Junction Coupling Conditions as Path Choices for Non-Conservative Products

Where pipes, arteries, or channels meet, a one-dimensional flow model needs an extra rule at the junction. Mass conservation alone does not fix the states on each branch. Gas-network modelers usually add "equal pressure": every pipe touching the node has the same pressure there. That rule is known to remove spurious pressure jumps (Banda, Herty, Klar). A separate literature deals with non-conservative hyperbolic systems. These are equations of the form $\partial_t u + A(u)\partial_x u = 0$, where the product of a discontinuous coefficient and a jump has no unique meaning. Fixing it requires choosing a "path," meaning an assumed internal profile of the jump through state space. Different paths give different shock speeds and post-shock states (Nkonga & Chandrashekar).

This note argues that these are the same problem seen from two sides. A junction where pipe area, wall stiffness, or sound speed changes abruptly contains a non-conservative term, for example pressure times the area gradient. Every candidate coupling rule corresponds to a particular path through that term. The candidates include continuity of momentum flux, continuity of total head (Bernoulli), and equal static pressure. We propose to make this correspondence exact and to extend it from two-pipe interfaces to multi-branch junctions through a "hub path." We would then test numerically which path emerges when the junction is resolved as a thin region with physical viscosity. The result would tell practitioners when equal pressure is the right closure, how large an error it introduces otherwise, and how to build junction solvers whose coupling rule matches the path used inside the branches.

Core Claim

Treat each geometric or material parameter $\theta$ as an extra state variable obeying $\partial_t \theta = 0$. Examples are cross-section, vessel stiffness, and sound speed. A junction then becomes a stationary discontinuity of an augmented non-conservative system.

For a pipe with area $\theta$, the momentum balance contains the term $p\,\partial_x\theta$, where $p$ is pressure. Under a path $\phi$ joining the left trace $u_L$ to the right trace $u_R$, the stationary generalized jump condition reads

$$ 0=\int_0^1 A\big(\phi(\tau;u_L,u_R)\big)\,\partial_\tau\phi\,d\tau . $$

Here $A$ is the augmented coefficient matrix and $\tau\in[0,1]$ parametrizes the path. This condition is the coupling rule, and different rules are different paths:

  • A path along which the flow stays at constant pressure gives equal pressure.
  • A path that follows smooth steady flow through the area change gives Bernoulli-type continuity.
  • The default straight segment in conserved variables gives a third rule, which matches neither of the others.

These rules differ by terms of order $M^2$, where $M$ is the Mach number, because the steady momentum balance trades pressure against $\rho u^2$. Equal pressure should therefore be the low-Mach limit of the physically selected path, not an exact law.

For a star junction with $K$ branches, we define a hub path. Each branch trace $u_k$ connects through its own path to a common hub state $u_*$. Mass balance is imposed at the hub, and equal pressure becomes the statement that all branch paths end at the same hub pressure. This replaces the pairwise path of Kolbe, Herty, Müller with an object defined for any number of branches.

Relation to Prior Work

Three papers motivate the proposal, two constrain it, and one supplies a method.

Motivation. Kolbe, Herty, Müller couple two non-conservative systems across a single interface using a linear relaxation system. Their test case is blood flow with a jump in vessel stiffness. They report that the direction of path integration is "crucial for maintaining physical consistency." We take that observation as the starting point but differ in three ways: we work with multi-branch junctions, we ask which path is physically selected rather than defaulting to segments, and we tie the path to the equal-pressure rule. Banda and coauthors analyze equal pressure for conservative isothermal gas, both in the isothermal Euler paper and in the drift-flux paper with unequal sound speeds. In both, the rule is imposed as an algebraic constraint. Neither asks what it means as a jump condition. Redle & Herty bring equal pressure into an asymptotic-preserving scheme for T-junctions at low Mach number. That is exactly the regime where our claim predicts equal pressure is accurate.

Constraints. Nkonga & Chandrashekar show that numerical schemes for non-conservative shocks can converge to a different limit than the assumed path, because numerical dispersion shapes the jump. This rules out using a numerical scheme's own dissipation as the arbiter of the physical path. Any resolved reference must use explicit physical viscosity, converged in mesh at fixed viscosity before viscosity is sent to zero. Herty, Kolbe, Müller show that linear coupling of relaxation variables creates boundary-layer artifacts and that consistent coupling must be nonlinear. Our relaxation-based solver must therefore impose the hub-path condition nonlinearly.

Method. The relaxation machinery and its subcharacteristic stability condition $\Lambda - A(u)^2 \ge 0$ are reused from Kolbe et al..

Research Program

Stage 1: theory for two-branch interfaces. For isothermal Euler with a jump in sound speed or pipe area, and for the blood-flow model with a stiffness jump, we compute the coupling rule induced by three path families. These are the straight segment in conserved variables, the constant-pressure path, and the smooth-steady-flow path. We will derive closed-form expansions of the pressure mismatch between them in $M$ and in the parameter ratio. We also determine where each induced coupling problem loses uniqueness as the flow approaches the sonic point.

Stage 2: hub paths for star junctions. We prove existence and uniqueness of the junction Riemann problem for 1-to-2 and 2-to-1 junctions under the hub-path condition. Following Banda et al., the proof intersects wave curves (the families of states reachable from a given state through a shock or rarefaction) with the hub constraints. We start with subsonic, unidirectional flow and then attempt flow reversal, where an outgoing branch becomes incoming. In the hub formulation, reversal only changes which branches are upstream of the hub.

Stage 3: physical selection. We resolve each junction as a transition region of width $\delta$, in which the parameter varies smoothly and a physical viscosity $\nu$ acts. For the T-junction, the region is a short merging segment, an assumption we will vary. We then take $\delta,\nu\to0$ numerically and identify the emerging coupling rule. This directly tests the tension the file raises: viscous junctions may generate pressure drops incompatible with exact equal pressure.

Stage 4: solver. We build a path-consistent junction solver inside the relaxation scheme. It uses the same path family at the junction as in the branch interiors, it preserves steady junction states exactly, and it accepts non-conservative branch dynamics. The target branch model is blood flow in a vessel bifurcation.

Baselines

The naive reference is the straight-segment path-conservative coupling. This is what most path-conservative codes would do at an interface, and it is what Kolbe et al. default to. It will run on identical meshes and time steps. Beating it means a smaller junction pressure error against the resolved reference at matched cost.

The strongest existing method for conservative gas networks is direct algebraic equal pressure in a half-Riemann solver, as in Banda et al. and Redle & Herty. It is the strongest competitor because it is cheap, well-posed in its regime, and physically motivated. Redle & Herty report a speedup of about 140 times over an explicit scheme at $\epsilon=0.001$, with first-order accuracy in time and second-order in space. Banda et al. report removal of the pressure jump at $a_1=6, a_2=8$, but give no error norms. No reported junction error values were found on Lacuna for either method. We will reimplement it with the same branch discretization, grid sequence, and time-step rule as our solver. Our solver beats it if it is closer to the resolved reference whenever $M$ is not small, and statistically indistinguishable when $M$ is small. Equality at low Mach is part of the prediction, not a failure.

For the non-conservative blood-flow case, the strongest baseline is the pairwise path-conservative Kirchhoff relaxation coupling of Kolbe et al.. They report only first-order convergence in $\epsilon$ and $\Delta x$, with no error table. At a bifurcation, this baseline is applied pairwise by splitting the junction into sequential interfaces.

An ablation keeps our solver unchanged but swaps the hub path for the straight segment. This isolates whether the path choice, rather than the solver structure, drives any gain.

Why not just…?

Why not just impose equal pressure algebraically and ignore paths? For conservative gas at low Mach number, this probably works as well as anything, and our Stage 1 expansions predict that it should. The threat is real in that regime, so we do not claim a gain there. The open cases are three. The first is moderate Mach number, where the $O(M^2)$ mismatch grows. The second is non-conservative branches, where the interior scheme uses one path and the junction implicitly uses another; this produces a jump condition inconsistent with the bulk equations, which is exactly the inconsistency Kolbe et al. warn about. The third is flow reversal, where existing proofs assume unidirectional flow. The Mach sweep in our experiments settles where algebraic equal pressure stops being adequate.

Why not just use a straight-segment path, since it is standard? Chandrashekar et al. note there is no physical reason for nature to follow a straight line in state space. Our Stage 1 calculation predicts that segment paths produce neither equal pressure nor Bernoulli continuity at a parameter jump. The ablation measures that directly.

Why not just resolve every junction with a fine mesh and physical viscosity? That is our ground truth, but it costs a fine-mesh, small-viscosity solve at every node of a network. It also inherits the dispersion issue identified by Nkonga & Chandrashekar, so it needs careful convergence in viscosity. The point of the project is to compress that limit into a closed-form coupling rule.

Why not chain pairwise Kirchhoff interfaces at a bifurcation? Chaining makes the result depend on the order in which branches are processed. A hub path removes that dependence by construction. The experiments test whether order dependence is numerically visible.

Proposed Experiments

Experiment 1: two-pipe sound-speed jump. We use the Banda et al. Riemann problem with $a_1=6, a_2=8$, extended to an area jump and swept over $M$ from 0.01 to 0.8. The metric is the junction pressure and mass-flux mismatch of each coupling rule against the $\delta,\nu\to0$ resolved reference, together with the scaling exponent of that mismatch in $M$. The decisive check is whether equal pressure's error scales as $M^2$ and whether the selected path matches one of the three analytic families.

Experiment 2: blood-flow stiffness jump. We use the Kolbe et al. benchmark and compare continuity of pressure and flow rate, wave speeds, and $L^1$ convergence order.

Experiment 3: T-junctions. We use the 1-to-2 and 2-to-1 configurations of Redle & Herty, first with isentropic gas and then with a non-conservative blood-flow bifurcation, including a transient that reverses flow in one branch. The metrics are junction pressure error, order dependence for the pairwise baseline, and exact preservation of steady junction states. The well-balancing target follows the machine-precision standard ($10^{-13}$–$10^{-15}$) of Zhang et al..

The project fails if two things hold together. The first is that the resolved limit yields exact equal pressure at all Mach numbers tested. The second is that algebraic equal pressure matches our solver on the non-conservative bifurcation. If both hold, the path view adds nothing operationally.