Classwise permanence for weakly reversible mass-action systems

With fixed rates, every weakly reversible chemical network herds each class of states into one common window.

2 papers September 25, 2026 Explained by AI

Chemicals in a reaction network can, in principle, die out or pile up without limit. A September 2026 paper proves that weakly reversible networks with fixed rates avoid both (Lean proof listed). The October paper claims to prove something stronger. All starts that share the same conserved totals end up in one common window. The time it takes to get there may still depend on the start.

First step

1987

Martin Feinberg, raised the question of zero-concentration limits

Significance

Major

It claims the fixed-rate weakly reversible case of the permanence question, which earlier work had settled only in special cases.

Trust

Formal proof listed

Field

Dynamical systems and ergodic theory

2 papers

The explanation

Picture a vat holding one chemical, A. Two reactions run at once. In one, a molecule of A copies itself: A → 2A. In the other, two molecules of A merge into one: 2A → A. Under mass action, a reaction runs at its rate constant times the concentrations it uses up, multiplied together. Set both rate constants to 1. Then the concentration x of A changes at speed x − x². Below 1, copying wins and x grows. Above 1, merging wins and x falls.

This toy system can be solved exactly. Start at x = 0.001, and x climbs to 0.5 after about 6.91 time units. Start at x = 1000, and x falls to 2 after about 0.69 units. After that, both stay between 0.5 and 2 for good. The waiting time depends on the start. The final window does not. Every positive start, however extreme, ends up in that one window. This is what mathematicians call permanence.

Real reaction networks have many chemicals and many reactions. A network is weakly reversible when every reaction has a directed return path. The path can take several steps, such as A → B → C → A. Many networks also keep some totals fixed. Take 2A ⇌ B, where two A molecules join into one B and split back. There the total a + 2b never changes. All positive states that share these fixed totals form one class (a stoichiometric compatibility class). In the A ⇌ 2A vat nothing is conserved. So its class holds every positive number, and it is unbounded.

Three questions sit side by side. Boundedness asks that no concentration blow up. Persistence asks that none drift toward zero. Permanence asks for more: one window that works for the whole class, after a waiting time. In 1987, Martin Feinberg raised the question of ruling out limits where some concentration is zero. Partial answers followed. Craciun, Nazarov and Pantea proved permanence for two-chemical weakly reversible systems in 2013. Gopalkrishnan, Miller and Shiu proved it in 2014 for another family, strongly endotactic networks. That family includes every weakly reversible network whose reactions form one connected piece. Our toy vat is one such piece, so it was already covered. Boros and Hofbauer gave another proof of that case in 2020.

A companion paper from September 2026 proves boundedness and persistence for every finite weakly reversible network with fixed positive rates. It notes that a 2010 paper by August and Barahona had stated this conclusionSource 1. It also says one step of that proof does not hold as writtenSource 1. The companion's bounds come with a catch. Each starting point gets its own window, which may differ from start to start. A start nearer zero might need a wider window. That paper does not promise one window for a whole classSource 2.

The October paper claims to prove that promise. For each class it builds one compact convex regionSource 3. Compact means closed and bounded. Convex means the straight segment between any two of its points stays inside. Call the path a starting state follows over time its trajectory. Every trajectory in the class enters this region in finite time and never leaves. The region depends only on the network, the rates and the classSource 4. This holds even when the class itself is unbounded, as in the toy vatSource 3.

The hard part is turning many personal windows into one shared window. The companion paper already traps each start in a region of its own. The new step links traps of different sizesSource 5. A trajectory caught in a wide trap is pushed into a narrower one. After finitely many such steps, it lands in one fixed trap chosen for the class.

Some limits stay. The waiting time still depends on where a trajectory starts. The theorem says a common window exists, but its statement gives no size for it. Rates must be fixed numbers. Several earlier results also allowed rates that change over time within fixed positive boundsSource 6. The theorem does not say that concentrations settle to a single resting stateSource 7.

How sure is this? A computer proof checker called Lean has checked the companion's main theorem. Every positive start has a solution for all time. One number keeps its concentrations above it and below its reciprocal from the start onward. That number may depend on the start. The October paper's common window for each class is outside that check. No specialist review has been published yet.

What it doesn't show. This does not show that concentrations settle to one resting state, or cover rates that change over time.

Where it came from

This result comes from an unreleased OpenAI model, published in the openai/math release on 6 October 2026.

Hype check

Major if it holds. Lean has checked the companion's start-dependent bounds, not the common window for each class. No specialist review has been published, and time-varying rates remain outside the result.

Words used

Mass-action kinetics: A rule where each reaction runs at a fixed rate constant times the product of the concentrations it uses up.
Weakly reversible: Every reaction in the network can be undone by some chain of reactions leading back to where it started.
Stoichiometric compatibility class: All positive states that share the totals the reactions can never change.
Boundedness: No concentration grows without limit along a trajectory.
Persistence: No concentration drifts toward zero along a trajectory.
Permanence: One window, the same for a whole class, that every trajectory in that class eventually enters and stays in.
Compact convex set: A closed, bounded region that contains the straight segment between any two of its points.
Linkage class: One connected piece of the network's reaction diagram.

The original papers (2)

Uniform Permanence in Weakly Reversible Mass-Action Systems (PDF, opens in a new tab)
October 5, 2026
Boundedness and persistence of weakly reversible mass-action systems (PDF, opens in a new tab)
September 25, 2026

Lean notes

All quoted lines (7)

That implication does not hold throughout the stated region.
Boundedness and persistence of weakly reversible mass-action systems Back to text
Such sets give boundedness and persistence, but may depend on the initial point.
Uniform Permanence in Weakly Reversible Mass-Action Systems Back to text
Thus the fixed-positive-rate, weakly reversible formulation of the permanence conjecture holds, including on unbounded classes.
Uniform Permanence in Weakly Reversible Mass-Action Systems Back to text
In particular, there is $\eps_P\in(0,1)$ , depending only on the network, the fixed rates, and $P$ , such that every such solution satisfies $\eps_P\leq x_i(t)\leq\eps_P^{-1} \qquad(1\leq i\leq d,\ t\geq T(x^0)).$
Uniform Permanence in Weakly Reversible Mass-Action Systems Back to text
They turn bounds for individual trajectories into permanence on a fixed class.
Uniform Permanence in Weakly Reversible Mass-Action Systems Back to text
These permanence results also permit rates varying in time within fixed positive bounds.
Uniform Permanence in Weakly Reversible Mass-Action Systems Back to text
We make no assertion in this generality that the trajectory converges to an equilibrium, or that the same $\varepsilon$ works for all positive initial points.
Boundedness and persistence of weakly reversible mass-action systems Back to text