How Lightning Routing Actually Works in 2026

The Core Problem#

Lightning is a network of payment channels. When you send a payment to someone who isn't your direct channel peer, your node has to find a path through the network — a series of connected channels that together bridge you to the destination.

This sounds simple. It isn't.

What Your Node Actually Knows#

Your node has a partial, stale view of the network. It knows:

• The channel graph (which nodes are connected to which)
• Public channel capacity (total capacity, not balance)
• Fee policies (base fee + proportional fee rate)

What it doesn't know:

• How balanced any given channel is right now
• Whether a channel was just drained by a large payment five minutes ago
• Whether a node is having connectivity issues

This is why pathfinding is probabilistic, not deterministic.

Pathfinding Algorithms#

LND uses a modified Dijkstra's algorithm with a probability model. When it finds a route, it assigns a success probability to each hop based on:

1. Channel capacity: Can this channel physically route the amount?
2. Historical success: Has this channel worked for similar amounts recently?
3. HTLC bounds: Most channels have min/max HTLC constraints

The algorithm finds routes that maximize the expected value (probability × payment amount), not just the minimum fee path.

Why Payments Fail#

The most common failure modes:

• Insufficient liquidity: Channel balance < payment amount on that side
• HTLC minimum threshold: Your payment is too small for a fee-conscious node
• Timeout: The path is too long (more than ~20 hops risks CLTV issues)
• Offline peer: Node is unreachable

MPP: Split and Conquer#

Multi-Path Payments (MPP) send the same payment in parallel shards across multiple routes. A 1M sat payment might split into 4 × 250k sat shards.

This dramatically improves success rates because:

• Each shard needs less liquidity from each channel
• Independent failure of one path doesn't fail the whole payment
• Paths can use different fee levels based on shard size

Fee Markets in Practice#

The mempool has nothing on Lightning fee dynamics. Node operators set fee policies autonomously, and the competitive pressure to route payments creates a fee market without a central authority.

Routing nodes that maintain balanced channels earn fees continuously. Nodes with chronically imbalanced channels earn nothing. The network self-corrects through fee differential — nodes lower fees on the direction they want to attract traffic from.

What This Means for Users#

If your payment fails, try again. The second attempt uses updated probability estimates. Most wallets do this automatically.

If it fails three times, the issue is likely structural — either the destination has poor connectivity, or you need more inbound capacity on your own node.

The Lightning Network in 2026 is mature enough that casual users rarely see these failures. But understanding the mechanics helps when you're building on top of it.