j , t \in [j\tau, (j+1)\tau) , \mathcal{u}_j \in \mathcal{U} , j = 0, 1, 2, ... , this system becomes

    \begin{equation} \begin{aligned} \mathcal{x}_{j+1} & = \hat{\mathcal{A}} \mathcal{x}_j + \hat{\mathcal{B}}\mathcal{u}_j \quad \mathcal{x}_0 \in \mathcal{H} \\ \mathcal{y}_j & = \hat{\mathcal{C}} \mathcal{x}_j \end{aligned} \end{equation} (3.9)

    for j = 0, 1, 2, \ldots where \hat{\mathcal{A}} = e^{\mathcal{A}\tau} \in \mathscr{L}(\mathcal{H}, \mathcal{H}) , \hat{\mathcal{B}} = \int_0^{\tau} e^{\mathcal{A}s} \mathcal{B} ds \in \mathscr{L}(\mathcal{U}, \mathcal{H}) , and \hat{\mathcal{C}} \in \mathscr{L}(\mathcal{H}, \mathbb{R}) .

    Now, with eq. (3.9), note that \{ \mathcal{u}_j\} \subset \mathcal{U} is obtained by zero-order hold sampling a continuous time signal. That is, the input to eq. (3.9) is \mathcal{u}(j\tau) = \mathcal{u}_j \in \mathcal{U} with \mathcal{u} at least continuous on [0, T] . We seek an estimate for the input based on this model, wherein the input estimate \mathcal{u} is a function of both time and the random parameters \mathit{\boldsymbol{\mathcal{q}}} . For optimization purposes (more precisely, to be able to include regularization) we require additional regularity. Given the time interval [0, T] , let \mathcal{u} \in S(0, T) = H^1(0, T; \mathcal{U}) and let \mathcal{U} be a compact subset of S(0, T) .

    The input estimation or deconvolution problem is then given by

    \begin{align} \mathcal{u}^* = \mathop {\arg \min }\limits_{\mathcal{U}} J( \mathcal{u}) = \mathop {\arg \min }\limits_{\mathcal{U}} \sum\limits_{k = 1}^K \vert \mathcal{y}_k( \mathcal{u}) - \hat{y}_k \vert^2 + \lVert \mathcal{u} \rVert^2_{S(0, T)} \end{align} (3.10)

    where \lVert \cdot \rVert^2_{S(0, T)} is a norm on S(0, T) that will be defined below, \{\hat{y}_k\} are measured output values, the term \lVert \mathcal{u} \rVert^2_{S(0, T)} serves as regularization, and \mathcal{y}_k(\mathcal{u}) = \sum\limits_{j = 0}^{k-1}\langle h_{k-j-1}, \mathcal{u}_j\rangle_{\mathcal{U}}, for k = 1, 2, \ldots, K with \mathcal{u}_j = \mathcal{u}(j\tau) for j = 1, 2, \ldots, K the zero-order hold input to the discrete time system eq. (3.9), and convolution filter h_{\ell} = \hat{\mathcal{C}}\hat{\mathcal{A}}^{\ell -1} \hat{\mathcal{B}} \in \mathscr{L}(\mathcal{U}, \mathbb{R}) = \mathcal{U}^\ast (which is equal to \mathcal{U} , by the Riesz Representation Theorem) where \hat{\mathcal{C}} \in \mathscr{L}(\mathcal{H}, \mathbb{R}) , \hat{\mathcal{A}} \in \mathscr{L}(\mathcal{H}, \mathcal{H}) , and \hat{\mathcal{B}} = \int_{Q} e^{\mathcal{A}s} \mathcal{B} ds \in \mathscr{L}(\mathcal{U}, \mathcal{H}) .

    Solving eq. (3.10) requires finite dimensional approximations. For index M , let \mathcal{U}^M define an approximating family of closed subsets of \mathcal{U} , where each subset is contained within a corresponding finite dimensional subspace, \mathcal{S}^M of S(0, T) . Further we require that for each \mathcal{u} \in \mathcal{U} there exists a sequence \{ \mathcal{u}^M \} with \mathcal{u}^M \in \mathcal{U}^M such that \mathcal{u}^M \rightarrow \mathcal{u} in S(0, T) as M\rightarrow \infty . For index N , let \mathcal{V}^N be an element of an approximating family of finite-dimensional subspaces of \mathcal{V} , and let \mathscr{P}^N_{\mathcal{H}}: \mathcal{H} \rightarrow \mathcal{V}^N be the orthogonal projection of \mathcal{H} onto \mathcal{V}^N . We also require of the spaces \mathcal{V}^N that for each \mathcal{v}\in\mathcal{V} , \mathscr{P}^N_{\mathcal{H}}\mathcal{v} \rightarrow \mathcal{v} in \mathcal{V} as N\rightarrow \infty .

    We next specify finite-dimensional operators \hat{\mathcal{A}}^N \in \mathscr{L}(\mathcal{V}^N, \mathcal{V}^N) , \hat{\mathcal{B}}^N \in \mathscr{L}(\mathcal{U}, \mathcal{V}^N) , and \hat{\mathcal{C}}^N \in \mathscr{L}(\mathcal{V}^N, \mathbb{R}) that define the finite-dimensional system analogous to eq. (3.9). That is, let \mathcal{A}^N: \mathcal{V}^N \rightarrow \mathcal{V}^N be given by \langle \mathcal{A}^N\varphi^N, \psi^N \rangle_{\mathcal{H}} = -\mathcal{a}(\varphi^N, \psi^N) for \varphi^N, \psi^N \in \mathcal{V}^N , \hat{\mathcal{A}}^N = e^{\mathcal{A}^N\tau} , \mathcal{B}^N = \mathscr{P}^N_{\mathcal{H}} \mathcal{B} , \hat{\mathcal{B}}^N = \int_0^{\tau} e^{\mathcal{A}^Ns}\mathcal{B}^N ds , and \hat{\mathcal{C}}^N = \mathcal{C} . In this way we obtain a doubly-indexed sequence of approximating finite-dimensional optimization or deconvolution problems given by

    \begin{align} \mathcal{u}^*_L = \mathop {\arg \min }\limits_{ \mathcal{U}^M} J^L( \mathcal{u}) = \mathop {\arg \min }\limits_{ \mathcal{U}^M} \sum\limits_{k = 1}^K \vert \mathcal{y}^N_k( \mathcal{u}) - \hat{y}_k \vert^2 + \lVert \mathcal{u} \rVert^2_{S(0, T)} \end{align} (3.11)

    where L = (M, N) , h^N_{\ell} = \hat{\mathcal{C}}^N(\hat{\mathcal{A}}^N)^{\ell}\hat{\mathcal{B}}^N \in \mathcal{U} , and

    \begin{align} \mathcal{y}^N_k( \mathcal{u}) = \sum\limits_{j = 0}^{k-1}\langle h^N_{k-j-1}, \mathcal{u}_j\rangle_{\mathcal{U}}, \quad k = 1, 2, \ldots, K. \end{align} (3.12)

    Using the approximation properties of the subspaces \mathcal{V}^N and \mathcal{U}^M (that is, that for each \mathcal{u} \in \mathcal{U}^M there exists a sequence \{ \mathcal{u}^M \} with \mathcal{u}^M \in \mathcal{U}^M and \lVert \mathcal{u}^M - \mathcal{u} \rVert_{S(0, T)} \rightarrow 0 as M\rightarrow \infty , and that for each v \in \mathcal{V} , \lVert \mathscr{P}^N_{\mathcal{H}}v - v \rVert_{\mathcal{V}} \rightarrow 0 as N \rightarrow \infty ), and the corresponding operators \hat{\mathcal{A}}^N \in \mathscr{L}(\mathcal{V}^N, \mathcal{V}^N) , \hat{\mathcal{B}}^N \in \mathscr{L}(\mathcal{U}, \mathcal{V}^N) , and \hat{\mathcal{C}}^N \in \mathscr{L}(\mathcal{V}^N, \mathbb{R}) , it can be shown that 1) for each multi-index L , eq. (3.11) admits a solution \mathcal{u}_L^{\ast} , and 2) there exists a subsequence of \{ \mathcal{u}_L^{\ast} \} , \{ \mathcal{u}_{L_k}^{\ast} \} \subset \{ \mathcal{u}_L^{\ast} \} with \mathcal{u}_{L_k}^{\ast} \rightarrow \mathcal{u}^{\ast} strongly as k\rightarrow \infty with \mathcal{u}^{\ast} a solution of eq. (3.10). Further, if in addition \mathcal{U} is assumed to be a closed and convex subset of S(0, T) , for each M , \mathcal{U}^M is a closed and convex subset of \mathcal{U} , and the optimization problem given in eq. (3.10) admits a unique (with respect to sampling) solution, then the sequence of solutions to eq. (3.11), \{ \mathcal{u}_L^{\ast} \} converges strongly, or in S(0, T) to the unique solution of eq. (3.10), \mathcal{u}^{\ast} . For the proofs of these results see Section 5 of [25]

    To numerically carry out the requisite computations to actually determine \mathcal{u}_L^{\ast} for given values of M , N and L = (M, N) , we continue to apply the results in [25] while also connecting them to our treatment in Sections 2.1 and 2.2 above. We assume that the feasible parameter set Q is a compact rectangle in the positive orthant of \mathbb{R}^2 , we set \tilde{H} = H_{ \mathit{\boldsymbol{q}}} and \tilde{V} = V as in eq. (2.4), and we identify the operators in eq. (3.8) with those in eq. (2.6). Our distribution over \mathit{\boldsymbol{\mathcal{q}}} , \tilde{\pi} , is the finite-dimensional posterior \pi^n(\cdot \vert \{V_j^i\}) for fixed n as in eq. (2.20) and we proceed with the Bochner spaces \mathcal{V} = L^2_{\pi^n(\cdot \vert \{V_j^i\})}(Q; V) and \mathcal{H} = L^2_{\pi^n(\cdot \vert \{V_j^i\})}(Q; H_{ \mathit{\boldsymbol{q}}}) to achieve eq. (3.9).

    For the state variables \mathcal{x}_j(\eta, \mathit{\boldsymbol{q}}) we have that \eta \in [0, 1] and \mathit{\boldsymbol{q}} \in Q = [a_1, b_1] \times [a_2, b_2] for 0 < a_i < b_i , i = 1, 2 . Further, for the inputs \mathcal{u}(t, \mathit{\boldsymbol{q}}) we have that t \in [0, T] and \mathit{\boldsymbol{q}} \in Q . Let n be as in eq. (2.8) and m a positive integer, and we discretize [0, 1] and [0, T] using the sets of linear B-splines, \{ \psi_j^{n} \}_{j = 0}^{n} and \{ \zeta_j^{m} \}_{j = 0}^{m} , respectively, on the uniform meshes, \{ \frac{j}{n} \}_{j = 0}^{n} and \{ \frac{i T}{m} \}_{j = 0}^{m} , respectively. Further, for positive integers m_1 and m_2 , we discretize Q with the 0^{\text{th}} -order B-splines \{ \chi_{i, j}^{m_i} \}_{j = 1}^{m_i} , i = 1, 2 defined with respect to the uniform grids \{ a_i - \frac{(b_i - a_i)j}{m_i} \}_{j = 0}^{m_i} , on [a_i, b_i] , i = 1, 2 .

    Then for multi-indices N = (n, m_1, m_2) and M = (m, m_1, m_2) we define the approximating subspaces \mathcal{V}^N and \mathcal{S}^M as follows using tensor products. That is, let \mathcal{V}^N = \text{span}\{ \hat{\psi}_i^N \}_{i = 1}^{^\mathcal{N}} = \text{span}\{(\psi^{n}_j(0) \chi^{m_1}_{1, j_1} \chi^{m_{2}}_{2, j_2}, \psi^{n}_j \chi^{m_1}_{1, j_1} \chi^{m_{2}}_{2, j_{2}}) \}_{j = 0, j_1 = 1, j_{2} = 1}^{n, m_1, m_{2}} and \mathcal{S}^M = \text{span}\{(\zeta^{m}_j \chi^{m_1}_{1, j_1} \chi^{m_{2}}_{2, j_2}\}_{j = 0, j_1 = 1, j_{2} = 1}^{m, m_1, m_{2}} = \text{span}\{ \hat{\zeta}_i^M \}_{i = 1}^\mathcal{M} , where \mathcal{N} = (n+1)m_1m_2 and \mathcal{M} = (m+1)m_1m_2 . Standard approximation theoretic arguments (see, for example, [33]) can be used to argue that the subspaces defined in above satisfy the required approximation assumptions on \mathcal{V^N} and \mathcal{S}^M . Then \mathcal{x}^N \in \mathcal{V^N} and \mathcal{u}^M \in \mathcal{S}^M , can be written as \mathcal{x}^N(\eta; \mathit{\boldsymbol{q}}) = \sum\limits_{i = 0, i_1 = 1, i_{2} = 1}^{n, m_1, m_{2}} \mathcal{x}^N_{i, i_1, i_{2}} \psi_i^{n}(\eta) \chi^{m_1}_{1, i_1}(q_1) \chi^{m_2}_{2, i_2}(q_2) and \mathcal{u}^M(t; \mathit{\boldsymbol{q}}) = \sum\limits_{i = 0, i_1 = 1, i_{2} = 1}^{m, m_1, m_{2}} \mathcal{u}^M_{i, i_1, i_{2}} \zeta_i^{m}(t) \chi^{m_1}_{1, i_1}(q_1) \chi^{m_2}_{2, i_2}(q_2) , respectively.

    Then with the bases for \mathcal{V}^N and \mathcal{S}^M as chosen above, it is an elementary exercise to determine the matrix representations for the operators \mathcal{A}^N , \mathcal{B}^N , \mathcal{C}^N , \hat{\mathcal{A}}^N , \hat{\mathcal{B}}^N , and \hat{\mathcal{C}}^N . It then follows that eq. (3.9) takes a matrix system where for k = 1, 2, \ldots, K , \mathbb{M}^N\mathbb{X}^N_{k+1} = \mathbb{K}^N\mathbb{X}^N_{k} + \mathbb{B}^L\mathbb{U}^N_{k} and \mathcal{y}^L_k = \mathbb{C}^N\mathbb{X}^N_k with \mathbb{X}^N_k \in \mathbb{R}^\mathcal{N} the coefficients of the basis elements \{ \hat{\psi}^N_i \} , \mathbb{U}^N_k \in \mathbb{R}^\mathcal{M} the coefficients of the basis elements \{ \hat{\zeta}^M_i\} as in the previously mentioned approximating subspaces, \mathbb{M}^N \in \mathbb{R}^{\mathcal{N}\times \mathcal{N}} a matrix with entries [\mathbb{M}^N]_{i, j} = \langle \hat{\psi}_i, \hat{\psi}_j\rangle_{\mathcal{H}} , \mathbb{K}^N \in \mathbb{R}^{\mathcal{N}\times \mathcal{N}} a matrix with entries [\mathbb{K}^N]_{i, j} = \langle \mathcal{A}^N\hat{\psi}^N_i, \hat{\psi}^N_j\rangle_{\mathcal{H}} , \mathbb{B}^L \in \mathbb{R}^{\mathcal{N}\times \mathcal{M}} a matrix with entries \langle \mathcal{B}^N \hat{\zeta}^{M}_i, \hat{\psi}^N_j \rangle_{\mathcal{H}} , and \mathbb{C}^N \in \mathbb{R}^{1\times \mathcal{M}} given by [1, 0, \ldots, 0] . From here the matrix representation of h_{k}^L (with L = (M, N) in place of N due to the joint dependence on the multi-indices M and N ) can be found using this matrix system.

    We note that the optimization problem eq. (3.11) is a constrained problem, in that \mathbb{U}^N_k of the previously stated matrix system are to be non-negative. With a proper placement of \{ h_k^L \} into the block matrix \mathbb{H}^L , the approximating deconvolution problem eq. (3.11) is now given by u_{L}^{*} = \underset{\mathscr{U}^{M}}{\arg \min } J^{L}(u) where we have that

    \begin{align} J^{L}(u;r_1, r_2) = \left.\left\|\left[\begin{array}{c} \mathbb{H}^{L} \\ \left(r_{1} \mathbb{Q}_{1}^{M}+r_{2} \mathbb{Q}_{2}^{M}\right)^{\frac{1}{2}} \end{array}\right] \mathbb{U}^{M}-\mathbb{Y}^{M}\right\|\right|^{2}_{\mathbb{R}^{K+K\mathcal{M}}}, \end{align} (3.13)

    where \mathbb{U}^M is the K\mathcal{M} dimensional column vector of the coefficients of \mathcal{u} \in \mathscr{U}^{M} , and \mathbb{Y}^M is the K + K\mathcal{M} column vector of measured output values \{ \hat{y}_k \} followed by K\mathcal{M} zeros. Further, \mathbb{Q}^M_i for i = 1, 2 are matrices with entries given by the \mathcal{U} inner products of the basis elements for the subspaces \mathcal{S}^M as determined by the regularization term \| \mathcal{u}\|_{S(0, T)}^{2} below. Note that the regularization term \| \mathcal{u}\|_{S(0, T)}^{2} is derived from a weighted inner product on S[0, T] and thus corresponds to a squared norm on S(0, T) given by \| \mathcal{u}\|_{S(0, T)}^{2} = r_{1} \int_{0}^{T}\| \mathcal{u}(t)\|_{\mathcal{U}}^{2} d t+r_{2} \int_{0}^{T}\|\dot{ \mathcal{u}}(t)\|_{\mathcal{U}}^{2} d t .

    The values of the regularization weights used in eq. (3.13), r_i = r^{\ast}_i > 0 for i = 1, 2 are chosen optimally. Indeed, in order to find (r_1^{\ast}, r_2^{\ast}) , BrAC-TAC input-output training data pairs, \{(u^i_{j}, V^i_{j})\}_{i = 0, j = 0}^{P, K} are used to optimize (r_1, r_2) via the following scheme:

    \begin{align} \left(r_{1}^{*}, r_{2}^{*}\right) = \underset{\left(r_{1}, r_{2}\right) \in \mathbb{R}^{+}\times \mathbb{R}^+}{\arg \min } \sum\limits_{i = 1}^{R} \sum\limits_{j = 1}^{K}\left(\left|\bar{u}_{L ; j-1}^{i, *}- u^i_{j-1}\right|^{2}+\left|\tilde{y}_{L ; j}^{i, *}- V^i_{j}\right|^{2}\right) \end{align} (3.14)

    where \bar{u}_{L; j}^{i, *} = \mathbb{E}_{\tilde{\pi}(\mathit{\boldsymbol{\theta}}^{\ast})}[\tilde{u}_{L; j}^{i, *}] , \tilde{u}_{L; j}^{i, *} are the predicted BrAC values found by finding the minimum of J^L(\ \cdot \; r_1, r_2) given by eq. (3.13) with r_1 and r_2 candidate values for the regularization weights from a specified feasible set in the positive orthant of \mathbb{R}^2 , \mathbb{R}^{+}\times \mathbb{R}^+ , and \tilde{y}_{L; j}^{i, *} are the TAC values found by using \tilde{u}_{L; j}^{i, *} as input to eq. (3.12).

    All of the data used in the studies detailed below, unless otherwise specifically stated (e.g., as in Section 3.4.2), were collected in USC IRB approved human subject experiments designed and run by researchers in the laboratory of one of the authors (S. E. L.) as part of a National Institutes of Health (NIH) funded investigation (see, [51]). These experiments were carried out in controlled environments wherein 40 participants completed one to four drinking episodes, with viable data recorded in 146 drinking episodes. BrAC was obtained using Alco-sensor IV breath analyzer devices from Intoximeters, Inc, St. Louis, MO, and participants each wore two SCRAM (Secure Continuous Remote Alcohol Monitoring) devices manufactured by Alcohol Monitoring Systems (AMS) in Littleton, Colorado (see Figure 1) simultaneously placed on the participants' left and right arms for TAC. For each separate SCRAM device, participants started their readings with a TAC and BrAC of 0.000, consumed alcohol (equivalent across all sessions per participant) in one of three different drinking patterns (single: over 15 minutes; dual: over two 15-min periods spaced 30-minutes apart; or steady: over 60 minutes), and then ended their session when their TAC and BrAC had returned to 0.000. We note that the placement of the two sensors challenges the independence assumption from Section 2.2, but for experimental purposes we will include all of the data as independently measured drinking episodes with this caveat in mind. In addition, we did not focus on any specific drinking pattern as including all possible patterns is in line with real-world, varying drinking patterns and may improve the generalizability of our model. In the calculations of Sections 3.4.1 and 3.4.2, as in eq. (2.5), time is discretized by a constant sampling time \tau of 5 minutes and is subject to our zero-order hold assumption. While this challenges the implications of our zero order hold assumption, namely that \tau = .0833 hours implies that subjects' BAC is constant for 5 minutes, this restriction is needed as computational complexity becomes unstable as \tau decreases. In order to achieve this sampling time, we first linearly interpolate all of the data (both BrAC and TAC), and then re-sample at our desired rate of \tau = 5 . For Section 3.4.3, a \tau will be discussed. Further, in all sections we assume a truncated multivariate normal (tMVN) prior \pi_0 (as in eq. (2.20)) on \mathit{\boldsymbol{q}} with mean \mu and covariance matrix \Sigma which varies from example to example.

    Figure 1.  SCRAM Systems transdermal continuous alcohol monitoring device.

    Unfortunately, the USC IRB approved experiments for collecting human subject data were not designed around the problem of estimating the sensor collection chamber inflow and outflow parameters, q_3 and q_4 as in eq. (2.3), nor do the authors have the laboratory facilities or expertise to determine them experimentally. Moreover, since the approach developed in sections 2.1 and 2.2, and in particular our underlying hybrid PDE/ODE model given in 2.3, are relatively novel, no values for q_3 and q_4 are available from either the manufacturers of the sensors or the current literature. Consequently, for the purposes of this study we have chosen values for q_3 and q_4 arbitrarily as q_3 = q_4 = 1 . However, we note that the precise values chosen for q_3 and q_4 had no perceptible qualitative effect on the results to be presented below. Finally we note that all computational work was done in Python 3.7.2 and includes ported MATLAB code from the work of [15,21,24,25], in particular with respect to the creation of the finite-dimensional, discrete-time kernel as in eq. (2.12). Ported code was verified against the original code through the use of unit tests.

    We used surface plots as well as Metropolis Hastings (MH) Markov Chain Monte Carlo (MCMC) methods to validate our convergence in distribution results. Throughout the results described here we have that from eq. (2.14) for all sample times the i.i.d. noise \varepsilon is distributed as N(0, 0.005^2) , and prior \pi_0 as in eq. (2.20) is distributed as the optimal distribution found in Section 6 of [25]. Specifically, the prior is a tMVN random variable with mean, \mu = \left(\begin{array}{l}0.6318\\ 1.0295\end{array} \right) and covariance matrix, \Sigma = \left(\begin{array}{l} 0.0259 & 0.0077\\ 0.0077 & 0.1232 \end{array} \right) with the feasible parameter set, Q , taken to be Q = [0.01, 2.2877] \times [0.01, 2.1410] for \mathit{\boldsymbol{q}} = [q_1, q_2]^T . The choice of 0.005 for the standard deviation of \varepsilon was made to limit the role of noise in our subsequent sampling algorithms so that we may focus on the role of the dimension of our approximating system in the resulting posterior distribution. In addition, when comparing this choice in standard deviation to the peak TAC values of our training dataset, we had a typical peak TAC to noise ratio of 20. For computational reasons, we limit ourselves to measurements from a random subgroup of R = 3 subject drinking episode measurements. Figure 2 contains the resulting surface plots for n values of 1, 3 , and 25 . Further, Table 1 contains the means and credible regions for n values of 1, 2, 3, and 25 as determined by respective 1000 sample (1100 draws with a 100 draw burn-in period) MH MCMC sampling runs. The MCMC sample size chosen here was due to computational complexities and runtimes. Figure 3 displays deconvolution results for a randomly chosen, non-training drinking episode for different values for the dimension of the approximating system, n . This figure used the method from Section 3.3 along with the resulting posteriors as shown in Figure 2 and Table 1.

    Figure 2.  Posterior distribution surface plots for varying finite dimensional approximations of the kernel from eq. (2.12).
    Table 1.  90\% credible circles for MCMC sampled posteriors with noise distribution \mathcal{N}(0, 0.005^2) .
    n Dimension 1 2 3 25
    Mean (q_1, q_2) (0.7185, 0.8512) (0.6829, 0.8651) (0.6776, 0.8686) (0.6719, 0.8716)
    Credible Circle Radius 0.1173 0.1097 0.1029 0.1289

     | Show Table
    DownLoad: CSV
    Figure 3.  Deconvolutions for differing approximating dimension values, n , associated with posteriors from Figure 2 and Table 1.

    We again used surface plots as well as MH MCMC sampling methods to verify our consistency results. For these studies we have assumed that the noise \varepsilon is now distributed as N(0, 0.025^2) while our prior \pi_0 from eq. (2.20) is still the optimal distribution found in Section 6 of [25]. That is, \pi_0 is a tMVN with \mu = \left(\begin{array}{l} 0.6318\\ 1.0295 \end{array} \right) and covariance matrix, \Sigma = \left(\begin{array}{l} 0.0259 & 0.0077\\ 0.0077 & 0.1232 \end{array} \right) with bounds [0.01, 2.2877] and [0.01, 2.1410] for q_1 and q_2 , respectively. The choice for a distribution for the random noise is meant to simulate a more realistic situation where little is assumed known about any external effects that play a role in perturbing the sensor measurements. Consequently the data is assumed noisy. When comparing this choice for the standard deviation of the noise process to the peak TAC values of our training dataset, we had a typical peak TAC to noise ratio of 8.

    To test Theorem 3.5, we generated 276 TAC values using subject-measured BrAC values via eq. (2.14) with a predetermined \mathit{\boldsymbol{q}}_0 value of [1, 1] , n = 24 , and noise variance of 0.025^2 . Table 2 displays the calculated means and 90\% credible circle radii for the posterior distribution eq. (2.22) for increasing amounts of idealized (BrAC, TAC) data pairs ( R ) all generated using the "true" \mathit{\boldsymbol{q}}_0 value previously stated. To calculate these values, MH MCMC samples were drawn with a sample of size 1400 (1500 data points with a 100 sample burn-in phase) where the MCMC sample size was increased from that of Section 3.4.1 due to the increase in noise variance.

    Table 2.  90\% credible circles for MCMC sampled posteriors from idealized TAC data with noise distributed \mathcal{N}(0, 0.025^2) and prior as determined in Section 3.4.2.
    R 1 11 26 101 276
    Mean (q_1, q_2) (0.683, 1.023) (0.734, 1.031) (0.776, 1.025) (0.877, 1.011) (0.942, 1.003)
    Cred. Circle Radius 0.2854 0.1963 0.1677 0.1590 0.0787

     | Show Table
    DownLoad: CSV

    We now investigate the results of Section 3.2 with respect to the field-measured (BrAC, TAC) data pairs. Note that we no longer are able to know the true value of the parameters, \mathit{\boldsymbol{q}}_0 . Surface plots for increasing amounts of subject drinking episode measurements, R = 1, 26, 76 , and 101 are contained within Figure 4. Table 3 displays the calculated means and 90\% credible circle radii for increasing numbers of subjects, and thus data (corresponding to R as in Section 3.2) included in determination of the prior. To calculate these values, for each R , we again used 1400 MH MCMC samples (1500 draws with a 100 sample burn-in phase) generated according to our chosen prior.

    Figure 4.  Posterior distribution surface plots for varying amounts of collected data, m . All images use noise and prior distributions as stated in Section 3.4.2.
    Table 3.  90\% credible circles for MCMC sampled posteriors with noise distributed \mathcal{N}(0, 0.025^2) and prior distribution equivalent to the one in Section 6 of [24], namely a tMVN with \mu = \left( \begin{array}{l}0.6318 \\ 1.0295\end{array} \right) , \Sigma = \left( \begin{array}{l}0.0259 & 0.0077 \\ 0.0077 & 0.1232\end{array} \right) , and \boldsymbol{q} bounds [0.01, 2.2877] and [0.01, 2.1410] .
    R 1 11 26 101
    Mean (q_1, q_2) (0.913, 1.251) (1.426, 1.629) (1.900, 1.551) (2.183, 1.231)
    Credible Circle Radius 0.2824 0.1497 0.1560 0.1254

     | Show Table
    DownLoad: CSV

    As in Section 3.3, we rely on the treatment in [25] for deconvolving BrAC from TAC using a distribution for \mathit{\boldsymbol{\mathcal{q}}} over Q . The chosen distribution was the posterior eq. (2.22) with n = 3 . To determine the posterior we elected to investigate the case where a non-informative, or what is more aptly described as an uneducated, prior was used. Thus we chose a prior of a tMVN random variable with bounds [0.01, 10] \times [0.01, 10] , and parameters \mu = \left(\begin{array}{l} 5\\ 5 \end{array} \right) and \Sigma = \left(\begin{array}{l} 0.7 & 0.1\\ 0.1 & 0.55 \end{array} \right) . The noise used was distributed as \mathcal{N}(0, 0.025^2) . Note that this choice in prior also highlights the effects of data on the posterior by not providing any initial information to the posterior. When comparing this choice in noise standard deviation to the peak TAC values of our training dataset, we had a typical peak TAC to noise ratio of 8. Further, for the subspaces from Section 3.3 we set our discretization to be n = 3 , time discretization as m = 1300 , and discretized Q with m_1 = m_2 = 20 . As with Section 3.4.2, the noise distribution is meant to simulate a situation where little is known about external effects that play a role in determining noise, and so the data is assumed noisy.

    In all of the numerical results presented and discussed in this section, the test dataset used consisted of five drinking episodes from four different participants. These drinking episodes were chosen heuristically so that the test dataset had two drinking episodes with peak BrAC greater than peak TAC, two drinking episodes with peak BrAC less than peak TAC, and one drinking episode with peak BrAC within 0.015 of peak TAC (deemed, "close"). The remaining drinking episodes were used as training data with the added restriction that whenever the desired number of training sets to be used was not too large, BrAC/TAC pairs from any participant who had a dataset included in the selected test data, would be excluded from being among the data used for determining the posterior. The primary exception to this restriction being Figure 6c, wherein we allowed all data that wasn't the current test data point to be included in the training set.

    By linearly interpolating the BrAC and TAC data for each subject in all test and training datasets, we are able to re-sample our data with sampling interval \tau = 45 seconds, and the time discretization m = 1300 previously mentioned. The associated participant IDs, TAC device placement (left vs. right arm), type of drinking pattern used (single, dual, or steady), and number of subjects used in posterior distribution determination ( R ) are labeled in Figures 5 and 6. As in eq. (3.14), we utilized all available non-test subject drinking episode measurements ( R = 136 ) to determine population parameters (r_1^{\ast}, r_2^{\ast}) to be (4.7733, 1.7020) .

    Figure 5.  BrAC deconvolution given TAC and predicted \boldsymbol{q} values for varying test data participants' right arm data with conservative credible regions shaded in gray. Across sub-figures, all training data remained constant with R = 25 . Prior used was tMVN with bounds [0.01, 10] \times [0.01, 10] , \mu = \left( \begin{array}{l}5 \\ 5\end{array} \right) , and \Sigma = \left( \begin{array}{l}0.7 & 0.1 \\ 0.1 & 0.55\end{array} \right) .
    Figure 6.  BrAC deconvolution given TAC and predicted \boldsymbol{q} values for a single test data drinking episode derived using varying amounts of training data, R , with conservative credible regions shaded in gray. All sub-figures use the same test TAC data from a single right arm session from BT333. Prior used was tMVN with bounds [0.01, 10] \times [0.01, 10] , \mu = \left( \begin{array}{l}5 \\ 5\end{array} \right) , and \Sigma = \left( \begin{array}{l}0.7 & 0.1 \\ 0.1 & 0.55\end{array} \right) . Associated data are contained in Table 4.
    Table 4.  Data associated with posterior determination and deconvolution used in Figure 6.
    \textbf{Arm} R \boldsymbol{q} \textbf{mean} \textbf{90.0% Credible} \textbf{Circle Radius} \boldsymbol{q_1} \textbf{range} \boldsymbol{q_2} \textbf{range}
    Right 25 [4.512, 1.346] 1.431 [3.082, 5.943] [0.01, 2.777]
    Right 75 [3.450, 1.215] 1.490 [2.006, 4.986] [0.01, 2.705]
    Right 145 [2.824, 1.000] 1.066 [1.758, 3.890] [0.01, 2.066]

     | Show Table
    DownLoad: CSV

    Figure 5, shows varying deconvolution attempts for three test data participants, whereas Figure 6 shows deconvolution attempts for the same test data participant reading (BT333), with varying amounts of training subject data within the posterior eq. (2.22), R = 25, 75,145 . In both Figures 5 and 6, gray bands represent 90% error regions that are determined by sampling respective parameter posterior distributions and utilizing these samples with (3.13) to determine estimated BrAC values. It follows that these error regions contain the 90% credible regions for the pointwise BrAC values as functions of the population parameters appearing in the model. This is the basis for our referring to them in what follows as conservative credible bands.

    Figure 2 illustrates rapid convergence in dimensionality of our spatial dimensions as n grows, thus bolstering the results of Theorem 3.1. Within two steps ( n = 3 ), we have a graph that visually differs from that of n = 25 in ways barely perceptible. Paired with the credible circles in Table 1, these provide evidence that after n = 3 the mean and radius of the \mathit{\boldsymbol{q}} credible circles stay consistent. Thus one can choose a computationally efficient n value that minimizes data lost when projecting eq. (2.6) into finite dimensions, eq. (2.11).

    For the consistency results, Table 2 exemplifies the theoretical prediction in Theorem 3.5 that as the amount of subject data R grows, the posterior distribution better predicts the true \mathit{\boldsymbol{q}} value by localizing the true parameter \mathit{\boldsymbol{q}}_0 in mean with higher confidence (smaller credible circles). This increasing confidence is backed by the decreasing variance results shown in Figure 4. Notice that although the variance decreases, the mean is allowed to shift as more data are incorporated, as evident from comparing Figure 4c to Figure 4d. This shifting mean is permitted by the theoretical results and is likely due to the incorporation of 70 extra data points. Table 3 displays the shifting of the mean as more data are incorporated while quantitatively displaying a decreasing 90% credible circle radius, as expected.

    As a final note, recall that TAC data were collected simultaneously from both the right and left arms of participants. For an investigation into this see [32].

    In Figure 5a, the deconvolved mean BrAC curve more closely resembles the overall curve of the measured TAC values rather than the desired BrAC, with its increased values towards the latter part of the curve. This is to be expected as the measured TAC plays a role in the Bayesian step, but notice that the severity of the increase in the mean value curve is attenuated when compared to that of the TAC curve (red vs. yellow curves at the five hour mark). A similar phenomenon also appears in Figure 6a. For Figures 6a to 6c, as the number of subject drinking episodes R increases, we find that the mean curve grows towards the actual BrAC curve, an expected convergence phenomenon given the theoretical consistency results from Section 3.2.

    Lastly, the 90% conservative credible bands about the deconvolved BrAC curves appear to always have a lower bound of zero. For the upper bound, the extreme case is shown in Figure 5c. These wide ranges in BrAC values allow us to capture the true BrAC value with high probability, but also leave us capturing far more area under the curve than needed. Thus, there are times when our two-step method would falsely signal that the TAC device wearer is far more inebriated than they actually are. This incorrect signaling might be due in part to the quantitative inaccurate readings in Figure 5c, wherein the TAC curve is greater than the BrAC curve. If our (training) data are mainly composed of the other cases (TAC following BrAC at an attenuated rate), then the algorithm will learn to "guess up" when turning the TAC back into BrAC. Lastly, this phenomenon may be due to the use of an uninformed prior as the credible regions in Table 3 do not approach zero. Hence, in the future use of an informed prior may be preferable.

    We believe that the i.n.i.d. assumption from Section 2.2 (specifically Section 3.2) may not reflect the realities of the data collection method wherein two sensors are worn simultaneously on participants' left and right arms. We are currently investigating the elimination of this i.n.i.d assumption. However, the results from Section 3.4 are quite reasonable and are extremely useful when seeking to use this approach computationally in practice. Further investigation is needed regarding the traveling mean exhibited in the numerical results and how it is related to the non-inclusion of other covariate data (age, height, weight, etc.). This investigation may also be aided by attempting to combine the results of Sections 3.1 and 3.2 and let both the approximating dimension of the kernel, as well as the amount of training data, go to infinity simultaneously.

    We also believe that the packaging of all error sources into a single random variable in Section 2.2 may yield larger uncertainties than formulations where many additive errors are considered. Namely, mixed-effects formulations may be utilized in order to separate errors and might lower overall uncertainty. However, the results from Section 3.4 are again quite reasonable, and the usage of mixed-effects formulations can be left as a design choice when considering the main goals and implementations of the PDE model from Section 2.1.

    When our approach and results are optimized for use in actual practice, some care will have to be taken in regard to the sampling methods used in Sections 3.4.1 and 3.4.2. If Markov Chain Monte Carlo methods are still the method of choice, then issues such as sample size, convergence of the chains, and randomized chain starting points will need to be taken into account. In addition, a laboratory protocol will be need to be developed to estimate the sensor-dependent values of q_3 and q_4 that appear in eq. (2.3). As far as the numerical results presented in Section 3.4 are concerned in regard to the values chosen for q_3 and q_4 , they primarily serve to reinforce the theoretical results in Sections 3.1 and 3.2.

    Finally, Of primary interest is the direct inversion of BrAC, u , given TAC as in eq. (2.12) without the need for a two-step process like that of the method used in this paper. We believe that a hierarchical model paired with a Gaussian Process framework may reduce the problem down to a single step (see, [52]). In such a framework, we place a prior on \mathit{\boldsymbol{q}} , as well as a function space prior over u . In this way, we obtain a method that statistically deconvolves BrAC from TAC while providing a distribution from which we may derive error bars on the estimated BrAC values. We are also currently examining the inclusion of another hierarchical Bayesian model that incorporates covariates in both priors placed over \mathit{\boldsymbol{q}} and u . We believe that this will improve the accuracy of our predictions by allowing the use of all available subject and environment data.

    The authors wish to acknowledge that this research was supported in part by the National Institute for Alcohol Abuse and Alcoholism (NIAAA) under grant no. R01AA026368.

    The authors declare no conflicts of interest.



    [1] M. Binkley et al., "Defining Twenty-First Century Skills BT - Assessment and Teaching of 21st Century Skills, " P. Griffin, B. McGaw, and E. Care, Eds. Dordrecht: Springer Netherlands, 2012, pp. 17–66.
    [2] Harman K, Bich NTN (2010) Reforming Teaching and Learning in Vietnam's Higher Education System BT - Reforming Higher Education in Vietnam: Challenges and Priorities. In: Harman G, Hayden M, Nghi PT (eds). Springer Netherlands, Dordrecht, pp 65–86
    [3] Pakpour AH, Zeidi IM, Emamjomeh MM, et al (2014) Household waste behaviours among a community sample in Iran: An application of the theory of planned behaviour. Waste Manag 34: 980–986. https://doi.org/https://doi.org/10.1016/j.wasman.2013.10.028 doi: 10.1016/j.wasman.2013.10.028
    [4] Peraturan Presiden Indonesia, Perpres No 83/2018. 2018.
    [5] Sugiyono A, Adiarso (2021) Development of Natural Gas Infrastructure to Enhance National Energy Security in Indonesia. IOP Conf Ser Mater Sci Eng 1053: 12099. https://doi.org/10.1088/1757-899x/1053/1/012099 doi: 10.1088/1757-899x/1053/1/012099
    [6] Erahman QF, Purwanto WW (2021) Energy Security: A Case Study of Indonesia BT - Energy and Environmental Security in Developing Countries. In: Asif M (ed). Springer International Publishing, Cham, pp 49–74
    [7] Kumara INS, Ariastina WG, Sukerayasa IW, Giriantari IAD (2014) On the potential and progress of renewable electricity generation in Bali. In: 2014 6th International Conference on Information Technology and Electrical Engineering (ICITEE). pp 1–6
    [8] Yana S, Nizar M, Irhamni, Mulyati D (2022) Biomass waste as a renewable energy in developing bio-based economies in Indonesia: A review. Renew Sustain Energy Rev 160: 112268. https://doi.org/https://doi.org/10.1016/j.rser.2022.112268 doi: 10.1016/j.rser.2022.112268
    [9] Sharvini SR, Noor ZZ, Chong CS, et al (2018) Energy consumption trends and their linkages with renewable energy policies in East and Southeast Asian countries: Challenges and opportunities. Sustain Environ Res 28: 257–266. https://doi.org/https://doi.org/10.1016/j.serj.2018.08.006 doi: 10.1016/j.serj.2018.08.006
    [10] Pambudi NA, Firdaus RA, Rizkiana R, et al (2023) Renewable Energy in Indonesia: Current Status, Potential, and Future Development. Sustainability 15
    [11] Soonmin H, Lomi A, Okoroigwe EC, Urrego LR (2019) Investigation of Solar Energy: The Case Study in. 9:
    [12] Anitha K, Verchot L V, Joseph S, et al (2015) A review of forest and tree plantation biomass equations in Indonesia. Ann For Sci 72: 981–997. https://doi.org/10.1007/s13595-015-0507-4 doi: 10.1007/s13595-015-0507-4
    [13] Afifah AS, Suryawan IWK, Sarwono A (2020) Microalgae production using photo-bioreactor with intermittent aeration for municipal wastewater substrate and nutrient removal. Commun Sci Technol 5: 107–111. https://doi.org/10.21924/cst.5.2.2020.200 doi: 10.21924/cst.5.2.2020.200
    [14] Sari MM, Inoue T, Harryes RK, et al (2022) Potential of Recycle Marine Debris in Pluit Emplacement, Jakarta to Achieve Sustainable Reduction of Marine Waste Generation. Int J Sustain Dev Plan 17: 119–125
    [15] Malico I, Nepomuceno Pereira R, Gonçalves AC, Sousa AMO (2019) Current status and future perspectives for energy production from solid biomass in the European industry. Renew Sustain Energy Rev 112: 960–977. https://doi.org/https://doi.org/10.1016/j.rser.2019.06.022 doi: 10.1016/j.rser.2019.06.022
    [16] Sirohi R, Kumar Gaur V, Kumar Pandey A, et al (2021) Harnessing fruit waste for poly-3-hydroxybutyrate production: A review. Bioresour Technol 326: 124734. https://doi.org/https://doi.org/10.1016/j.biortech.2021.124734 doi: 10.1016/j.biortech.2021.124734
    [17] Amen R, Hameed J, Albashar G, et al (2021) Modelling the higher heating value of municipal solid waste for assessment of waste-to-energy potential: A sustainable case study. J Clean Prod 287: 125575. https://doi.org/https://doi.org/10.1016/j.jclepro.2020.125575 doi: 10.1016/j.jclepro.2020.125575
    [18] Ali Shah SA, Longsheng C, Solangi YA, et al (2021) Energy trilemma based prioritization of waste-to-energy technologies: Implications for post-COVID-19 green economic recovery in Pakistan. J Clean Prod 284: 124729. https://doi.org/https://doi.org/10.1016/j.jclepro.2020.124729 doi: 10.1016/j.jclepro.2020.124729
    [19] Jakobsson C, Fujii S, Gä rling T (2000) Determinants of private car users' acceptance of road pricing. Transp Policy 7: 153–158. https://doi.org/https://doi.org/10.1016/S0967-070X(00)00005-6 doi: 10.1016/S0967-070X(00)00005-6
    [20] Mills B, Schleich J (2010) What's driving energy efficient appliance label awareness and purchase propensity? Energy Policy 38: 814–825. https://doi.org/https://doi.org/10.1016/j.enpol.2009.10.028 doi: 10.1016/j.enpol.2009.10.028
    [21] Achillas C, Vlachokostas C, Moussiopoulos N, et al (2011) Social acceptance for the development of a waste-to-energy plant in an urban area. Resour Conserv Recycl 55: 857–863. https://doi.org/https://doi.org/10.1016/j.resconrec.2011.04.012 doi: 10.1016/j.resconrec.2011.04.012
    [22] Upreti BR, van der Horst D (2004) National renewable energy policy and local opposition in the UK: the failed development of a biomass electricity plant. Biomass Bioenergy 26: 61–69. https://doi.org/https://doi.org/10.1016/S0961-9534(03)00099-0 doi: 10.1016/S0961-9534(03)00099-0
    [23] Baxter J, Ho Y, Rollins Y, Maclaren V (2016) Attitudes toward waste to energy facilities and impacts on diversion in Ontario, Canada. Waste Manag 50: 75–85. https://doi.org/https://doi.org/10.1016/j.wasman.2016.02.017 doi: 10.1016/j.wasman.2016.02.017
    [24] He J, Lin B (2019) Assessment of waste incineration power with considerations of subsidies and emissions in China. Energy Policy 126: 190–199. https://doi.org/https://doi.org/10.1016/j.enpol.2018.11.025 doi: 10.1016/j.enpol.2018.11.025
    [25] Bondes M, Johnson T (2017) Beyond Localized Environmental Contention: Horizontal and Vertical Diffusion in a Chinese Anti-Incinerator Campaign. J Contemp China 26: 504–520. https://doi.org/10.1080/10670564.2017.1275079 doi: 10.1080/10670564.2017.1275079
    [26] Damayanti P, Moersidik SS, Haryanto JT (2021) Waste to Energy in Sunter, Jakarta, Indonesia: Plans and Challenges. IOP Conf Ser Earth Environ Sci 940: 012033. https://doi.org/10.1088/1755-1315/940/1/012033 doi: 10.1088/1755-1315/940/1/012033
    [27] Mulyadin R M, Ariawan K, Iqbal M (2018)Conflict of waste management in DKI Jakarta and its recomended solutions. J Analisis Kebijakan Kehutanan 15: 179–191. https://doi.org/10.20886/jakk.2018.15.2.179-191 doi: 10.20886/jakk.2018.15.2.179-191
    [28] Sagala G, Kristanto GA, Kusuma MA, Rizki S (2018) Assessment of municipal solid waste as refuse derived fuel in the cement industry. Int J Adv Sci Eng Inf Technol 8: 1062–1070. https://doi.org/10.18517/ijaseit.8.4.3469 doi: 10.18517/ijaseit.8.4.3469
    [29] Guo M, Nowakowska-Grunt J, Gorbanyov V, Egorova M (2020) Green Technology and Sustainable Development: Assessment and Green Growth Frameworks. Sustain. 12
    [30] Haghani M, Bliemer MCJ, Rose JM, et al (2021) Hypothetical bias in stated choice experiments: Part I. Macro-scale analysis of literature and integrative synthesis of empirical evidence from applied economics, experimental psychology and neuroimaging. J Choice Model 41: 100309. https://doi.org/https://doi.org/10.1016/j.jocm.2021.100309 doi: 10.1016/j.jocm.2021.100309
    [31] Rahaman MM, Iqbal MH (2021) Willingness-to-pay for improved cyclone early warning services across coastal Bangladesh: Application of choice experiment. Int J Disaster Risk Reduct 61: 102344. https://doi.org/https://doi.org/10.1016/j.ijdrr.2021.102344 doi: 10.1016/j.ijdrr.2021.102344
    [32] Mazzocchi C, Orsi L, Sali G (2021) Consumers' Attitudes for Sustainable Mountain Cheese. Sustain. 13
    [33] Medeiros G, Florindo T, Talamini E, et al (2020) Optimising Tree Plantation Land Use in Brazil by Analysing Trade-Offs between Economic and Environmental Factors Using Multi-Objective Programming. For. 11
    [34] Grilli G, Tyllianakis E, Luisetti T, et al (2021) Prospective tourist preferences for sustainable tourism development in Small Island Developing States. Tour Manag 82: 104178. https://doi.org/https://doi.org/10.1016/j.tourman.2020.104178 doi: 10.1016/j.tourman.2020.104178
    [35] Tomić T, Schneider DR (2020) Circular economy in waste management – Socio-economic effect of changes in waste management system structure. J Environ Manage 267: 110564. https://doi.org/https://doi.org/10.1016/j.jenvman.2020.110564 doi: 10.1016/j.jenvman.2020.110564
    [36] Khan AH, López-Maldonado EA, Alam SS, et al (2022) Municipal solid waste generation and the current state of waste-to-energy potential: State of art review. Energy Convers Manag 267: 115905. https://doi.org/https://doi.org/10.1016/j.enconman.2022.115905 doi: 10.1016/j.enconman.2022.115905
    [37] Iqbal A, Liu X, Chen G-H (2020) Municipal solid waste: Review of best practices in application of life cycle assessment and sustainable management techniques. Sci Total Environ 729: 138622. https://doi.org/https://doi.org/10.1016/j.scitotenv.2020.138622 doi: 10.1016/j.scitotenv.2020.138622
    [38] Istrate I-R, Iribarren D, Gálvez-Martos J-L, Dufour J (2020) Review of life-cycle environmental consequences of waste-to-energy solutions on the municipal solid waste management system. Resour Conserv Recycl 157: 104778. https://doi.org/https://doi.org/10.1016/j.resconrec.2020.104778 doi: 10.1016/j.resconrec.2020.104778
    [39] Al-Shetwi AQ (2022) Sustainable development of renewable energy integrated power sector: Trends, environmental impacts, and recent challenges. Sci Total Environ 822: 153645. https://doi.org/https://doi.org/10.1016/j.scitotenv.2022.153645 doi: 10.1016/j.scitotenv.2022.153645
    [40] Gutberlet J, Sorroche S, Martins Baeder A, et al (2021) Waste Pickers and Their Practices of Insurgency and Environmental Stewardship. J Environ Dev 30: 369–394. https://doi.org/10.1177/10704965211055328 doi: 10.1177/10704965211055328
    [41] Mö slinger M, Ulpiani G, Vetters N (2023) Circular economy and waste management to empower a climate-neutral urban future. J Clean Prod 421: 138454. https://doi.org/https://doi.org/10.1016/j.jclepro.2023.138454 doi: 10.1016/j.jclepro.2023.138454
    [42] Nie P, Dahanayake KC, Sumanarathna N (2023) Exploring UAE's transition towards circular economy through construction and demolition waste management in the pre-construction stage–A case study approach. Smart Sustain Built Environ ahead-of-p: https://doi.org/10.1108/SASBE-06-2022-0115
    [43] Suryawan IWK, Lee C-H (2023a) Citizens' willingness to pay for adaptive municipal solid waste management services in Jakarta, Indonesia. Sustain Cities Soc 97. https://doi.org/https://doi.org/10.1016/j.scs.2023.104765 doi: 10.1016/j.scs.2023.104765
    [44] Montorsi L, Milani M, Venturelli M (2018) Economic assessment of an integrated waste to energy system for an urban sewage treatment plant: A numerical approach Energy, 158: 105-110.
    [45] Malinauskaite J, Jouhara H, Czajczyńska D, et al (2017) Municipal solid waste management and waste-to-energy in the context of a circular economy and energy recycling in Europe. Energy 141: 2013–2044. https://doi.org/https://doi.org/10.1016/j.energy.2017.11.128 doi: 10.1016/j.energy.2017.11.128
    [46] Suryawan IWK, Septiariva IY, Fauziah EN, et al (2022) Municipal Solid Waste to Energy : Palletization of Paper and Garden Waste into Refuse Derived Fuel. J Ecol Eng 23: 64–74
    [47] Tozlu A, Ö zahi E, Abuşoğlu A (2016) Waste to energy technologies for municipal solid waste management in Gaziantep. Renew Sustain Energy Rev 54: 809–815. https://doi.org/https://doi.org/10.1016/j.rser.2015.10.097 doi: 10.1016/j.rser.2015.10.097
    [48] Duan Z, Kjeldsen P, Scheutz C (2021) Trace gas composition in landfill gas at Danish landfills receiving low-organic waste. Waste Manag 122: 113–123. https://doi.org/https://doi.org/10.1016/j.wasman.2021.01.001 doi: 10.1016/j.wasman.2021.01.001
    [49] Fetanat A, Mofid H, Mehrannia M, Shafipour G (2019) Informing energy justice based decision-making framework for waste-to-energy technologies selection in sustainable waste management: A case of Iran. J Clean Prod 228: 1377–1390. https://doi.org/https://doi.org/10.1016/j.jclepro.2019.04.215 doi: 10.1016/j.jclepro.2019.04.215
    [50] Huang GC-L, Chen R-Y (2021) Injustices in phasing out nuclear power?: Exploring limited public participation and transparency in Taiwan's transition away from nuclear energy. Energy Res Soc Sci 71: 101808. https://doi.org/https://doi.org/10.1016/j.erss.2020.101808 doi: 10.1016/j.erss.2020.101808
    [51] Badan Pusat Statistik provinsi DKI Jakarta, "Provinsi DKI Jakarta Dalam Angka 2021, " Provinsi DKI Jakarta Dalam Angka 2021, 2021.
    [52] Suryawan IWK, Lee C-H (2023) Community preferences in carbon reduction: Unveiling the importance of adaptive capacity for solid waste management. Ecol Indic 157: 111226. https://doi.org/https://doi.org/10.1016/j.ecolind.2023.111226 doi: 10.1016/j.ecolind.2023.111226
    [53] Suryawan IWK, Septiariva IY, Sari MM, et al (2023) Acceptance of Waste to Energy (WtE) Technology by Local Residents of Jakarta City, Indonesia to Achieve Sustainable Clean and Environmentally Friendly Energy. J Sustain Dev Energy, Water Environ Syst 11: 1004
    [54] Shipe ME, Deppen SA, Farjah F, Grogan EL (2019) Developing prediction models for clinical use using logistic regression: an overview. J Thorac Dis 11: S574–S584. https://doi.org/10.21037/jtd.2019.01.25
    [55] Li W, Long R, Chen H, et al (2020) Willingness to pay for hydrogen fuel cell electric vehicles in China: A choice experiment analysis. Int J Hydrogen Energy 45: 34346–34353. https://doi.org/https://doi.org/10.1016/j.ijhydene.2020.01.046 doi: 10.1016/j.ijhydene.2020.01.046
    [56] Crouch GI, Del Chiappa G, Perdue RR (2019) International convention tourism: A choice modelling experiment of host city competition. Tour Manag 71: 530–542. https://doi.org/https://doi.org/10.1016/j.tourman.2018.10.002 doi: 10.1016/j.tourman.2018.10.002
    [57] Deliormanlı AH (2012) Cerchar abrasivity index (CAI) and its relation to strength and abrasion test methods for marble stones. Constr Build Mater 30: 16–21. https://doi.org/https://doi.org/10.1016/j.conbuildmat.2011.11.023 doi: 10.1016/j.conbuildmat.2011.11.023
    [58] McDermott R, O'Dell C (2001) Overcoming cultural barriers to sharing knowledge. J Knowl Manag 5: 76–85. https://doi.org/10.1108/13673270110384428 doi: 10.1108/13673270110384428
    [59] Iorfa SK, Ottu IFA, Oguntayo R, et al (2020) COVID-19 Knowledge, Risk Perception, and Precautionary Behavior Among Nigerians: A Moderated Mediation Approach. Front Psychol 11: 1–10. https://doi.org/10.3389/fpsyg.2020.566773 doi: 10.3389/fpsyg.2020.566773
    [60] Gubbels J, van der Put CE, Assink M (2019) Risk Factors for School Absenteeism and Dropout: A Meta-Analytic Review. J Youth Adolesc 48: 1637–1667. https://doi.org/10.1007/s10964-019-01072-5 doi: 10.1007/s10964-019-01072-5
    [61] Ulhasanah N, Goto N (2018) Assessment of citizens' environmental behavior toward municipal solid waste management for a better and appropriate system in Indonesia: a case study of Padang City. J Mater Cycles Waste Manag 20: 1257–1272. https://doi.org/10.1007/s10163-017-0691-4 doi: 10.1007/s10163-017-0691-4
    [62] Singh A, Sushil (2021) Integrated approach for finding the causal effect of waste management over sustainability in the organization. Benchmarking An Int J 28: 3040–3073. https://doi.org/10.1108/BIJ-08-2020-0419 doi: 10.1108/BIJ-08-2020-0419
    [63] Nurany F, Ismail I, Ferdaus Noor Aulady M, et al (2020) Article Review: The Policy Implementation of Waste to Energy Power Plant – Pilot Project in Surabaya, Indonesia. E3S Web Conf 190:
    [64] Mustia DI, Edy S, Nurul A (2021) Analysis of waste composition as a source of refuse-derived fuel in Cilacap. IOP Conf Ser Earth Environ Sci 896: 12063. https://doi.org/10.1088/1755-1315/896/1/012063 doi: 10.1088/1755-1315/896/1/012063
    [65] Rachman SA, Hamdi M, Djaenuri A, Sartika I (2020) Model of Public Policy Implementation for Refused Derived Fuel (RDF) Waste Management in Cilacap Regency. Int J Sci Soc 2. https://doi.org/10.200609/ijsoc.v2i4.239
    [66] Siregar SRH, Saragih BR, Surjosatyo A (2018) Evaluation of Waste Energy Conversion Technology using Analitycal Hierarchy Process in Bantargebang Landfill, Indonesia. E3S Web Conf 67: 1–5. https://doi.org/10.1051/e3sconf/20186702012 doi: 10.1051/e3sconf/20186702012
    [67] Kristanto GA, Koven W (2019) Estimating greenhouse gas emissions from municipal solid waste management in Depok, Indonesia. City Environ Interact 4: 100027. https://doi.org/https://doi.org/10.1016/j.cacint.2020.100027 doi: 10.1016/j.cacint.2020.100027
    [68] Legino S, Arianto R, Pasra N (2019) The attainment of 100 percent electrification ratio in the archipelago of Indonesia by people way electricity initiative. J Phys Conf Ser 1282. https://doi.org/10.1088/1742-6596/1282/1/012057
    [69] Legino S, Hidayawanti R, Putra IS, Pribadi A (2019) Reducing coal consumption by people empowerment using local waste processing unit. J Phys Conf Ser 1217: 12028. https://doi.org/10.1088/1742-6596/1217/1/012028 doi: 10.1088/1742-6596/1217/1/012028
    [70] Brunner IMIM, Norhidayat A, Brunner SM (2021) Processing of Organic Waste and Biomass Waste with Waste Processing Technology at the Source. VI: 2085–2095. https://doi.org/10.32672/jse.v6i3.3120
    [71] Nasruddin, Idrus Alhamid M, Daud Y, et al (2016) Potential of geothermal energy for electricity generation in Indonesia: A review. Renew Sustain Energy Rev 53: 733–740. https://doi.org/https://doi.org/10.1016/j.rser.2015.09.032 doi: 10.1016/j.rser.2015.09.032
    [72] Saxinger G, Krasnoshtanova NE, Illmeier G (2018) In limbo between state and corporate responsibility: Transport infrastructure in the oil village Verkhnemarkovo, Irkutskaya Oblast in Russia. IOP Conf Ser Earth Environ Sci 190: 12062. https://doi.org/10.1088/1755-1315/190/1/012062 doi: 10.1088/1755-1315/190/1/012062
    [73] Gupta D, Garg A (2020) Sustainable development and carbon neutrality: Integrated assessment of transport transitions in India. Transp Res Part D Transp Environ 85: 102474. https://doi.org/https://doi.org/10.1016/j.trd.2020.102474 doi: 10.1016/j.trd.2020.102474
    [74] Persyn D, Díaz-Lanchas J, Barbero J (2022) Estimating road transport costs between and within European Union regions. Transp Policy 124: 33–42. https://doi.org/https://doi.org/10.1016/j.tranpol.2020.04.006 doi: 10.1016/j.tranpol.2020.04.006
    [75] Chu Van T, Ramirez J, Rainey T, et al (2019) Global impacts of recent IMO regulations on marine fuel oil refining processes and ship emissions. Transp Res Part D Transp Environ 70: 123–134. https://doi.org/https://doi.org/10.1016/j.trd.2019.04.001 doi: 10.1016/j.trd.2019.04.001
    [76] Marafi A, Albazzaz H, Rana MS (2019) Hydroprocessing of heavy residual oil: Opportunities and challenges. Catal Today 329: 125–134. https://doi.org/https://doi.org/10.1016/j.cattod.2018.10.067 doi: 10.1016/j.cattod.2018.10.067
    [77] Gabdulkhakov RR, Rudko VA, Pyagay IN (2022) Methods for modifying needle coke raw materials by introducing additives of various origin (review). Fuel 310: 122265. https://doi.org/https://doi.org/10.1016/j.fuel.2021.122265 doi: 10.1016/j.fuel.2021.122265
    [78] Mbazima SJ, Masekameni MD, Mmereki D (2022) Waste-to-energy in a developing country: The state of landfill gas to energy in the Republic of South Africa. Energy Explor Exploit 40: 1287–1312. https://doi.org/10.1177/01445987221084376 doi: 10.1177/01445987221084376
    [79] Palaniandy P, Aziz HA, Wang LK, et al (2022) Sanitary Landfill Types and Design BT - Solid Waste Engineering and Management: Volume 2. In: Wang LK, Wang M-HS, Hung Y-T (eds). Springer International Publishing, Cham, pp 543–597
    [80] Roswulandari A, Daerobi A, Gravitiani E (2019) Waste to Energy (WTE) Putri Cempo As Urban Innovation: A Financial Analysis. 8th International Conference on Sustainable Environment and Architecture. https://doi.org/10.2991/senvar-18.2019.25
    [81] Sunarto, Sulistyaningsih T (2018) Integrated sustainable waste management in Malang City, East Java, Indonesia. AIP Conf Proc 1977: 30043. https://doi.org/10.1063/1.5042963 doi: 10.1063/1.5042963
    [82] Prasetya DA, Sanusi A, Chandrarin G, et al (2019) Small and Medium Enterprises Problem and Potential Solutions for Waste Management. J Southwest Jiaotong Univ 54. https://doi.org/10.35741/issn.0258-2724.54.6.21
    [83] Prayitno G, Hakim AN, Meidiana C (2020) Community Participation on the Self Help Group of Methane Gas (Biogas) Management as Renewable Energy in Indonesia. Int J Energy Econ Policy 11: 200–211
    [84] Purmessur B, Surroop D (2019) Power generation using landfill gas generated from new cell at the existing landfill site. J Environ Chem Eng 7: 103060. https://doi.org/https://doi.org/10.1016/j.jece.2019.103060 doi: 10.1016/j.jece.2019.103060
    [85] USEPA, "Bioreactors, " 2018. https://archive.epa.gov/epawaste/nonhaz/municipal/web/html/bioreactors.html. (accessed Apr. 25, 2022).
    [86] Grossule V, Morello L, Cossu R, Lavagnolo M (2018) BIOREACTOR LANDFILLS: COMPARISON AND KINETICS OF THE DIFFERENT SYSTEMS. Detritus In Press: 1. https://doi.org/10.31025/2611-4135/2018.13703
    [87] Lin C-Y, Chai WS, Lay C-H, et al (2021) Optimization of Hydrolysis-Acidogenesis Phase of Swine Manure for Biogas Production Using Two-Stage Anaerobic Fermentation. Process. 9
    [88] Kumarasiri B, Dissanayake P (2021) Barriers to implementing waste-to-energy projects in Sri Lanka: a PESTEL analysis. Built Environ Proj Asset Manag 11: 544–558. https://doi.org/10.1108/BEPAM-03-2020-0051 doi: 10.1108/BEPAM-03-2020-0051
    [89] Panepinto D, Zanetti M (2021) Technical and Environmental Comparison among Different Municipal Solid Waste Management Scenarios. Sustain. 13
    [90] Sütterlin B, Siegrist M (2017) Public acceptance of renewable energy technologies from an abstract versus concrete perspective and the positive imagery of solar power. Energy Policy 106: 356–366. https://doi.org/https://doi.org/10.1016/j.enpol.2017.03.061 doi: 10.1016/j.enpol.2017.03.061
    [91] Gayatri SU (2019) ANALISIS JEJARING KEBIJAKAN DALAM PENGELOLAAN SAMPAH MENJADI ENERGI LISTRIK DI DKI JAKARTA. Universitas Gadjah Mada, Yogyakarta
    [92] Come Zebra EI, van der Windt HJ, Nhumaio G, Faaij APC (2021) A review of hybrid renewable energy systems in mini-grids for off-grid electrification in developing countries. Renew Sustain Energy Rev 144: 111036. https://doi.org/https://doi.org/10.1016/j.rser.2021.111036 doi: 10.1016/j.rser.2021.111036
    [93] Carley S, Konisky DM (2020) The justice and equity implications of the clean energy transition. Nat Energy 5: 569–577. https://doi.org/10.1038/s41560-020-0641-6 doi: 10.1038/s41560-020-0641-6
    [94] Hoang AT, Sandro Nižetić, Olcer AI, et al (2021) Impacts of COVID-19 pandemic on the global energy system and the shift progress to renewable energy: Opportunities, challenges, and policy implications. Energy Policy 154: 112322. https://doi.org/https://doi.org/10.1016/j.enpol.2021.112322 doi: 10.1016/j.enpol.2021.112322
    [95] Koengkan M, Poveda YE, Fuinhas JA (2020) Globalisation as a motor of renewable energy development in Latin America countries. GeoJournal 85: 1591–1602. https://doi.org/10.1007/s10708-019-10042-0 doi: 10.1007/s10708-019-10042-0
    [96] Sari MM, Suryawan IWK, Septiariva IY (2023) Processing of Biodegradable Waste from Ceremonial Activities in Bali with Black Soldier Fly (BSF) Larvae. J Multidiscip Appl Nat Sci 3: 138–148. https://doi.org/10.47352/jmans.2774-3047.175 doi: 10.47352/jmans.2774-3047.175
  • This article has been cited by:

    1. Clemens Oszkinat, Susan E. Luczak, I. G. Rosen, 2022, Physics-Informed Learning: Distributed Parameter Systems, Hidden Markov Models, and the Viterbi Algorithm, 978-1-6654-5196-3, 266, 10.23919/ACC53348.2022.9867145
    2. Clemens Oszkinat, Tianlan Shao, Chunming Wang, I G Rosen, Allison D Rosen, Emily B Saldich, Susan E Luczak, Blood and breath alcohol concentration from transdermal alcohol biosensor data: estimation and uncertainty quantification via forward and inverse filtering for a covariate-dependent, physics-informed, hidden Markov model* , 2022, 38, 0266-5611, 055002, 10.1088/1361-6420/ac5ac7
    3. Mengsha Yao, Susan E. Luczak, Emily B. Saldich, I. Gary Rosen, A population model‐based linear‐quadratic Gaussian compensator for the control of intravenously infused alcohol studies and withdrawal symptom prophylaxis using transdermal sensing, 2022, 0143-2087, 10.1002/oca.2934
    4. Clemens Oszkinat, Susan E. Luczak, I. Gary Rosen, An abstract parabolic system-based physics-informed long short-term memory network for estimating breath alcohol concentration from transdermal alcohol biosensor data, 2022, 34, 0941-0643, 18933, 10.1007/s00521-022-07505-w
    5. Jiachen Yu, Catharine E. Fairbairn, Laura Gurrieri, Eddie P. Caumiant, Validating transdermal alcohol biosensors: a meta‐analysis of associations between blood/breath‐based measures and transdermal alcohol sensor output, 2022, 117, 0965-2140, 2805, 10.1111/add.15953
    6. Haoxing Liu, Larry Goldstein, Susan E. Luczak, I. G. Rosen, 2023, Delta-Method Induced Confidence Bands for a Parameter-Dependent Evolution System with Application to Transdermal Alcohol Concentration Monitoring, 979-8-3503-0124-3, 6211, 10.1109/CDC49753.2023.10383767
    7. Bob M. Lansdorp, Flux-Type versus Concentration-Type Sensors in Transdermal Measurements, 2023, 13, 2079-6374, 845, 10.3390/bios13090845
    8. Kyla-Rose Walden, Emily B. Saldich, Georgia Wong, Haoxing Liu, Chunming Wang, I. Gary Rosen, Susan E. Luczak, 2023, 79, 9780443193866, 271, 10.1016/bs.plm.2023.06.002
    9. Clemens Oszkinat, Susan E. Luczak, I. G. Rosen, Uncertainty Quantification in Estimating Blood Alcohol Concentration From Transdermal Alcohol Level With Physics-Informed Neural Networks, 2023, 34, 2162-237X, 8094, 10.1109/TNNLS.2022.3140726
    10. Lernik Asserian, Susan E. Luczak, I. G. Rosen, Computation of nonparametric, mixed effects, maximum likelihood, biosensor data based-estimators for the distributions of random parameters in an abstract parabolic model for the transdermal transport of alcohol, 2023, 20, 1551-0018, 20345, 10.3934/mbe.2023900
    11. Maria Allayioti, Clemens Oszkinat, Emily Saldich, Larry Goldstein, Susan E. Luczak, Chunming Wang, I. G. Rosen, 2023, Parametric and Non-Parametric Estimation of a Random Diffusion Equation-Based Population Model for Deconvolving Blood/Breath Alcohol Concentration from Transdermal Alcohol Biosensor Data with Uncertainty Quantification*, 979-8-3503-2806-6, 313, 10.23919/ACC55779.2023.10156287
  • Reader Comments
  • © 2023 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(1779) PDF downloads(117) Cited by(1)

Figures and Tables

Figures(3)  /  Tables(4)

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog