Appendix A. Mathematical Modeling of a Nonlinear System with Imprecisely Known Parameters and Domain Boundaries
The computational model of a physical system comprises equations that relate the system’s independent variables and parameters to the system’s state variables. The model parameters usually stem from processes that are external to the system under consideration and are seldom, if ever, known precisely. The known characteristics of the model parameters may include their nominal (expected/mean) values and, possibly, higher-order moments or cumulants (i.e., variance/covariances, skewness, kurtosis), which are usually determined from experimental data and/or processes external to the physical system under consideration. Occasionally, just the lower and the upper bounds may be known for some model parameters, expressed by inequality and/or equality constraints that delimit the ranges of the system’s parameters that are known. Without loss of generality, the imprecisely known model parameters can be considered to be real-valued scalar quantities. These model parameters will be denoted as ,…,, where denotes the total number of imprecisely known parameters underlying the model under consideration. For subsequent developments, it is convenient to consider that these parameters are components of a vector of parameters, denoted as , where is also a normed linear space and where denotes the -dimensional subset of the set of real scalars. The components of the -dimensional column vector are considered to include imprecisely known geometrical parameters that characterize the physical system’s boundaries in the phase space of the model’s independent variables. The nominal parameter values will be denoted as ; the superscript “0” will be used throughout this work to denote nominal values. Matrices will be denoted using capital bold letters, whereas vectors will be denoted using either capital or lower-case bold letters. The symbol “” will be used to denote “is defined as” or “is by definition equal to.” Transposition will be indicated by a dagger superscript.
The generic nonlinear model is considered to comprise independent variables, which will be denoted as and that are considered to be components of a -dimensional column vector denoted as , where the sub/superscript “TI” denotes the total number of independent variables. The vector of independent variables is considered to be defined on a phase-space domain that will be denoted as and that is defined as follows: . The lower boundary point of an independent variable is denoted as and the corresponding upper boundary point is denoted as . The boundary is also considered to be imprecisely known since it may depend on both geometrical parameters and material properties. A typical example of boundaries that depend on both geometrical parameters and material properties are the boundaries facing vacuum in models based on diffusion theory, where conditions are imposed on the extrapolated boundary of the respective spatial domain. The extrapolated boundary depends both on the imprecisely known physical dimensions of the problem’s domain and also on the medium’s properties, such as atomic number densities and microscopic transport cross-sections. The boundary of , which will be denoted as , comprises the set of all of the endpoints of the respective intervals on which the components of are defined, i.e., .
A nonlinear physical system can be generally modeled by means of coupled equations, which can be represented in operator form as follows:
The quantities that appear in Equation (A1) are defined as follows:
is a -dimensional column vector of dependent variables; the abbreviation denotes “total number of dependent variables.” The functions , denote the system’s dependent variables (also called “state functions”); , where is a normed linear space over the scalar field of real numbers.
denotes a -dimensional column vector. The components are operators (including differential, difference, integral, distributions, and/or finite or infinite matrices) acting (usually) nonlinearly on the dependent variables , the independent variables , and the model parameters . The mapping is defined on the combined domains of the model’s parameters and state functions, i.e., , where , , , .
is a -dimensional column vector that represents inhomogeneous source terms, which usually depend nonlinearly on the uncertain parameters . The vector is defined on a normed linear space denoted as , i.e., .
The equalities in this work are considered to hold in the weak (distributional) sense. The right sides of Equation (A1) and of other various equations to be derived in this work may contain generalized functions/functionals, particularly Dirac-distributions and derivatives thereof.
Boundary and/or initial conditions must also be provided if differential operators appear in Equation (A1). In operator form, these boundaries and/or initial conditions are represented as follows:
where the column vector
has
components, all of which are identically zero, i.e.,
In Equation (A2), the components of are nonlinear operators in and , which are defined on the boundary of the model’s domain . The components of comprise inhomogeneous boundary sources that are nonlinear functions of .
Solving Equations (A1) and (A2) at the nominal parameter values, denoted as
, provides the nominal solution
, i.e., the vectors
and
satisfy the following equations:
The results computed using a mathematical model are customarily called “model responses” (or “system responses,” “objective functions,” or “indices of performance”). In general, a function-valued (i.e., operator-type) response
can be represented by a spectral expansion in multidimensional orthogonal polynomials or Fourier series of the form:
where the quantities
,
, denote the corresponding spectral functions (e.g., orthogonal polynomials or Fourier exponential/trigonometric functions) and where the spectral Fourier coefficients
are defined as follows:
The coefficients can themselves be considered model responses since the spectral polynomials are perfectly well known, whereas the expansion coefficients will contain all of the dependencies (directly or indirectly through the state functions) of the respective response on the imprecisely known model parameters. This way, the sensitivity analysis of an operator-valued response can be reduced to the sensitivity analysis of the scalar-valued responses .
A measurement of a physical quantity that depends on the model’s state functions and parameters can be considered to be a response denoted as
, which is to be evaluated at
, where
denotes the location in phase space of the specific measurement point. Such a measurement (or measurement-like) response can be represented mathematically as follows:
where the function
denotes the mathematical dependence of the measurement device on the model’s dependent variable(s), and where the quantity
denotes the Dirac-delta functional. The measurement’s location in phase space,
, may itself be afflicted by measurement (experimental) uncertainties. Hence, it is convenient to consider the components of
to be included among the components of the vector
of model parameters, even though
appears only in the definition of the response but does not appear in Equations (A1) and (A2), which mathematically define the physical model. Thus, the physical system is defined to comprise both the system’s computational model and the system’s response. In most cases, the coordinates
,
, will be independent (albeit uncertain) model parameters, in which case
and
.
The representations shown in Equations (A6)–(A8) indicate that model responses can be fundamentally analyzed by considering the following generic integral representation:
where
is a suitably differentiable nonlinear function of
and of
. It is important to note that the components of
include not only parameters that appear in the equations defining the computational model per se, i.e., in Equations (A1) and (A2), but also include parameters that specifically occur only in the definition of the response under consideration.
It is also important to note that the system’s definition domain, , in phase space is considered to be imprecisely known and subject to uncertainties in the components of the vector of model parameters . Therefore, the system domain’s boundary, , as well as the model response , will be affected by the boundary uncertainties that affect the endpoints . Such boundary uncertainties stem most often from manufacturing uncertainties.
Appendix B. The nth-CASAM-N for n = 1, 2, 3
The model and boundary parameters are considered to be uncertain quantities, with unknown true values. The nominal (or mean) parameter values are considered to be known, and these will differ from the true values by variations denoted as , where . Since the forward state functions are related to the model and boundary parameters through Equations (A1) and (A2), variations in the model and boundary parameters will cause corresponding variations around the nominal solution in the forward state functions. The variations and induce variations in the system’s response.
The first-order sensitivities of a model response , where , with respect to variations in the model parameters and state functions in a neighborhood around the nominal functions and parameter values , will exist if the first-order Gateaux (G) variation of the response will exist and will be linear in , which will occur if and only if the following two conditions are satisfied by :
- (i)
satisfies a weak Lipschitz condition at , namely,
- (ii)
satisfies the following condition for a scalar , where the symbol denotes the underlying field of scalars:
Numerical methods (e.g., Newton’s method and variants thereof) for solving Equations (A1) and (A2) also require the existence of the first-order G-derivatives of original model equations. Therefore, the conditions provided in Equations (A10) and (A11) are henceforth considered to be satisfied by the model responses and also by the operators underlying the physical system modeled by Equations (A1) and (A2). When the first-order G-differential
satisfies the conditions provided in Equations (A10) and (A11), it can be written as follows:
In Equation (A12), the direct-effect term
comprises only dependencies on
and is defined as follows:
where
denotes the partial G-derivatives of
with respect to
, evaluated at the nominal parameter values, and where the following definitions are used:
The notation on the left side of Equation (A14) comprises an implied transposed vector (since it represents an inner product), but the respective dagger (), which indicates transposition, has been omitted in order to keep the notation as simple as possible. Daggers indicating transposition will also be omitted in other inner products whenever possible without causing ambiguities.
The direct-effect term can be computed once the nominal values are available. The notation will be used to indicate that the quantity enclosed within the bracket is to be evaluated at the respective nominal parameter and state function values.
On the other hand, the quantity
in Equation (A12) comprises only variations in the state functions and is therefore called the indirect-effect term, with the following expression:
where
The notation on the left side of Equation (A17) comprises an implied transposed vector (since it represents an inner product), but the respective dagger (), which indicates transposition, has been omitted to simplify the notation.
The indirect-effect term induces variations in the response through the variations in the state functions, which are, in turn, caused by the parameter variations through the equations underlying the model. Evidently, the indirect-effect term can be quantified only after having determined the variations
in terms of the variations
. The first-order relationship between the vectors
and
is determined by solving the equations obtained by applying the definition of the G-differential to Equations (A1) and (A2), which yields the following equations:
In Equations (A18) and (A19), the superscript “(1)” indicates “first-level” and the various quantities that appear in these equations are defined as follows:
The system comprising Equations (A81) and (A82) is called the 1st-Level Variational Sensitivity System (1st-LVSS). In order to determine the solutions of the 1st-LVSS that would correspond to every parameter variation
,
, the 1st-LVSS would need to be solved
times, with distinct right sides for each
, thus requiring
large-scale computations. In other words, the actual form of the 1st-LVSS that would need to be solved in practice is as follows:
Appendix B.1. The 1st-CASAM-N for Computing Exactly and Efficiently the 1st-Order Sensitivities ,
The need for computing the vectors
,
, is eliminated by expressing the indirect-effect term defined in Equation (A16) in terms of the solutions of the 1st-Level Adjoint Sensitivity System (1st-LASS), the construction of which requires the introduction of adjoint operators. This is accomplished by introducing a real Hilbert space, denoted as
, endowed with an inner product of two vectors
and
, denoted as
and defined as follows:
where
, and where the dagger (
), which indicates transposition, has been omitted (to simplify the notation) in the representation of this scalar product.
Using the definition of the adjoint operator in
, the left side of Equation (A81) is transformed as follows:
where
denotes the associated bilinear concomitant evaluated on the space/time domain’s boundary
and where
is the operator adjoint to
. The symbol
indicates an adjoint operator. In certain situations, it might be computationally advantageous to include certain boundary components of
in the components of
. The first term on the right side of Equation (A21) is required to represent the indirect-effect term defined in Equation (A16) by imposing the following relationship:
The domain of
is determined by selecting appropriate adjoint boundary and/or initial conditions, which will be denoted in operator form as:
The above boundary conditions for are usually inhomogeneous, i.e., , and are obtained by imposing the following requirements: (i) They must be independent of unknown values of and , and (ii) the substitution of the boundary and/or initial conditions represented by Equations (A25) and (A29) in the expression of must cause all terms containing unknown values of to vanish. Constructing the adjoint initial and/or boundary conditions for as described above and implementing them together with the variational boundary and initial conditions represented by Equation (A25) in Equation (A27) reduces the bilinear concomitant to a quantity denoted as , which will contain boundary terms involving only known values of , , , and . Since is linear in , it can be expressed in the following form: .
The results obtained in Equations (A27) and (A28) are now replaced in Equation (A16) to obtain the following expression of the indirect-effect term as a function of
:
Replacing in Equation (A12) the result obtained in Equation (A30) together with the expression for the direct-effect term provided in Equation (A13) yields the following expression for the first G-differential of the response
:
where for each
, the quantity
denotes the first-order sensitivities of the response
with respect to the model parameters
and has the following expression:
As indicated by Equation (A32), each of the first-order sensitivities of the response with respect to the model parameters (including boundary and initial conditions) can be computed inexpensively after having obtained the function , using just quadrature formulas to evaluate the various inner products involving . The function is obtained by numerically solving Equations (A28) and (A29), which is the only large-scale computation needed to obtain all of the first-order sensitivities. Equations (A28) and (A29) are called the 1st-Level Adjoint Sensitivity System (1st-LASS), and its solution, , is called the first-level adjoint function. It is very important to note that the 1st-LASS is independent of parameter variation ,, and therefore needs to be solved only once, regardless of the number of model parameters under consideration. Furthermore, since Equation (A28) is linear in , solving it requires less computational effort than solving the original Equation (A1), which is nonlinear in . Finally, the function , which is defined and constructed by using the left side of Equation (A32), will contain Dirac-delta functions that are needed to represent the non-zero residual boundary terms contained in as a volume integral over . The function will be used to determine the second- and higher-order sensitivities of the response with respect to the model’s parameters.
Appendix B.2. The 2nd-CASAM-N for Computing Exactly and Efficiently the Second-Order Sensitivities, ,
2nd-CASAM-N determines the second-order sensitivities by treating them as the first-order sensitivities of the first-order sensitivities. Thus, the first-order G-differential of a first-order sensitivity
,
is given by the following expression:
where the indirect-effect term
comprises all dependencies on the vectors
and
of variations in the state functions
and
, respectively, and is defined as follows:
The functions
and
are obtained by solving the following 2nd-Level Variational Sensitivity System (2nd-LVSS), which comprises the concatenation of the 1st-LVSS with the G-differentiated 1st-LASS:
The argument “2” that appears in the list of arguments of the vector
and the variational vector
in Equation (A35) indicate that each of these vectors is a two-block column vector, each block comprising a column vector of dimension
, defined as follows:
To distinguish block vectors from block matrices, two bold capital letters have been used (and will henceforth be used) to denote block matrices, as in the case of the second-level variational matrix
. The second level is indicated by the superscript “(2)”. The argument
, which appears in the list of arguments of
, indicates that this matrix is a
-dimensional block matrix comprising four matrices, each of dimensions
, with the following structure:
The other quantities that appear in Equations (A35) and (A36) are two-block vectors with the same structure as
and are defined as follows:
The need to solve the 2nd-LVSS is circumvented by deriving an alternative expression for the indirect-effect term defined in Equation (A34), in which the function is replaced by a second-level adjoint function that is independent of variations in the model parameter and state functions and is the solution of a 2nd-Level Adjoint Sensitivity System (2nd-LASS) that is constructed by using the same principles as employed for deriving the 1st-LASS.
The 2nd-LASS is constructed in a Hilbert space, denoted as
, which comprises as elements block vectors of the same form as
, and is endowed with the following inner product of two vectors
and
:
The inner product defined in Equation (A48) is used to construct the following 2nd-Level Adjoint Sensitivity System (2nd-LASS) for the second-level adjoint function
for each
:
subject to boundary conditions represented as follows:
where
The matrix
comprises
block matrices, each of dimensions
, thus comprising a total of
components (or
elements), and is obtained from the following relation:
where the quantity
denotes the corresponding bilinear concomitant on the domain’s boundary, evaluated at the nominal values for the parameters and respective state functions. The second-level adjoint boundary/initial conditions represented by Equation (A50) are determined by requiring that (a) they be independent of unknown values of
, and (b) the substitution of the boundary and/or initial conditions represented by Equations (A50) and (A36) in the expression of
cause all terms containing unknown values of
to vanish. Using in Equation (A34) the relations defining 2nd-LASS, the 2nd-LVSS and the relation provided in Equation (A53) yields the following alternative expression for the indirect-effect term in terms of the second-level adjoint sensitivity function
:
where
denotes residual boundary terms that may not have vanished after having used the boundary and/or initial conditions represented by Equations (A36) and (A50). Replacing the expression obtained in Equation (A54) in Equation (A33) yields the following expression:
where the quantity
denotes the second-order sensitivity of the generic scalar-valued response
with respect to the parameters
and
computed at the nominal values of the parameters and respective state functions, and has the following expression for
:
If the 2nd-LASS is solved times, the second-order mixed sensitivities will be computed twice, in two different ways, in terms of two distinct second-level adjoint functions. Consequently, the symmetry property enjoyed by the second-order sensitivities provides an intrinsic (numerical) verification that the components of the second-level adjoint function , as well as the first-level adjoint function , are computed
accurately.
Appendix B.3. The 3rd-CASAM-N for Computing Exactly and Efficiently the 3rd-Order Sensitivities, ,
The third-order sensitivities are computed by considering them to be the first-order sensitivities of a second-order sensitivity. Thus, each of the second-order sensitivities
will be considered to be a model response that is assumed to satisfy the conditions stated in Equations (A10) and (A11) for each
, so that the first-order total G-differential of
will exist and will be linear in the variations
and
in a neighborhood around the nominal values of the parameters and the respective state functions. By definition, the first-order total G-differential of
, which will be denoted as
, is given by the
following expression:
where:
The indirect-effect term
can be computed after having determined the vectors
and
, which are the
solutions of the following 3rd-Level Variational Sensitivity System (3rd-LVSS):
where
Solving the 3rd-LVSS would require
large-scale computations, which is unrealistic for large-scale systems comprising many parameters. The 3rd-CASAM-N circumvents the need to solve the 3rd-LVSS by deriving an alternative expression for the indirect-effect term defined in Equation (A58), in which the function
is replaced by a third-level adjoint function that is independent of parameter variations. This third-level adjoint function is the solution of a 3rd-Level Adjoint Sensitivity System (3rd-LASS), which is constructed by applying the same principles as those used for constructing the 1st-LASS and the 2nd-LASS. The Hilbert space appropriate for constructing the 3rd-LASS, denoted as
, comprises as elements block vectors of the same form as
. Thus, a generic block vector in
, denoted as
, comprises four
-dimensional vector components of the form
,
, where each of these four components is a
-dimensional column vector. The inner product of two vectors
and
in the Hilbert space
will be denoted as
and defined as
follows:
The steps for constructing the 3rd-LASS are conceptually similar to those underlying the construction of the 1st-LASS and 2nd-LASS. The final expressions for the third-order sensitivity results are as follows:
where
denotes residual boundary terms that may have not vanished automatically and where the third-level adjoint function
is the
solution of the following 3rd-LASS:
where
The boundary conditions to be satisfied by each of the third-level adjoint functions
can be represented
in operator form as follows:
In component form, the total differential expressed by Equation (A70) has the following expression:
where the quantity
denotes the third-order sensitivity of the generic scalar-valued response
with respect to any three model parameters
,
,
, and has the following expression for
:
Appendix B.4. The Mathematical Framework of the 1st-CASAM-N in Block Matrix/Vector Notation
It is apparent from the formulations of the 2nd-CASAM-N and 3rd-CASAM-N that the general mathematical framework of the n
th-CASAM-N will involve block matrices and block vectors. The 1st-CASAM-N, however, only involves
-dimensional vectors and matrices but does not involve block matrices or block vectors. In order to facilitate the comparison of the formulas obtained when particularizing the general mathematical framework of the n
th-CASAM-N for the particular case
, it is convenient to recast the formulas obtained for the 1st-CASAM-N in block-vector and block-matrix form of the types used in the formulation of the 2nd- and 3rd-CASAM-N. Thus, Equations (A18) and (A19) take on the following forms in block-vector/-matrix notation:
In Equations (A81) and (A82), the superscript “(1)” indicates “first-level,”
(evidently), and
the various quantities that appear in these equations are defined as follows:
In block-vector/-matrix form, the 1st-LASS has the following expression:
where