Governing Equations

The equation set and solution procedure used by MAESTROeX has changed and improved over time. In this chapter, we outline the model equations and algorithmic options in the code. The latest published references for MAESTROeX are the multilevel paper [NonakaAlmgrenBell+10] and the more recent [FanNonakaAlmgren+19]. These two papers use the same model equations, however the more recent paper, in addition to retaining the original algorithmic capability of the previous code, includes an option to use a new, simplified temporal integration scheme. We distinguish between the two temporal integration strategies by referring to them as the “original temporal scheme” and “new temporal scheme”. In this description, we make frequent reference to papers I-IV and the multilevel paper (see § Introduction to MAESTROeX), which describe the developments of the original temporal scheme.

Summary of the MAESTROeX Equation Set

Here we summarize the equations solved by MAESTROeX. We refer the reader to papers I through IV for the derivation and motivation of the equation set. We take the standard equations of reacting, compressible flow, and recast the equation of state (EOS) as a divergence constraint on the velocity field. The resulting model is a series of evolution equations for mass, energy, and momentum, subject to an additional constraint on velocity, with a base state density and pressure linked via hydrostatic equilibrium:

(14)$$\frac{\partial \rho X_k}{\partial t}+\mathrm{\nabla }\cdot (\rho \mathbf{U}{X}_k)=\rho {\dot{\omega }}_k$$

(15)$$\frac{\partial (\rho h)}{\partial t}=-\mathrm{\nabla }\cdot (\rho h\mathbf{U})+\frac{D{p}_0}{Dt}+\rho H_{\mathrm{n}\mathrm{u}\mathrm{c}}+\rho H_{\mathrm{e}\mathrm{x}\mathrm{t}},$$

(16)$$\frac{\partial \mathbf{U}}{\partial t}+\mathbf{U}\cdot \mathrm{\nabla }\mathbf{U}+\frac{{\beta }_0}{\rho }\mathrm{\nabla }\left(\frac{p^{\prime }}{{\beta }_0}\right)=-\frac{{\rho }^{\prime }}{\rho }|g|{\mathbf{e}}_r$$

(17)$$\mathrm{\nabla }\cdot ({\beta }_0\mathbf{U})={\beta }_0(S-\frac{1}{{\bar{\mathrm{\Gamma }}}_1{p}_0}\frac{\partial p_0}{\partial t})$$

(18)$$\mathrm{\nabla }{p}_0=-{\rho }_0|g|{\mathbf{e}}_r$$

We discuss each of these equations in further detail below.

Lateral Average

A key concept in the MAESTROeX equation set and algorithm is the lateral average. The lateral average represents the average value of a quantity at a given radius in spherical simulations (or a given height in planar simulations). We denote the lateral average of a quantity with an overline, e.g., for any quantity $\varphi$, we denote the average of $\varphi$ over a layer at constant radius as $\bar{\varphi }$. For planar problems this routine is a trivial average of all the values at a given height. For spherical problems there is a novel interpolation routine we use to average 3D data representing a full spherical star into a 1D array representing the average. Details can be found in [NonakaAlmgrenBell+10] and [FanNonakaAlmgren+19].

For the velocity field, we can decompose the full velocity field into a base state velocity and a local velocity,

(19)$$\mathbf{U}=w_0(r,t){\mathbf{e}}_r+\tilde{\mathbf{U}}(\mathbf{x},t).$$

where $r$ is a 1D radial coordinate, $\mathbf{x}$ is a 3D Cartesian grid coordinate, and ${\mathbf{e}}_r$ is the unit vector in the outward radial direction. Note that $\overline{(\stackrel{~}{\mathbf{U}}\cdot {\mathbf{e}}_r)}=0$ and $w_0=\overline{(\mathbf{U}\cdot {\mathbf{e}}_r)}$. In other words, the base state velocity can be thought of as the lateral average of the outward radial velocity. For the velocity decompsotion, we do not use the same spatial averaging operators used for all other variables; instead we derive an analytic expression for the average expansion velocity and numerically integrate this expression to obtain $w_0$.

Mass

Conservation of mass gives the same continuity equation we have with compressible flow:

(20)$$\frac{\partial \rho }{\partial t}+\mathrm{\nabla }\cdot (\rho \mathbf{U})=0,$$

where $\rho$ is the total mass density. Additionally, we model the evolution of individual species that advect and react. The creation and destruction of the species is described by their create rate, ${\dot{\omega }}_k$, and the species are defined by their mass fractions, $X_k\equiv {\rho }_k/\rho$, giving

(21)$$\frac{\partial \rho X_k}{\partial t}+\mathrm{\nabla }\cdot (\rho \mathbf{U}{X}_k)=\rho {\dot{\omega }}_k$$

and

(22)$$\sum _k{X}_k=1$$

In the original temporal scheme, we need to model the evolution of a base state density, ${\rho }_0$. The governing equation can be obtained by laterally averaging the full continuity equation, giving:

(23)$$\frac{\partial {\rho }_0}{\partial t}=-\mathrm{\nabla }\cdot ({\rho }_0{w}_0{\mathbf{e}}_r),$$

Subtracting these two yields the evolution equation for the perturbational density, ${\rho }^{\prime }\equiv \rho -{\rho }_0$:

(24)$$\frac{\partial {\rho }^{\prime }}{\partial t}=-\mathbf{U}\cdot \mathrm{\nabla }{\rho }^{\prime }-{\rho }^{\prime }\mathrm{\nabla }\cdot \mathbf{U}-\mathrm{\nabla }\cdot \left({\rho }_0\tilde{\mathbf{U}}\right)$$

As first discussed in paper III and then refined in the multilevel paper, we capture the changes that can occur due to significant convective overturning by imposing the constraint that $\overline{{\rho }^{\prime }}=0$ for all time. This gives

(25)$$\frac{\partial \overline{{\rho }^{\prime }}}{\partial t}=-\mathrm{\nabla }\cdot ({\eta }_{\rho }{\mathbf{e}}_r).$$

where

(26)$${\eta }_{\rho }=\overline{({\rho }^{\prime }\mathbf{U}\cdot {\mathbf{e}}_r)}$$

In practice, we correct the drift by simply setting ${\rho }_0=\bar{\rho }$ after the advective update of $\rho$. However we still need to explicitly compute ${\eta }_{\rho }$ since it appears in other equations.

Energy

We model the evolution of specific enthalpy, $h$. Strictly speaking this is not necessary to close the system, but a user can enable the option to couple the energy with the rest of the system by using the enthalpy to define the temperature. The advantages of this coupling is an area of active research. The evolution equation is

(27)$$\frac{\partial (\rho h)}{\partial t}=-\mathrm{\nabla }\cdot (\rho h\mathbf{U})+\frac{D{p}_0}{Dt}+\rho H_{\mathrm{n}\mathrm{u}\mathrm{c}}+\rho H_{\mathrm{e}\mathrm{x}\mathrm{t}},$$

where $p_0$ is the 1D base state pressure, $H_{\mathrm{n}\mathrm{u}\mathrm{c}}$ and $H_{\mathrm{e}\mathrm{x}\mathrm{t}}$ are energy sources due to reactions and user-defined external heating.

When we are using thermal diffusion, there will be an additional term in the enthalpy equation (see § Thermal Diffusion Changes).

In the original temporal scheme, we utilized a base state enthlpy that effectively represents the average over a layer; its evolution equation can be found by laterally averaging Eq.27

(28)$$\frac{\partial (\rho h{)}_0}{\partial t}=-\mathrm{\nabla }\cdot [(\rho h{)}_0{w}_0{\mathbf{e}}_r]+\psi +\overline{\rho H_{\mathrm{n}\mathrm{u}\mathrm{c}}}+\overline{\rho H_{\mathrm{e}\mathrm{x}\mathrm{t}}}.$$

We will often expand $D{p}_0/Dt$ as

(29)$$\frac{D{p}_0}{Dt}=\psi +(\tilde{\mathbf{U}}\cdot {\mathbf{e}}_r)\frac{\partial p_0}{\partial r}$$

where we defined

(30)$$\psi \equiv \frac{\partial p_0}{\partial t}+w_0\frac{\partial p_0}{\partial r}$$

In paper III, we showed that for a plane-parallel atmosphere with constant gravity, $\psi ={\eta }_{\rho }g$

At times, we will define a temperature equation by writing $h=h(T,p,X_k)$ and differentiating:

(31)$$\frac{DT}{Dt}=\frac{1}{\rho c_p}\{(1-\rho h_p)[\psi +(\tilde{\mathbf{U}}\cdot {\mathbf{e}}_r)\frac{\partial p_0}{\partial r}]-\sum _k\rho {\xi }_k{\dot{\omega }}_k+\rho H_{\mathrm{n}\mathrm{u}\mathrm{c}}+\rho H_{\mathrm{e}\mathrm{x}\mathrm{t}}\}.$$

Subtracting it from the full enthalpy equation gives:

(32)$$\begin{array}{r}\begin{array}{rl}\frac{\partial (\rho h{)}^{\prime }}{\partial t}=& -\mathbf{U}\cdot \mathrm{\nabla }(\rho h{)}^{\prime }-(\rho h{)}^{\prime }\mathrm{\nabla }\cdot \mathbf{U}-\mathrm{\nabla }\cdot [(\rho h{)}_0\tilde{\mathbf{U}}]+\\ & \tilde{\mathbf{U}}\cdot \mathrm{\nabla }{p}_0+(\rho H_{\mathrm{n}\mathrm{u}\mathrm{c}}-\overline{\rho H_{\mathrm{n}\mathrm{u}\mathrm{c}}})+(\rho H_{\mathrm{e}\mathrm{x}\mathrm{t}}-\overline{\rho H_{\mathrm{e}\mathrm{x}\mathrm{t}}})\end{array}\end{array}$$

Base State

The stratified atmosphere is characterized by a one-dimensional time-dependent base state, defined by a base state density, ${\rho }_0$, and a base state pressure, $p_0$, in hydrostatic equilibrium:

(33)$$\mathrm{\nabla }{p}_0=-{\rho }_0|g|{\mathbf{e}}_r$$

The gravitational acceleration, $g$ is either constant or a point-mass with a $1/r^2$ dependence (see § Modifications for a 1/r^2 Plane-Parallel Basestate) for plane-parallel geometries, or a monopole constructed by integrating the base state density for spherical geometries.

For the time-dependence, we will define a base state velocity, $w_0$, which will adjust the base state from one hydrostatic equilibrium to another in response to heating.

For convenience, we define a base state enthalphy, $h_0$, as needed by laterally averaging the full enthalpy, $h$.

Base State Expansion

In practice, we calculate $w_0$ by integrating the one-dimensional divergence constraint. For a plane-parallel atmosphere, the evolution is:

(34)$$\frac{\partial w_0}{\partial r}=\bar{S}-\frac{1}{{\bar{\mathrm{\Gamma }}}_1{p}_0}{\eta }_{\rho }g$$

Then we define

(35)$$-\frac{{\beta }_0}{{\rho }_0}\frac{\partial ({\pi }_0/{\beta }_0)}{\partial r}=\frac{\partial w_0}{\partial t}+w_0\frac{\partial w_0}{\partial r},$$

once $w_0$ at the old and new times is known, and the advective term is computed explicitly. Then we can include this for completeness in the update for $\tilde{u}.$

Momentum

The compressible momentum equation (written in terms of velocity is):

(36)$$\rho \frac{\partial \mathbf{U}}{\partial t}+\rho \mathbf{U}\cdot \mathrm{\nabla }\mathbf{U}+\mathrm{\nabla }p=-\rho |g|{\mathbf{e}}_r$$

Subtracting off the equation of hydrostatic equilibrium, and defining the perturbational pressure (sometimes called the dynamic pressure) as $\pi \equiv p-p_0$, and perturbational density as ${\rho }^{\prime }\equiv \rho -{\rho }_0$, we have:

(37)$$\rho \frac{\partial \mathbf{U}}{\partial t}+\rho \mathbf{U}\cdot \mathrm{\nabla }\mathbf{U}+\mathrm{\nabla }\pi =-{\rho }^{\prime }|g|{\mathbf{e}}_r$$

or

(38)$$\frac{\partial \mathbf{U}}{\partial t}+\mathbf{U}\cdot \mathrm{\nabla }\mathbf{U}+\frac{1}{\rho }\mathrm{\nabla }\pi =-\frac{{\rho }^{\prime }}{\rho }|g|{\mathbf{e}}_r$$

This is the form of the momentum equation that we solved in papers I–IV and in the multilevel paper.

Several authors [KP12, VLB+13] explored the idea of energy conservation in a low Mach number system and found that an additional term (which can look like a buoyancy) is needed in the low Mach number formulation, yielding:

(39)$$\frac{\partial \mathbf{U}}{\partial t}+\mathbf{U}\cdot \mathrm{\nabla }\mathbf{U}+\frac{{\beta }_0}{\rho }\mathrm{\nabla }\left(\frac{p^{\prime }}{{\beta }_0}\right)=-\frac{{\rho }^{\prime }}{\rho }|g|{\mathbf{e}}_r$$

This is the form that we enforce in MAESTROeX, and the choice is controlled by use_alt_energy_fix.

We decompose the full velocity field into a base state velocity, $w_0$, that governs the base state dynamics, and a local velocity, $\tilde{\mathbf{U}}$, that governs the local dynamics, i.e.,

(40)$$\mathbf{U}=w_0(r,t){\mathbf{e}}_r+\tilde{\mathbf{U}}(\mathbf{x},t).$$

with $\overline{(\stackrel{~}{\mathbf{U}}\cdot {\mathbf{e}}_r)}=0$ and $w_0=\overline{(\mathbf{U}\cdot {\mathbf{e}}_r)}$—the motivation for this splitting was given in paper II. The velocity evolution equations are then

(41)$$\frac{\partial w_0}{\partial t}=-w_0\frac{\partial w_0}{\partial r}-\frac{{\beta }_0}{{\rho }_0}\frac{\partial ({\pi }_0/{\beta }_0)}{\partial r}$$

(42)$$\frac{\partial \tilde{\mathbf{U}}}{\partial t}=-(\tilde{\mathbf{U}}+w_0{\mathbf{e}}_r)\cdot \mathrm{\nabla }\tilde{\mathbf{U}}-(\tilde{\mathbf{U}}\cdot {\mathbf{e}}_r)\frac{\partial w_0}{\partial r}{\mathbf{e}}_r-\frac{{\beta }_0}{\rho }\mathrm{\nabla }\left(\frac{\pi }{{\beta }_0}\right)+\frac{{\beta }_0}{{\rho }_0}\frac{\partial ({\pi }_0/{\beta }_0)}{\partial r}{\mathbf{e}}_r-\frac{\rho -{\rho }_0}{\rho }g{\mathbf{e}}_r.$$

where ${\pi }_0$ is the base state component of the perturbational pressure. By laterally averaging to Eq.45, we obtain a divergence constraint for $w_0$:

(43)$$\mathrm{\nabla }\cdot ({\beta }_0{w}_0{\mathbf{e}}_r)={\beta }_0(\bar{S}-\frac{1}{{\bar{\mathrm{\Gamma }}}_1{p}_0}\frac{\partial p_0}{\partial t}).$$

The divergence constraint for $\tilde{\mathbf{U}}$ can be found by subtracting Eq.43 into Eq.45, resulting in

(44)$$\mathrm{\nabla }\cdot \left({\beta }_0\tilde{\mathbf{U}}\right)={\beta }_0(S-\bar{S}).$$

Velocity Constraint

The equation of state is cast into an elliptic constraint on the velocity field by differentiating $p_0(\rho ,s,X_k)$ along particle paths, giving:

(45)$$\mathrm{\nabla }\cdot ({\beta }_0\mathbf{U})={\beta }_0(S-\frac{1}{{\bar{\mathrm{\Gamma }}}_1{p}_0}\frac{\partial p_0}{\partial t})$$

where ${\beta }_0$ is a density-like variable that carries background stratification, defined as

(46)$${\beta }_0(r,t)={\rho }_0(0,t)\exp \left({\int }_0^r\frac{1}{{\bar{\mathrm{\Gamma }}}_1{p}_0}\frac{\partial p_0}{\partial r^{\prime }}d{r}^{\prime }\right),$$

and

(47)$$S=-\sigma \sum _k{\xi }_k{\dot{\omega }}_k+\frac{1}{\rho p_{\rho }}\sum _k{p}_{X_k}{\dot{\omega }}_k+\sigma H_{\mathrm{n}\mathrm{u}\mathrm{c}}+\sigma H_{\mathrm{e}\mathrm{x}\mathrm{t}}+\frac{\sigma }{\rho }\mathrm{\nabla }\cdot k_{\mathrm{t}\mathrm{h}}\mathrm{\nabla }T$$

where $p_{X_k}\equiv {\partial p/\partial X_k|}_{\rho ,T,X_{j,j\ne k}}$, ${\xi }_k\equiv {\partial h/\partial X_k|}_{p,T,X_{j,j\ne k}},p_{\rho }\equiv {\partial p/\partial \rho |}_{T,X_k}$, and $\sigma \equiv p_T/(\rho c_p{p}_{\rho })$, with $p_T\equiv {\partial p/\partial T|}_{\rho ,X_k}$ and $c_p\equiv {\partial h/\partial T|}_{p,X_k}$ is the specific heat at constant pressure. The last term is only present if we are using thermal diffusion (use_thermal_diffusion = T). In this term, $k_{\mathrm{t}\mathrm{h}}$ is the thermal conductivity.

In this constraint, ${\bar{\mathrm{\Gamma }}}_1$ is the lateral average of ${\mathrm{\Gamma }}_1\equiv d\log p/d\log \rho {|}_s$. Using the lateral average here makes it possible to cast the constraint as a divergence. [KP12] discuss the general case where we want to keep the local variations of ${\mathrm{\Gamma }}_1$ (and we explored this in paper III). We also look at this in § \Gamma_1 Variation Changes

Notation

Throughout the papers describing MAESTROeX, we’ve largely kept our notation consistent. The table below defines the frequently-used quantities and provides their units.

Table 1 Definition of symbols.

symbol	description	units
$c_p$	specific heat at constant pressure ($c_p\equiv \partial h/\partial T{|}_{p,X_k}$)	erg g ${}^{-1}$ K ${}^{-1}$
$f$	volume discrepancy factor ($0\le f\le 1$)	–
$g$	gravitational acceleration	cm s ${}^{-2}$
$h$	specific enthalpy	erg g ${}^{-1}$
$H_{\mathrm{e}\mathrm{x}\mathrm{t}}$	external heating energy generation rate	erg g ${}^{-1}$ s ${}^{-1}$
$H_{\mathrm{n}\mathrm{u}\mathrm{c}}$	nuclear energy generation rate	erg g ${}^{-1}$ s ${}^{-1}$
$h_p$	$h_p\equiv \partial h/\partial p{|}_{T,X_k}$	cm ${}^3$ g ${}^{-1}$
$k_{\mathrm{t}\mathrm{h}}$	thermal conductivity	erg cm ${}^{-1}$ s ${}^{-1}$ K ${}^{-1}$
$p_0$	base state pressure	erg cm ${}^{-3}$
$p_T$	$p_T\equiv \partial p/\partial T{|}_{\rho ,X_k}$	erg cm ${}^{-3}$ K ${}^{-1}$
$p_{X_k}$	$p_{X_k}\equiv \partial p/\partial X_k{|}_{p,T,X_{j,j\ne k}}$	erg cm ${}^{-3}$
$p_{\rho }$	$p_{\rho }\equiv \partial p/\partial \rho {|}_{T,X_k}$	erg g ${}^{-1}$
$q_k$	specific nuclear binding energy	erg g${}^{-1}$
$r$	radial coordinate (direction of gravity)	cm
$s$	specific entropy	erg g ${}^{-1}$ K ${}^{-1}$
$S$	source term to the divergence constraint	s ${}^{-1}$
$t$	time	s
$T$	temperature	K
$\mathbf{U}$	total velocity ($\mathbf{U}=\tilde{\mathbf{U}}+w_0{\mathbf{e}}_r$	cm s ${}^{-1}$
$\tilde{\mathbf{U}}$	local velocity	cm s ${}^{-1}$
${\tilde{\mathbf{U}}}^{\mathrm{ADV}}$	advective velocity (edge-centered)	cm s ${}^{-1}$
$w_0$	base state expansion velocity	cm s${}^{-1}$
$X_k$	mass fraction of the species ($\sum _k{X}_k=1$)	–
${\beta }_0$	coefficient to velocity in velocity constraint equation	g cm ${}^{-3}$
${\mathrm{\Gamma }}_1$	first adiabatic exponent (${\mathrm{\Gamma }}_1\equiv d\log p/d\log \rho {|}_s$)	–
${\eta }_{\rho }$	${\eta }_{\rho }\equiv \overline{({\rho }^{\prime }\mathbf{U}\cdot {\mathbf{e}}_r)}$	g cm ${}^{-2}$ s ${}^{-1}$
${\xi }_k$	${\xi }_k\equiv \partial h/\partial X_k{|}_{p,T,X_{j,j\ne k}}$	erg g ${}^{-1}$
$\pi$	dynamic pressure	erg cm ${}^{-3}$
${\pi }_0$	base state dynamic pressure	erg cm ${}^{-3}$
$\rho$	mass density	g cm ${}^{-3}$
${\rho }_0$	base state mass density	g cm ${}^{-3}$
${\rho }^{\prime }$	perturbational density (${\rho }^{\prime }=\rho -{\rho }_0$)	g cm ${}^{-3}$
$(\rho h{)}_0$	base state enthalpy density	erg cm ${}^{-3}$
$(\rho h{)}^{\prime }$	perturbational enthalpy density $[(\rho h{)}^{\prime }=\rho h-(\rho h{)}_0]$	erg cm ${}^{-3}$
$\sigma$	$\sigma \equiv p_T/(\rho c_p{p}_{\rho })$	erg ${}^{-1}$ g
$\psi$	$\psi \equiv D_0{p}_0/Dt=\partial p_0/\partial t+w_0\partial p_0/\partial r$	erg cm ${}^{-3}$
${\dot{\omega }}_k$	creation rate for species $k$ (${\dot{\omega }}_k\equiv D{X}_k/Dt$)	s ${}^{-1}$

[1]

Here we see an unfortunate conflict of notation between the compressible hydro community and the incompressible community. In papers on compressible hydrodynamics, $\mathbf{U}$ will usually mean the vector of conserved quantities. In incompressible / low speed papers, $\mathbf{U}$ will mean the velocity vector.

Time Advancement Algorithm Ingredients

The full time advancement algorithm is detailed in [NonakaAlmgrenBell+10] for the original algorithm and [FanNonakaAlmgren+19] for the new, simplified algorithm. We do not repeat that here.

The overall flow of the algorithm is depicted in the following flowcharts:

Fig. 1 A flowchart of the algorithm. The thermodynamic state variables, base state variables, and local velocity are indicated in each step. Red text indicates that quantity was updated during that step. The predictor-corrector steps are outlined by the dotted box. The blue text indicates state variables that are the same in Step 6 as they are in Step 2, i.e., they are unchanged by the predictor steps. The diffusion steps (4a and 8a) are optional, depending on use_thermal_diffusion.

Fig. 2 A flowchart for Steps 4 and 8. The thermodynamic state variables and base state variables are indicated in each step. Red text indicates that quantity was updated during that step. Note, for Step 4, the updated quantities should also have a $\star$ superscript, e.g., Step 8I defines $T^{(2)}$ while Step 4I defines $T^{(2),\star }$.

Here are some of the basic ingredients to the solver:

Definitions

Below we define operations that will be referenced in § Volume Discrepancy Changes.

React State$[{\rho }^{\mathrm{in}},(\rho h{)}^{\mathrm{in}},X_k^{\mathrm{in}},T^{\mathrm{in}},(\rho H_{\mathrm{e}\mathrm{x}\mathrm{t}}{)}^{\mathrm{in}},p_0^{\mathrm{in}}]\to [{\rho }^{\mathrm{out}},(\rho h{)}^{\mathrm{out}},X_k^{\mathrm{out}},T^{\mathrm{out}},(\rho {\dot{\omega }}_k{)}^{\mathrm{out}},(\rho H_{\mathrm{n}\mathrm{u}\mathrm{c}}{)}^{\mathrm{out}}]$ evolves the species and enthalpy due to reactions through $\mathrm{\Delta }t/2$ according to:

(48)$$\frac{d{X}_k}{dt}={\dot{\omega }}_k(\rho ,X_k,T);\frac{dT}{dt}=\frac{1}{c_p}(-\sum _k{\xi }_k{\dot{\omega }}_k+H_{\mathrm{n}\mathrm{u}\mathrm{c}}).$$

Here the temperature equation comes from Eq.31 with $D{p}_0/Dt=0$ for the burning part of the evolution.

Full details of the solution procedure can be found in Paper III. We then define:

(49)$$\begin{array}{r}\begin{array}{rlr}(\rho {\dot{\omega }}_k{)}^{\mathrm{out}}& =& \frac{{\rho }^{\mathrm{out}}(X_k^{\mathrm{out}}-X_k^{\mathrm{in}})}{\mathrm{\Delta }t/2},\\ (\rho h{)}^{\mathrm{out}}& =& (\rho h{)}^{\mathrm{in}}+\frac{\mathrm{\Delta }t}{2}(\rho H_{\mathrm{n}\mathrm{u}\mathrm{c}}{)}^{\mathrm{out}}+\frac{\mathrm{\Delta }t}{2}(\rho H_{\mathrm{e}\mathrm{x}\mathrm{t}}{)}^{\mathrm{in}}.\end{array}\end{array}$$

where the enthalpy update includes external heat sources $(\rho H_{\mathrm{e}\mathrm{x}\mathrm{t}}{)}^{\mathrm{in}}$. As introduced in Paper IV, we update the temperature using $T^{\mathrm{out}}=T({\rho }^{\mathrm{out}},h^{\mathrm{out}},X_k^{\mathrm{out}})$ for planar geometry or $T^{\mathrm{out}}=T({\rho }^{\mathrm{out}},p_0^{\mathrm{in}},X_k^{\mathrm{out}})$ for spherical geometry, with this behavior controlled by use_tfromp. Note that the density remains unchanged within React State, i.e., ${\rho }^{\mathrm{out}}={\rho }^{\mathrm{in}}$.

The source code for this operation can be found in react_state.f90.

Advect Base Density

$[{\rho }_0^{\mathrm{in}},w_0^{\mathrm{in}}]\to [{\rho }_0^{\mathrm{out}},{\rho }_0^{\mathrm{out},n+\frac{1}{2}}]$ is the process by which we update the base state density through $\mathrm{\Delta }t$ in time. We keep the time-centered edge states, ${\rho }_0^{\mathrm{out},n+\frac{1}{2}}$, since they are used later in discretization of ${\eta }_{\rho }$ for planar problems.

• planar:

  We discretize equation Eq.23 to compute the new base state density,

  (50)$${\rho }_{0,j}^{\mathrm{out}}={\rho }_{0,j}^{\mathrm{in}}-\frac{\mathrm{\Delta }t}{\mathrm{\Delta }r}[{\left({\rho }_0^{\mathrm{out},n+\frac{1}{2}}{w}_0^{\mathrm{in}}\right)}_{j+\frac{1}{2}}-{\left({\rho }_0^{\mathrm{out},n+\frac{1}{2}}{w}_0^{\mathrm{in}}\right)}_{j-\frac{1}{2}}].$$

  We compute the time-centered edge states, ${{\rho }_0}_{j\pm \frac{1}{2}}^{\mathrm{out},n+\frac{1}{2}}$, by discretizing an expanded form of Eq.23:

  (51)$$\frac{\partial {\rho }_0}{\partial t}+w_0\frac{\partial {\rho }_0}{\partial r}=-{\rho }_0\frac{\partial w_0}{\partial r},$$

  where the right hand side is used as the force term.

• spherical:

  The base state density update now includes the area factors in the divergences:

  (52)$${\rho }_{0,j}^{\mathrm{out}}={\rho }_{0,j}^{\mathrm{in}}-\frac{1}{r_j^2}\frac{\mathrm{\Delta }t}{\mathrm{\Delta }r}[{\left(r^2{\rho }_0^{\mathrm{out},n+\frac{1}{2}}{w}_0^{\mathrm{in}}\right)}_{j+\frac{1}{2}}-{\left(r^2{\rho }_0^{\mathrm{out},n+\frac{1}{2}}{w}_0^{\mathrm{in}}\right)}_{j-\frac{1}{2}}].$$

  In order to compute the time-centered edge states, an additional geometric term is added to the forcing, due to the spherical discretization of Eq.23:

  (53)$$\frac{\partial {\rho }_0}{\partial t}+w_0\frac{\partial {\rho }_0}{\partial r}=-{\rho }_0\frac{\partial w_0}{\partial r}-\frac{2{\rho }_0{w}_0}{r}.$$

Enforce HSE

$[p_0^{\mathrm{in}},{\rho }_0^{\mathrm{in}}]\to [p_0^{\mathrm{out}}]$ has replaced Advect Base Pressure from Paper III as the process by which we update the base state pressure. Rather than discretizing the evolution equation for $p_0$, we enforce hydrostatic equilibrium directly, which is numerically simpler and analytically equivalent. We first set $p_{0,j=0}^{\mathrm{out}}=p_{0,j=0}^{\mathrm{in}}$ and then update $p_0^{\mathrm{out}}$ using:

(54)$$p_{0,j+1}^{\mathrm{out}}=p_{0,j}^{\mathrm{out}}+\mathrm{\Delta }r{g}_{j+\frac{1}{2}}\frac{({\rho }_{0,j+1}^{\mathrm{in}}+{\rho }_{0,j}^{\mathrm{in}})}{2},$$

where $g=g({\rho }_0^{\mathrm{in}})$. As soon as ${\rho }_{0,j}^{\mathrm{in}}<{\rho }_{\mathrm{c}\mathrm{u}\mathrm{t}\mathrm{o}\mathrm{f}\mathrm{f}}$, we set $p_{0,j+1}^{\mathrm{out}}=p_{0,j}^{\mathrm{out}}$ for all remaining values of $j$. Then we compare $p_{0,j_{\mathrm{m}\mathrm{a}\mathrm{x}}}^{\mathrm{out}}$ with $p_{0,j_{\mathrm{m}\mathrm{a}\mathrm{x}}}^{\mathrm{in}}$ and offset every element in $p_0^{\mathrm{out}}$ so that $p_{0,j_{\mathrm{m}\mathrm{a}\mathrm{x}}}^{\mathrm{out}}=p_{0,j_{\mathrm{m}\mathrm{a}\mathrm{x}}}^{\mathrm{in}}$. We are effectively using the location where the ${\rho }_0^{\mathrm{in}}$ drops below ${\rho }_{\mathrm{c}\mathrm{u}\mathrm{t}\mathrm{o}\mathrm{f}\mathrm{f}}$ as the starting point for integration.

Advect Base Enthalpy

$[(\rho h{)}_0^{\mathrm{in}},w_0^{\mathrm{in}},{\psi }^{\mathrm{in}}]\to [(\rho h{)}_0^{\mathrm{out}}]$ is the process by which we update the base state enthalpy through $\mathrm{\Delta }t$ in time.

• planar:

  We discretize Eq.28, neglecting reaction source terms, to compute the new base state enthalpy,

  (55)$$(\rho h{)}_{0,j}^{\mathrm{out}}=(\rho h{)}_{0,j}^{\mathrm{in}}-\frac{\mathrm{\Delta }t}{\mathrm{\Delta }r}\{{[(\rho h{)}_0^{n+\frac{1}{2}}{w}_0^{\mathrm{in}}]}_{j+\frac{1}{2}}-{[(\rho h{)}_0^{n+\frac{1}{2}}{w}_0^{\mathrm{in}}]}_{j-\frac{1}{2}}\}+\mathrm{\Delta }t{\psi }_j^{\mathrm{in}}.$$

  We compute the time-centered edge states, $(\rho h{)}_0^{n+\frac{1}{2}}$, by discretizing an expanded form of Eq.28:

  (56)$$\frac{\partial (\rho h{)}_0}{\partial t}+w_0\frac{\partial (\rho h{)}_0}{\partial r}=-(\rho h{)}_0\frac{\partial w_0}{\partial r}+\psi .$$

• spherical:

  The base state enthalpy update now includes the area factors in the divergences:

  (57)$$\begin{array}{r}\begin{array}{rl}(\rho h{)}_{0,j}^{\mathrm{out}}& =(\rho h{)}_{0,j}^{\mathrm{in}}\\ & -\frac{1}{r_j^2}\frac{\mathrm{\Delta }t}{\mathrm{\Delta }r}\{{[r^2(\rho h{)}_0^{n+\frac{1}{2}}{w}_0^{\mathrm{in}}]}_{j+\frac{1}{2}}-{[r^2(\rho h{)}_0^{n+\frac{1}{2}}{w}_0^{\mathrm{in}}]}_{j-\frac{1}{2}}\}+\mathrm{\Delta }t{\psi }^{\mathrm{in},n+\frac{1}{2}}.\end{array}\end{array}$$

  In order to compute the time-centered edge states, an additional geometric term is added to the forcing, due to the spherical discretization of Eq.28:

  (58)$$\frac{\partial (\rho h{)}_0}{\partial t}+w_0\frac{\partial (\rho h{)}_0}{\partial r}=-(\rho h{)}_0\frac{\partial w_0}{\partial r}-\frac{2(\rho h{)}_0{w}_0}{r}+\psi .$$

Computing $w_0$

Here we describe the process by which we compute $w_0$. The arguments are different for planar and spherical geometries.

• planar:

  $[{\bar{S}}^{\mathrm{in}},{{\bar{\mathrm{\Gamma }}}_1}^{\mathrm{in}},p_0^{\mathrm{in}},{\psi }^{\mathrm{in}}]\to [w_0^{\mathrm{out}}]$:

  In Paper III, we showed that $\psi ={\eta }_{\rho }g$ for planar geometries, and derived derived Eq.34 as an alternate expression for eq:eq:w0 divergence. We discretize this as:

  (59)$$\frac{w_{0,j+\frac{1}{2}}^{\mathrm{out}}-w_{0,j-\frac{1}{2}}^{\mathrm{out}}}{\mathrm{\Delta }r}={({\bar{S}}^{\mathrm{in}}-\frac{1}{{{\bar{\mathrm{\Gamma }}}_1}^{\mathrm{in}}{p}_0^{\mathrm{in}}}{\psi }^{\mathrm{in}})}_j,$$

  with $w_{0,-\frac{1}{2}}=0$.

• spherical:

  $[{\bar{S}}^{\mathrm{in}},{{\bar{\mathrm{\Gamma }}}_1}^{\mathrm{in}},{\rho }_0^{\mathrm{in}},p_0^{\mathrm{in}},{{\eta }_{\rho }}^{\mathrm{in}}]\to [w_0^{\mathrm{out}}]$:

  We begin with Eq.43 written in spherical coordinates:

  (60)$$\frac{1}{r^2}\frac{\partial }{\partial r}\left(r^2{\beta }_0{w}_0\right)={\beta }_0(\bar{S}-\frac{1}{{\bar{\mathrm{\Gamma }}}_1{p}_0}\frac{\partial p_0}{\partial t}).$$

  We expand the spatial derivative and recall from Paper I that

  (61)$$\frac{1}{{\bar{\mathrm{\Gamma }}}_1{p}_0}\frac{\partial p_0}{\partial r}=\frac{1}{{\beta }_0}\frac{\partial {\beta }_0}{\partial r},$$

  giving:

  (62)$$\frac{1}{r^2}\frac{\partial }{\partial r}\left(r^2{w}_0\right)=\bar{S}-\frac{1}{{\bar{\mathrm{\Gamma }}}_1{p}_0}\underset{\psi }{\underbrace{(\frac{\partial p_0}{\partial t}+w_0\frac{\partial p_0}{\partial r})}}.$$

  We solve this equation for $w_0$ as described in Appendix B of the multilevel paper.

Volume Discrepancy Changes

Chapter [ch:volume] describes the reasoning behind the volume discrepancy term—a forcing term added to the constraint equation to bring us back to the equation of state. This addition of this term (enabled with dpdt_factor= T) modifies our equation set in the following way:

• In Step 2B, to compute $w_0$, we need to account for the volume discrepancy term by first defining $p_{\mathrm{E}\mathrm{O}\mathrm{S}}^n=\overline{p(\rho ,h,X_k{)}^n}$, and then using:

  (63)$$\frac{\partial w_0^{n+\frac{1}{2},\star }}{\partial r}=\overline{S^{n+\frac{1}{2},\star \star }}-\frac{1}{\overline{{\mathrm{\Gamma }}_1^n}{p}_0^n}{\psi }^{n-\frac{1}{2}}-\underset{\delta {\chi }_{w_0}}{\underbrace{\frac{f}{\overline{{\mathrm{\Gamma }}_1^n}{p}_0^n}\left(\frac{p_0^n-\overline{p_{\mathrm{E}\mathrm{O}\mathrm{S}}^n}}{\mathrm{\Delta }t}\right)}}.$$

• In Step 3, the MAC projection should account for the volume discrepancy term:

  (64)$$\mathrm{\nabla }\cdot \left({\beta }_0^n{\tilde{\mathbf{U}}}^{\mathrm{ADV},\star }\right)={\beta }_0^n[(S^{n+\frac{1}{2},\star \star }-\overline{S^{n+\frac{1}{2},\star \star }})+\underset{\delta \chi }{\underbrace{\frac{f}{{{\bar{\mathrm{\Gamma }}}_1}^n{p}_0^n}\left(\frac{p_{\mathrm{E}\mathrm{O}\mathrm{S}}^n-\overline{p_{\mathrm{E}\mathrm{O}\mathrm{S}}^n}}{\mathrm{\Delta }{t}^n}\right)}}].$$

• In Step 6B, to compute $w_0$, we need to account for the volume discrepancy term by first defining $p_{\mathrm{E}\mathrm{O}\mathrm{S}}^{n+1,\star }=p(\rho ,h,X_k{)}^{n+1,\star }$, $\overline{{\mathrm{\Gamma }}_1^{n+\frac{1}{2},\star }}=(\overline{{\mathrm{\Gamma }}_1^n}+\overline{{\mathrm{\Gamma }}_1^{n+1,\star }})/2$, and $p^{n+\frac{1}{2},\star }=(p^n+p^{n+1,\star })/2$, and then using:

  (65)$$\frac{\partial w_0^{n+\frac{1}{2}}}{\partial r}=\overline{S^{n+\frac{1}{2},\star }}-\frac{1}{\overline{{\mathrm{\Gamma }}_1^{n+\frac{1}{2},\star }}{p}_0^{n+\frac{1}{2},\star }}{\psi }^{n+\frac{1}{2},\star }-\frac{f}{\overline{{\mathrm{\Gamma }}_1^{n+1,\star }}{p}_0^{n+1,\star }}\left(\frac{p_0^{n+1,\star }-\overline{p_{\mathrm{E}\mathrm{O}\mathrm{S}}^{n+1,\star }}}{\mathrm{\Delta }t}\right)-\delta {\chi }_{w_0}$$

• In Step 7, the MAC projection should account for the volume discrepancy term:

  (66)$$\mathrm{\nabla }\cdot \left({\beta }_0^{n+\frac{1}{2},\star }{\tilde{\mathbf{U}}}^{\mathrm{ADV}}\right)={\beta }_0^{n+\frac{1}{2},\star }[(S^{n+\frac{1}{2},\star }-\overline{S^{n+\frac{1}{2},\star }})+\frac{f}{\overline{{\mathrm{\Gamma }}_1^{n+1,\star }}{p}_0^{n+1,\star }}\left(\frac{p_{\mathrm{E}\mathrm{O}\mathrm{S}}^{n+1,\star }-\overline{p_{\mathrm{E}\mathrm{O}\mathrm{S}}^{n+1,\star }}}{\mathrm{\Delta }{t}^n}\right)+\delta \chi ],$$

  where $p(\rho ,h,X_k{)}^{n+\frac{1}{2},\star }=[p(\rho ,h,X_k{)}^n+p(\rho ,h,X_k{)}^{n+1,\star }]/2$.

• In Step 11, the approximate projection should account for the volume discrepancy term:

  (67)$$\mathrm{\nabla }\cdot \left({\beta }_0^{n+\frac{1}{2}}{\tilde{\mathbf{U}}}^{n+1}\right)={\beta }_0^{n+\frac{1}{2}}\{(S^{n+1}-\overline{S^{n+1}})+\frac{f}{\overline{{\mathrm{\Gamma }}_1^{n+1}}{p}_0^{n+1}}\left[\frac{p(\rho ,h,X_k{)}^{n+1}-\overline{p(\rho ,h,X_k{)}^{n+1}}}{\mathrm{\Delta }{t}^n}\right]\}.$$

${\mathrm{\Gamma }}_1$ Variation Changes

The constraint we derive from requiring the pressure to be close to the background hydrostatic pressure takes the form:

(68)$$\mathrm{\nabla }\cdot \mathbf{U}+\frac{1}{{\mathrm{\Gamma }}_1{p}_0}\frac{D{p}_0}{Dt}=S.$$

The default MAESTROeX algorithm replaces ${\mathrm{\Gamma }}_1$ with ${\bar{\mathrm{\Gamma }}}_1$, allowing us to write this as a divergence constraint. In paper III, we explored the effects of localized variations in ${\mathrm{\Gamma }}_1$ by writing ${\mathrm{\Gamma }}_1={\bar{\mathrm{\Gamma }}}_1+\delta {\mathrm{\Gamma }}_1$. This gives us:

(69)$$\mathrm{\nabla }\cdot \mathbf{U}+\frac{1}{({\bar{\mathrm{\Gamma }}}_1+\delta {\mathrm{\Gamma }}_1)p_0}\mathbf{U}\cdot \mathrm{\nabla }{p}_0=S-\frac{1}{({\bar{\mathrm{\Gamma }}}_1+\delta {\mathrm{\Gamma }}_1)p_0}\frac{\partial p_0}{\partial t}.$$

Assuming that $\delta {\mathrm{\Gamma }}_1\ll {\bar{\mathrm{\Gamma }}}_1$, we then have

(70)$$\begin{array}{r}\mathrm{\nabla }\cdot \mathbf{U}+\frac{1}{{\bar{\mathrm{\Gamma }}}_1{p}_0}\mathbf{U}\cdot \mathrm{\nabla }{p}_0[1-\frac{\delta {\mathrm{\Gamma }}_1}{{\bar{\mathrm{\Gamma }}}_1}+\frac{(\delta {\mathrm{\Gamma }}_1{)}^2}{{{\bar{\mathrm{\Gamma }}}_1}^2}]=S-\frac{1}{{\bar{\mathrm{\Gamma }}}_1{p}_0}\frac{\partial p_0}{\partial t}[1-\frac{\delta {\mathrm{\Gamma }}_1}{{\bar{\mathrm{\Gamma }}}_1}+\frac{(\delta {\mathrm{\Gamma }}_1{)}^2}{{{\bar{\mathrm{\Gamma }}}_1}^2}]\end{array}$$

Grouping by order of the correction, we have

(71)$$\mathrm{\nabla }\cdot \mathbf{U}+\frac{1}{{\bar{\mathrm{\Gamma }}}_1{p}_0}\mathbf{U}\cdot \mathrm{\nabla }{p}_0=S-\frac{1}{{\bar{\mathrm{\Gamma }}}_1{p}_0}\frac{\partial p_0}{\partial t}+\underset{\text{first order corrections}}{\underbrace{\frac{\delta {\mathrm{\Gamma }}_1}{{{\bar{\mathrm{\Gamma }}}_1}^2{p}_0}[\frac{\partial p_0}{\partial t}+\mathbf{U}\cdot \mathrm{\nabla }{p}_0]}}-\underset{\text{second order corrections}}{\underbrace{\frac{(\delta {\mathrm{\Gamma }}_1{)}^2}{{{\bar{\mathrm{\Gamma }}}_1}^3{p}_0}[\frac{\partial p_0}{\partial t}+\mathbf{U}\cdot \mathrm{\nabla }{p}_0]}}$$

Keeping to First Order in $\delta {\mathrm{\Gamma }}_1$

The base state evolution equation is the average of Eq.71 over a layer

(72)$$\mathrm{\nabla }\cdot w_0{\mathbf{e}}_r+\frac{1}{{\bar{\mathrm{\Gamma }}}_1{p}_0}{w}_0{\mathbf{e}}_r\cdot \mathrm{\nabla }{p}_0=\bar{S}-\frac{1}{{\bar{\mathrm{\Gamma }}}_1{p}_0}\frac{\partial p_0}{\partial t}+\stackrel{―}{\left(\frac{\delta {\mathrm{\Gamma }}_1}{{{\bar{\mathrm{\Gamma }}}_1}^2{p}_0}\tilde{\mathbf{U}}\cdot \mathrm{\nabla }{p}_0\right)}.$$

where we see that the $[\delta {\mathrm{\Gamma }}_1/({\mathrm{\Gamma }}_1^2{p}_0)]\partial p_0/\partial t$ terms averages to zero, since the average of $\delta {\mathrm{\Gamma }}_1$ term is zero. Subtracting this from Eq.71, we have

(73)$$\mathrm{\nabla }\cdot \tilde{\mathbf{U}}+\frac{1}{{\bar{\mathrm{\Gamma }}}_1{p}_0}\tilde{\mathbf{U}}\cdot \mathrm{\nabla }{p}_0=S-\bar{S}+\frac{\delta {\mathrm{\Gamma }}_1}{{{\bar{\mathrm{\Gamma }}}_1}^2{p}_0}(\psi +\tilde{\mathbf{U}}\cdot \mathrm{\nabla }{p}_0)-\stackrel{―}{\left(\frac{\delta {\mathrm{\Gamma }}_1}{{{\bar{\mathrm{\Gamma }}}_1}^2{p}_0}\tilde{\mathbf{U}}\cdot \mathrm{\nabla }{p}_0\right)}.$$

These can be written more compactly as:

(74)$$\frac{\partial w_0}{\partial r}=\bar{S}-\frac{1}{{\bar{\mathrm{\Gamma }}}_1{p}_0}\psi +\stackrel{―}{\left(\frac{\delta {\mathrm{\Gamma }}_1}{{{\bar{\mathrm{\Gamma }}}_1}^2{p}_0}\tilde{\mathbf{U}}\cdot \mathrm{\nabla }{p}_0\right)},$$

for plane-parallel geometries (analogous to Eq.34, and

(75)$$\mathrm{\nabla }\cdot ({\beta }_0\tilde{\mathbf{U}})={\beta }_0[S-\bar{S}+\frac{\delta {\mathrm{\Gamma }}_1}{{{\bar{\mathrm{\Gamma }}}_1}^2{p}_0}\psi +\frac{\delta {\mathrm{\Gamma }}_1}{{{\bar{\mathrm{\Gamma }}}_1}^2{p}_0}\tilde{\mathbf{U}}\cdot \mathrm{\nabla }{p}_0-\stackrel{―}{\left(\frac{\delta {\mathrm{\Gamma }}_1}{{{\bar{\mathrm{\Gamma }}}_1}^2{p}_0}\tilde{\mathbf{U}}\cdot \mathrm{\nabla }{p}_0\right)}]$$

This constraint is not in a form that can be projected. To solve this form, we need to use a lagged $\tilde{\mathbf{U}}$ in the righthand side.

This change comes into MAESTROeX in a variety of steps, summarized here. To enable this portion of the algorithm, set use_delta_gamma1_term = T.

• In Step 3, we are doing the “predictor” portion of the MAESTROeX algorithm, getting the MAC velocity that satisfies the constraint, so we do not try to incorporate the $\delta {\mathrm{\Gamma }}_1$ effect. We set all the $\delta {\mathrm{\Gamma }}_1$ terms in Eq.75 to zero.

• In Step 6, we are computing the new time-centered source, $S^{n+\frac{1}{2},\star }$ and the base state velocity, $w_0^{n+\frac{1}{2}}$. Now we can incorporate the $\delta {\mathrm{\Gamma }}_1$ effect. First we construct:

  (76)$$\frac{\delta {\mathrm{\Gamma }}_1}{{\mathrm{\Gamma }}_1^2{p}_0}\tilde{\mathbf{U}}\cdot \mathrm{\nabla }{p}_0\approx \frac{{\mathrm{\Gamma }}_1^{n+1,\star }-\overline{{\mathrm{\Gamma }}_1^{n+1,\star }}}{{\overline{{\mathrm{\Gamma }}_1^{n+1,\star }}}^2}\frac{1}{p_0^n}{\tilde{\mathbf{U}}}^n\cdot \mathrm{\nabla }{p}_0^n$$

  Then we call average to construct the lateral average of this

  (77)$$\stackrel{―}{\frac{\delta {\mathrm{\Gamma }}_1}{{\mathrm{\Gamma }}_1^2{p}_0}\tilde{\mathbf{U}}\cdot \mathrm{\nabla }{p}_0}=\mathbf{Avg}\left(\frac{{\mathrm{\Gamma }}_1^{n+1,\star }-\overline{{\mathrm{\Gamma }}_1^{n+1,\star }}}{{\overline{{\mathrm{\Gamma }}_1^{n+1,\star }}}^2}\frac{1}{p_0^n}{\tilde{\mathbf{U}}}^n\cdot \mathrm{\nabla }{p}_0^n\right)$$

  Since the average of this is needed in advancing $w_0$, we modify $\bar{S}$ to include this average:

  (78)$$\overline{S^{n+\frac{1}{2},\star }}\leftarrow \overline{S^{n+\frac{1}{2},\star }}+\stackrel{―}{\frac{\delta {\mathrm{\Gamma }}_1}{{\mathrm{\Gamma }}_1^2{p}_0}\tilde{\mathbf{U}}\cdot \mathrm{\nabla }{p}_0}$$

• In Step 7, we now include the $\delta {\mathrm{\Gamma }}_1$ term in the righthand side for the constraint by solving:

  (79)$$\mathrm{\nabla }\cdot \left({\beta }_0^{n+\frac{1}{2},\star }{\tilde{\mathbf{U}}}^{\mathrm{ADV}}\right)={\beta }_0^{n+\frac{1}{2},\star }(S^{n+\frac{1}{2},\star }-\overline{S^{n+\frac{1}{2},\star }}+\frac{{\mathrm{\Gamma }}_1^{n+1,\star }-\overline{{\mathrm{\Gamma }}_1^{n+1,\star }}}{{\overline{{\mathrm{\Gamma }}_1^{n+1,\star }}}^2}\frac{1}{p_0^n}({\psi }^{n+\frac{1}{2},\star }+{\tilde{\mathbf{U}}}^n\cdot \mathrm{\nabla }{p}_0^n))$$

  We note that this includes the average of the correction term as shown in Eq.75 because we modified $\overline{S}$ to include this already.

• In Step 10, we do a construction much like that done in Step 6, but with the time-centerings of some of the quantities changed. First we construct:

  (80)$$\frac{\delta {\mathrm{\Gamma }}_1}{{\mathrm{\Gamma }}_1^2{p}_0}\tilde{\mathbf{U}}\cdot \mathrm{\nabla }{p}_0\approx \frac{{\mathrm{\Gamma }}_1^{n+1}-\overline{{\mathrm{\Gamma }}_1^{n+1}}}{{\overline{{\mathrm{\Gamma }}_1^{n+1}}}^2}\frac{1}{p_0^{n+1}}{\tilde{\mathbf{U}}}^n\cdot \mathrm{\nabla }{p}_0^{n+1}$$

  Then we call average to construct the lateral average of this

  (81)$$\stackrel{―}{\frac{\delta {\mathrm{\Gamma }}_1}{{\mathrm{\Gamma }}_1^2{p}_0}\tilde{\mathbf{U}}\cdot \mathrm{\nabla }{p}_0}=\mathbf{Avg}\left(\frac{{\mathrm{\Gamma }}_1^{n+1}-\overline{{\mathrm{\Gamma }}_1^{n+1}}}{{\overline{{\mathrm{\Gamma }}_1^{n+1}}}^2}\frac{1}{p_0^{n+1}}{\tilde{\mathbf{U}}}^n\cdot \mathrm{\nabla }{p}_0^{n+1}\right)$$

  Again we modify $\bar{S}$ to include this average:

  (82)$$\overline{S^{n+1}}\leftarrow \overline{S^{n+1}}+\stackrel{―}{\frac{\delta {\mathrm{\Gamma }}_1}{{\mathrm{\Gamma }}_1^2{p}_0}\tilde{\mathbf{U}}\cdot \mathrm{\nabla }{p}_0}$$

• In Step 11, we modify the source of the constraint to include the $\delta {\mathrm{\Gamma }}_1$ information. In particular, we solve:

  (83)$$\mathrm{\nabla }\cdot \left({\beta }_0^{n+\frac{1}{2}}{\tilde{\mathbf{U}}}^{n+1}\right)={\beta }_0^{n+\frac{1}{2}}(S^{n+1}-\overline{S^{n+1}}+\frac{{\mathrm{\Gamma }}_1^{n+1}-\overline{{\mathrm{\Gamma }}_1^{n+1}}}{{\overline{{\mathrm{\Gamma }}_1^{n+1}}}^2}\frac{1}{p_0^{n+1}}({\psi }^{n+\frac{1}{2}}+{\tilde{\mathbf{U}}}^n\cdot \mathrm{\nabla }{p}_0^{n+1}))$$

Thermal Diffusion Changes

Thermal diffusion was introduced in the XRB [MaloneNonakaAlmgren+11]. This introduces a new term to $S$ as well as the enthalpy equation. Treating the enthalpy equation now requires a parabolic solve. We describe that process here.

Immediately after Step 4H, diffuse the enthalpy through a time interval of $\mathrm{\Delta }t$. First, define $(\rho h{)}^{(1a),\star }=(\rho h{)}^{(2),\star }$. We recompute $(\rho h{)}^{(2),\star }$ to account for thermal diffusion. Here we begin with the enthalpy equation, but consider only the diffusion terms,

(84)$$\frac{\partial (\rho h)}{\partial t}=\mathrm{\nabla }\cdot k_{\mathrm{t}\mathrm{h}}\mathrm{\nabla }T.$$

We can recast this as an enthalpy-diffusion equation by applying the chain-rule to $h(p_0,T,X_k)$,

(85)$$\mathrm{\nabla }h=h_p\mathrm{\nabla }{p}_0+c_p\mathrm{\nabla }T+\sum _k{\xi }_k\mathrm{\nabla }{X}_k,$$

giving

(86)$$\frac{\partial (\rho h)}{\partial t}=\mathrm{\nabla }\cdot \frac{k_{\mathrm{t}\mathrm{h}}}{c_p}\mathrm{\nabla }h-\sum _k\mathrm{\nabla }\cdot \frac{{\xi }_k{k}_{\mathrm{t}\mathrm{h}}}{c_p}\mathrm{\nabla }{X}_k-\mathrm{\nabla }\cdot \frac{h_p{k}_{\mathrm{t}\mathrm{h}}}{c_p}\mathrm{\nabla }{p}_0.$$

Compute ${k_{\mathrm{t}\mathrm{h}}}^{(1)},c_p^{(1)}$, and ${\xi }_k^{(1)}$ from ${\rho }^{(1)},T^{(1)}$, and $X_k^{(1)}$ as inputs to the equation of state. The update is given by

(87)$$\begin{array}{r}\begin{array}{rl}(\rho h{)}^{(2),\star }& =(\rho h{)}^{(1a),\star }+\frac{\mathrm{\Delta }t}{2}\mathrm{\nabla }\cdot (\frac{{k_{\mathrm{t}\mathrm{h}}}^{(1)}}{c_p^{(1)}}\mathrm{\nabla }{h}^{(2),\star }+\frac{{k_{\mathrm{t}\mathrm{h}}}^{(1)}}{c_p^{(1)}}\mathrm{\nabla }{h}^{(1)})\\ & -\frac{\mathrm{\Delta }t}{2}\sum _k\mathrm{\nabla }\cdot (\frac{{\xi }_k^{(1)}{k_{\mathrm{t}\mathrm{h}}}^{(1)}}{c_p^{(1)}}\mathrm{\nabla }{X}_k^{(2),\star }+\frac{{\xi }_k^{(1)}{k_{\mathrm{t}\mathrm{h}}}^{(1)}}{c_p^{(1)}}\mathrm{\nabla }{X}_k^{(1)})\\ & -\frac{\mathrm{\Delta }t}{2}\mathrm{\nabla }\cdot (\frac{h_p^{(1)}{k_{\mathrm{t}\mathrm{h}}}^{(1)}}{c_p^{(1)}}\mathrm{\nabla }{p}_0^{n+1,\star }+\frac{h_p^{(1)}{k_{\mathrm{t}\mathrm{h}}}^{(1)}}{c_p^{(1)}}\mathrm{\nabla }{p}_0^n),\end{array}\end{array}$$

which is numerically implemented as a diffusion equation for $h^{(2),\star }$,

(88)$$\begin{array}{r}\begin{array}{rl}({\rho }^{(2),\star }-\frac{\mathrm{\Delta }t}{2}\mathrm{\nabla }\cdot \frac{{k_{\mathrm{t}\mathrm{h}}}^{(1)}}{c_p^{(1)}}\mathrm{\nabla })h^{(2),\star }& =(\rho h{)}^{(1a),\star }+\frac{\mathrm{\Delta }t}{2}\mathrm{\nabla }\cdot \frac{{k_{\mathrm{t}\mathrm{h}}}^{(1)}}{c_p^{(1)}}\mathrm{\nabla }{h}^{(1)}\\ & -\frac{\mathrm{\Delta }t}{2}\sum _k\mathrm{\nabla }\cdot (\frac{{\xi }_k^{(1)}{k_{\mathrm{t}\mathrm{h}}}^{(1)}}{c_p^{(1)}}\mathrm{\nabla }{X}_k^{(2),\star }+\frac{{\xi }_k^{(1)}{k_{\mathrm{t}\mathrm{h}}}^{(1)}}{c_p^{(1)}}\mathrm{\nabla }{X}_k^{(1)})\\ & -\frac{\mathrm{\Delta }t}{2}\mathrm{\nabla }\cdot (\frac{h_p^{(1)}{k_{\mathrm{t}\mathrm{h}}}^{(1)}}{c_p^{(1)}}\mathrm{\nabla }{p}_0^{n+1,\star }+\frac{h_p^{(1)}{k_{\mathrm{t}\mathrm{h}}}^{(1)}}{c_p^{(1)}}\mathrm{\nabla }{p}_0^n),\end{array}\end{array}$$

Immediately after Step 8H, diffuse the enthalpy through a time interval of $\mathrm{\Delta }t$. First, define $(\rho h{)}^{(1a)}=(\rho h{)}^{(2)}$. We recompute $(\rho h{)}^{(2)}$ to account for thermal diffusion. Compute ${k_{\mathrm{t}\mathrm{h}}}^{(2),\star },c_p^{(2),\star }$, and ${\xi }_k^{(2),\star }$, from ${\rho }^{(2),\star },T^{(2),\star }$, and $X_k^{(2),\star }$ as inputs to the equation of state. The update is given by

(89)$$\begin{array}{r}\begin{array}{rl}(\rho h{)}^{(2)}& =(\rho h{)}^{(1a)}+\frac{\mathrm{\Delta }t}{2}\mathrm{\nabla }\cdot (\frac{{k_{\mathrm{t}\mathrm{h}}}^{(2),\star }}{c_p^{(2),\star }}\mathrm{\nabla }{h}^{(2)}+\frac{{k_{\mathrm{t}\mathrm{h}}}^{(1)}}{c_p^{(1)}}\mathrm{\nabla }{h}^{(1)})\\ & -\frac{\mathrm{\Delta }t}{2}\sum _k\mathrm{\nabla }\cdot (\frac{{\xi }_k^{(2),\star }{k_{\mathrm{t}\mathrm{h}}}^{(2),\star }}{c_p^{(2),\star }}\mathrm{\nabla }{X}_k^{(2)}+\frac{{\xi }_k^{(1)}{k_{\mathrm{t}\mathrm{h}}}^{(1)}}{c_p^{(1)}}\mathrm{\nabla }{X}_k^{(1)})\\ & -\frac{\mathrm{\Delta }t}{2}\mathrm{\nabla }\cdot (\frac{h_p^{(2),\star }{k_{\mathrm{t}\mathrm{h}}}^{(2),\star }}{c_p^{(2),\star }}\mathrm{\nabla }{p}_0^{n+1}+\frac{h_p^{(1)}{k_{\mathrm{t}\mathrm{h}}}^{(1)}}{c_p^{(1)}}\mathrm{\nabla }{p}_0^n),\end{array}\end{array}$$

which is numerically implemented as a diffusion equation for $h^{(2)}$.

(90)$$\begin{array}{r}\begin{array}{rl}({\rho }^{(2)}-\frac{\mathrm{\Delta }t}{2}\mathrm{\nabla }\cdot \frac{{k_{\mathrm{t}\mathrm{h}}}^{(2),\star }}{c_p^{(2),\star }}\mathrm{\nabla })h^{(2)}& =(\rho h{)}^{(1a)}+\frac{\mathrm{\Delta }t}{2}\mathrm{\nabla }\cdot \frac{{k_{\mathrm{t}\mathrm{h}}}^{(1)}}{c_p^{(1)}}\mathrm{\nabla }{h}^{(1)}\\ & -\frac{\mathrm{\Delta }t}{2}\sum _k\mathrm{\nabla }\cdot (\frac{{\xi }_k^{(2),\star }{k_{\mathrm{t}\mathrm{h}}}^{(2),\star }}{c_p^{(2),\star }}\mathrm{\nabla }{X}_k^{(2)}+\frac{{\xi }_k^{(1)}{k_{\mathrm{t}\mathrm{h}}}^{(1)}}{c_p^{(1)}}\mathrm{\nabla }{X}_k^{(1)})\\ & -\frac{\mathrm{\Delta }t}{2}\mathrm{\nabla }\cdot (\frac{h_p^{(2),\star }{k_{\mathrm{t}\mathrm{h}}}^{(2),\star }}{c_p^{(2),\star }}\mathrm{\nabla }{p}_0^{n+1}+\frac{h_p^{(1)}{k_{\mathrm{t}\mathrm{h}}}^{(1)}}{c_p^{(1)}}\mathrm{\nabla }{p}_0^n),\end{array}\end{array}$$

Initialization

We start each calculation with user-specified initial values for $\rho$, $X_k$ and $T,$ as well as an initial background state. In order for the low Mach number assumption to hold, the initial data must be thermodynamically consistent with the initial background state. In addition, the initial velocity field must satisfy an initial approximation to the divergence constraint. We use an iterative procedure to compute both an initial right-hand-side for the constraint equation and an initial velocity field that satisfies the constraint. The user specifies the number of iterations, $N_{\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{s}}^S,$ in this first step of the initialization procedure.

The initial perturbational pressure also needs to be determined for use in Steps 3, 7, and 11. This is done through a second iterative procedure which follows the time advancement algorithm as described in Steps 1-11 in §2. The user specifies the number of iterations, $N_{\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{s}}^{\pi },$ in this second step of the initialization procedure. The details for both iterations are given below.

First, we need to construct approximations to $S^0,w_0^{-\frac{1}{2}},\mathrm{\Delta }{t}^0$, and ${\mathbf{U}}^0$. Start with initial data $X_k^{\mathrm{init}},{\rho }^{\mathrm{init}},$ $T^{\mathrm{init}},$ an initial base state, and an initial guess for the velocity, ${\mathbf{U}}^{\mathrm{init}}$. Set $w_0^0=0$ as an initial approximation. Use the equation of state to determine $(\rho h{)}^{\mathrm{init}}$. Compute ${\beta }_0^0$ as a function of the initial data. The next part of the initialization process proceeds as follows.

1. Initial Projection: if do_initial_projection = T, then we first project the velocity field with $\rho =1$ and ${\beta }_0^0$. The initial projection does not see reactions or external heating, and thus we set $\dot{\omega }=H_{\mathrm{n}\mathrm{u}\mathrm{c}}=H_{\mathrm{e}\mathrm{x}\mathrm{t}}=0$ in $S$. The reason for ignoring reactions and heating is that we need some kind of time scale over which to compute the effect of reactions, but we first need an estimate of the velocity field in order to derive the time step that will be used as a time scale. The elliptic equation we solve is

   (91)$$\mathrm{\nabla }\cdot {\beta }_0^0\mathrm{\nabla }\varphi =\underset{0\mathrm{except}\mathrm{for}\mathrm{diffusion}}{\underbrace{{\beta }_0^0(S-\bar{S})}}-\mathrm{\nabla }\cdot ({\beta }_0^0{\mathbf{U}}^{\mathrm{init}})$$

   This $\varphi$ is then used to correct the velocity field to obtain ${\mathbf{U}}^{0,0}$. If do_initial_projection = F, set ${\mathbf{U}}^{0,0}={\mathbf{U}}^{\mathrm{init}}$.

2. “Divu” iterations: Next we do init_divu_iter iterations to project the velocity field using a constraint that sees reactions and external heating. The initial timestep estimate is provided by firstdt and ${\mathbf{U}}^{0,0}$, to allow us to compute the effect of reactions over $\mathrm{\Delta }t/2$.

   Do $\nu =1,...,N_{\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{s}}^S$.

   1. Estimate $\mathrm{\Delta }{t}^{\nu }$ using ${\mathbf{U}}^{0,\nu -1}$ and $w_0^{\nu -1}.$

   2. React State$[{\rho }^{\mathrm{init}},(\rho h{)}^{\mathrm{init}},X_k^{\mathrm{init}},T^{\mathrm{init}},({\rho }^{\mathrm{init}}{H}_{\mathrm{e}\mathrm{x}\mathrm{t}}),p_0^{\mathrm{init}}]\to [{\rho }^{\mathrm{out}},(\rho h{)}^{\mathrm{out}},X_k^{\mathrm{out}},T^{\mathrm{out}},(\rho {\dot{\omega }}_k{)}^{0,\nu }].$

   3. Compute $S^{0,\nu }$ using $(\rho {\dot{\omega }}_k{)}^{0,\nu }$ and the initial data.

   4. Compute $\overline{S^{0,\nu }}=\mathbf{Avg}(S^{0,\nu }).$

   5. Compute $w_0^{\nu }$ as in Step 2B using $\overline{S^{0,\nu }}$ and $\psi =0$.

   6. Project ${\mathbf{U}}^{0,\nu -1}$ using ${\beta }_0^0$ and $(S^{0,\nu }-\overline{S^{0,\nu }})$ as the source term. This yields ${\mathbf{U}}^{0,\nu }.$ In this projection, again the density is set to 1, and the elliptic equation we solve is:

      (92)$$\mathrm{\nabla }\cdot {\beta }_0^0\mathrm{\nabla }\varphi ={\beta }_0(S-\bar{S})-\mathrm{\nabla }\cdot ({\beta }_0^0{\mathbf{U}}^{0,\nu -1})$$

   End do.

   Define $S^0=S^{0,N_{\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{s}}^S}$, $w_0^{-\frac{1}{2}}=w_0^{N_{\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{s}}^S}$, ${\mathrm{\Delta }t}^0=\mathrm{\Delta }{t}^{N_{\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{s}}^S},$ and ${\mathbf{U}}^0={\mathbf{U}}^{0,N_{\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{s}}^S}.$

Next, we need to construct approximations to ${{\eta }_{\rho }}^{-\frac{1}{2}},{\psi }^{-\frac{1}{2}},S^1$, and ${\pi }^{-\frac{1}{2}}$. As initial approximations, set ${{\eta }_{\rho }}^{-\frac{1}{2}}=0,{\psi }^{-\frac{1}{2}}=0,S^{1,0}=S^0$, and ${\pi }^{-\frac{1}{2}}=0.$

1. Pressure iterations: Here we do init_iter iterations to get an approximation for the lagged pressure:

   Do $\nu =1,...,N_{\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{s}}^{\pi }$.

   1. Perform Steps 1-11 as described above, using $S^{\frac{1}{2},\star \star }=(S^0+S^{1,\nu -1})/2$ in Step 2 as described. The only other difference in the time advancement is that in Step 11 we define $\mathbf{V}=({\tilde{\mathbf{U}}}^{1,\star }-{\tilde{\mathbf{U}}}^0)$ and solve

      (93)$$L_{\beta }^{\rho }\varphi =D\left({\beta }_0^{\frac{1}{2}}\mathbf{V}\right)-{\beta }_0^{\frac{1}{2}}[(S^1-\overline{S^1})-(S^0-\overline{S^0})].$$

      (The motivation for this form of the projection in the initial pressure iterations is discussed in [ABC00].) We discard the new velocity resulting from this, but keep the new value for ${\pi }^{\frac{1}{2}}={\pi }^{-\frac{1}{2}}+(1/\mathrm{\Delta }t)\varphi .$ These steps also yield new scalar data at time $\mathrm{\Delta }t,$ which we discard, and new values for ${{\eta }_{\rho }}^{\frac{1}{2}}$ (Step 8C), ${\psi }^{\frac{1}{2}}$ (Step 8F), $S^{1,\nu }$ (Step 10A), and ${\pi }^{\frac{1}{2}}$ (Step 11), which we keep.

   2. Set ${\pi }^{-\frac{1}{2}}={\pi }^{\frac{1}{2}}$, ${{\eta }_{\rho }}^{-\frac{1}{2}}={{\eta }_{\rho }}^{\frac{1}{2}}$, and ${\psi }^{-\frac{1}{2}}={\psi }^{\frac{1}{2}}$.

   End do.

   Finally, we define $S^1=S^{1,N_{\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{s}}^{\pi }}.$

The tolerances for these elliptic solves are described in § Multigrid.

Changes from Earlier Implementations

Changes Between Paper 3 and Paper 4

1. We defined the mapping of data between a 1D radial array and the 3D Cartesian grid for spherical problems (which we improve upon in the multilevel paper).

2. We update $T$ after the call to React State.

3. We have created a burning_cutoff_density, where the burning does not happen below this density. It is presently set to base_cutoff_density.

4. Use corner coupling in advection.

5. We have an option, controlled by use_tfromp, to update temperature using $T=T(\rho ,X_k,p_0)$ rather than $T=T(\rho ,h,X_k)$. The former completely decouples enthalpy from our system. For spherical problems, we use use_tfromp = TRUE, for planar problems, we use use_tfromp = FALSE.

6. For spherical problems, we have changed the discretization of $\tilde{\mathbf{U}}\cdot \mathrm{\nabla }{p}_0$ in the enthalpy update to $\mathrm{\nabla }\cdot (\tilde{\mathbf{U}}{p}_0)-p_0\mathrm{\nabla }\cdot \tilde{\mathbf{U}}$.

7. In paper III we discretized the enthalpy evolution equation in terms of $T$. Since then we have discovered that discretizing the enthalpy evolution in perturbational form, $(\rho h{)}^{\prime }$, leads to better numerical properties. We use enthalpy_pred_type= 1. This is more like paper II.

8. We have turned off the evolution of $h$ above the atmosphere and instead compute $h$ with the EOS using do_eos_h_above_cutoff = T.

Changes Between Paper 4 and the Multilevel Paper

See the multilevel paper for the latest.

Changes Between the Multilevel Paper and Paper 5

1. Added rotation.

Changes Between Paper 5 and the XRB Paper

1. We have added thermal diffusion, controlled by use_thermal_diffusion, temp_diffusion_formulation, and thermal_diffusion_type.

2. We added the volume discrepancy term to the velocity constraint equation, controlled by the input parameter, dpdt_factor.

3. For certain problems, we need to set do_eos_h_above_cutoff = F to prevent large, unphysical velocities from appearing near the edge of the star.

Changes Since the XRB Paper

1. We switched to the new form of the momentum equation to Eq.39 to conserve the low-Mach number form of energy.

2. We changed the form of the volume discrepancy term to get better agreement between the two temperatures.

Future Considerations

• We are still exploring the effects of use_tfromp = F for spherical problems. We would eventually like to run in this mode, but $T=T(\rho ,X_k,p_0)$ and $T=T(\rho ,h,X_k)$ drift away from each other more than we would like. Our attempts at incorporating a dpdt_factor for spherical problems have not been successful.

Numerical Methodology

We give an overview of the original MAESTRO algorithm [NonakaAlmgrenBell+10], as well as the recently introduced new temporal integration scheme [FanNonakaAlmgren+19]. These schemes share many of the same algorithmic modules, however the new scheme is significantly simpler.

To summarize, in the original scheme we integrated the base state evolution over the time step, which requires splitting the velocity equation into more complicated average and perturbational equations. In the new temporal integrator, we eliminated much of this complication by introducing a predictor-corrector approach for the base state. The key observation is that the base state density is simply the lateral average of the full density, so we simply update the base state density through averaging routines, rather than predicting the evolution with a split velocity formulation, only to have to correct it with the averaging operator in the end regardless.

Original Temporal Integrator

New Temporal Integrator