There is a fault in electrical work that no breaker will ever catch. It doesn't trip anything. It doesn't blow a fuse. It doesn't show up on the panel schedule or the load calc or the inspector's punch list. The conductor gets hot, the insulation slowly cooks, the wire nut discolors, and the overcurrent protection sits there doing exactly what it was designed to do — which is nothing, because the thing it protects isn't the wire that's burning.

The neutral has no breaker. And on a modern branch circuit feeding LED drivers, computers, or a bank of switch-mode power supplies, the neutral can be carrying more current than either hot conductor it's supposed to be relieving.

Most electricians learn the neutral as a subtraction problem. Balance the phases and the neutral carries the difference. Balance them perfectly and the neutral carries nothing. That's true — and it was reliably true for most of the twentieth century, when the load on the other end of the wire was a filament, a heating element, or a motor. It stopped being reliably true somewhere around the time we started plugging in everything that has a circuit board inside it.

The load stopped drawing a sine wave

A resistive load — an incandescent bulb, a baseboard heater — draws current in the same shape as the voltage that pushes it. Voltage rises, current rises. Voltage crosses zero, current crosses zero. Two clean sine waves, locked together. Everything you were taught about phase cancellation assumes this shape.

A switch-mode power supply does not work that way. It rectifies the incoming AC and dumps it into a capacitor, and that capacitor only accepts charge when the instantaneous line voltage exceeds what's already stored in it. Which means for most of each half-cycle, the supply draws nothing at all. Then, near the peak of the waveform, it draws a hard, narrow gulp of current — and shuts off again.

That pulse is what we call a nonlinear load. And a pulse is not a sine wave. Fourier told us that any repeating waveform can be decomposed into a sum of sine waves at integer multiples of the fundamental frequency. So that spiky 60 Hz current pulse is, mathematically and physically, a 60 Hz sine wave plus a 180 Hz sine wave plus a 300 Hz sine wave, and so on up. The odd harmonics. The wire doesn't know it's carrying a distorted waveform — it just carries all of those component currents at once, and heats according to all of them.

Why the third harmonic refuses to cancel

Here is the part that breaks the intuition.

In a three-phase wye system, the fundamental currents are 120 degrees apart. That separation is the whole reason a balanced neutral carries zero: three equal currents, spaced evenly around the circle, sum to nothing at the center point.

Now take the third harmonic. It oscillates three times for every one cycle of the fundamental. So when you shift the fundamental by 120 degrees, you shift its third harmonic by 3 × 120 = 360 degrees — which is a full revolution, which is no shift at all. The third harmonic on phase A, the third harmonic on phase B, and the third harmonic on phase C are all in phase with each other.

They don't cancel at the neutral. They stack.

The same argument holds for the ninth, the fifteenth, the twenty-first — every odd multiple of three. Electricians call them the triplen harmonics, and in a four-wire wye system they behave like zero-sequence currents: they arrive at the neutral together and add arithmetically instead of vectorially. Perfectly balance your phases and you have done nothing whatsoever to the triplens. You have only cancelled the fundamental, leaving the harmonic content behind with nowhere to go but back down the shared neutral.

In the theoretical worst case — a load drawing pure third harmonic — the neutral current reaches √3 times the phase current. About 1.73 to 1. Real-world circuits don't hit that ceiling, because real distorted waveforms still contain a large fundamental component. But neutral currents exceeding phase currents on heavily nonlinear circuits are a documented, measured, unremarkable reality in commercial buildings, and they were common enough that the Code went and wrote a rule about it.

What the Code actually says about it

Two rules matter, and they sit in different chapters, which is part of why they get missed.

The neutral counts as a current-carrying conductor. NEC 310.15(E)(3) says that where a major portion of the load consists of nonlinear loads, harmonic currents are present in the neutral conductor, and the neutral shall therefore be considered a current-carrying conductor. That's not a footnote — that's the trigger for the adjustment factors in 310.15(C)(1). Run three phases and a neutral in one raceway with a linear load and you have three current-carrying conductors, no derating. Run the same four wires to a rack of power supplies and you have four current-carrying conductors, and every ampacity in that raceway drops to 80 percent.

The neutral can't be reduced. 220.61(C)(2) prohibits applying the usual neutral reduction to the portion of the load that consists of nonlinear loads supplied from a 4-wire, three-phase wye system. The old habit of running a half-size neutral on a feeder because "it only carries the imbalance" is precisely the assumption that harmonics invalidate.

Notice what these rules are actually doing. They are not asking you to measure total harmonic distortion. They are not asking you to own a power quality analyzer. They are asking you to look at what's on the other end of the pipe and make a judgment about whether it draws sine waves or pulses. Fluorescent ballasts, LED drivers, VFDs, UPS systems, servers, EV chargers, anything with a rectifier front end — pulses. That's the whole test.

The failure you'll actually see

The neutral doesn't announce itself. It fails the way slow thermal problems always fail: an intermittent that comes and goes with building occupancy, a panel that runs warm, a neutral bar screw that discolors, insulation that turns brittle and brown ten years before its time. On a shared-neutral multiwire branch circuit feeding modern electronics, the neutral is quietly carrying more heat than either of the conductors that a 20-amp breaker is faithfully watching.

And when that neutral finally opens — a loose lug, a cooked splice, a nicked strand — the line-to-neutral loads on that circuit stop seeing 120 volts. They see a voltage divider between two phases, split according to load impedance. The lightly loaded side goes high. Equipment dies in the exact way that makes a homeowner say the lights got bright right before everything stopped working. The overloaded neutral is the cause. The dead electronics are the symptom that finally gets you called.

Your next moves

  • Count your current-carrying conductors again on every raceway feeding electronics. Three phases plus a neutral to a server rack, a lighting panel full of LED drivers, or a bank of EV chargers is four CCCs, not three. Apply the 80 percent adjustment factor from 310.15(C)(1) before you sign off on the wire size.
  • Stop half-sizing feeder neutrals to panels that serve nonlinear load. Run a full-size neutral. If the load is heavily nonlinear, oversizing beyond full size is a defensible engineering choice — and 220.61(C)(2) already forbids the reduction you might have been tempted to take.
  • Clamp the neutral, not just the hots, on your next commercial service call. Put an ammeter on a four-wire wye neutral in a lighting panel or IT closet and compare it to the phase readings. Do this once and the concept stops being theoretical forever. Use a true-RMS meter — an averaging meter will read a distorted waveform low and lie to you about the size of the problem.
  • Torque and inspect neutral terminations on nonlinear panels first. The neutral bar is the one carrying current no breaker is protecting. It deserves the same wrench and the same torque spec as the phase lugs, per NEC 110.14.
  • When a client reports "weird" intermittent electronics failures on one circuit, suspect the shared neutral before you suspect the equipment. Open-neutral voltage divider behavior explains far more service calls than bad appliances do.

The math you shouldn't be doing on a scrap of Romex

None of this is hard arithmetic. It's just arithmetic that changes depending on the raceway, the ambient temperature, the conductor count, the insulation rating, and which column of 310.16 you're actually allowed to read from — and it tends to come up while you're standing in a ceiling on a Friday afternoon with a phone that has no service. That's the situation Voltly was built for: ampacity with derating and adjustment factors applied, conduit fill, box fill, voltage drop, and NEC references, all running offline in your pocket where the job actually happens. It won't clamp the neutral for you. But when you've counted four current-carrying conductors instead of three, it will tell you exactly what that costs you.

If you'd rather spend your Friday afternoon finishing the job than second-guessing the table, take a look at Voltly.