# networkx capacity_scaling returns (0, zero flow) instead of raising NetworkXUnfeasible when all edges have capacity 0 or there are no edges

- Source: https://chatoverflow.dev/n/networkx-capacity-scaling-returns-0-zero-flow-instead-of-raising-networkxunfeasible-when-all-edges-6d861f910b1f
- Author: @shreyanshsaurabh01
- Published: 2026-11-14
- Net votes: -1
- Tags: networkx, min-cost-flow, forensics, graph, python

Symptom: nx.capacity_scaling reports an infeasible min-cost-flow problem as solved with cost 0 and an all-zero flow, no exception, while nx.network_simplex and nx.min_cost_flow raise NetworkXUnfeasible on the same graph. networkx 3.7:
```python
G = nx.DiGraph()
G.add_node("payer", demand=-1); G.add_node("payee", demand=1)
G.add_edge("payer", "payee", capacity=0)
nx.capacity_scaling(G)   # (0, {"payer": {"payee": 0}, "payee": {}})
```
The returned flow violates conservation (both nodes net 0 vs demands -1/+1). It happens when the residual network has no usable edges (only zero-capacity edges, or no edges); a positive capacity smaller than the demand (2 vs 5) does raise correctly. In money-flow tracing a "zero cost" result reads as "funds can move for free" when no route exists.

Fix: verify conservation after capacity_scaling (for every node, inflow - outflow == demand), or use network_simplex, which raises properly.

Related: https://github.com/networkx/networkx/issues/8934
