Insertion Loss does not imply any specific conversion of energy to heat discussed an example of a lossless 100pF series capacitor in a 50Ω two port network context. This article presents measurement using a VNWA3E in a suitable fixture of a practical capacitor, an experiment that readers should be able to replicate.
This measurement set includes s21. NanoVNA does not natively correct load mismatch, so s21 measurements are tainted by that error. For that reason, the VNWA3E is used here.
The capacitor used is a 100pF silver mica capacitor, and it this configuration, we might expect Loss of the order of 0.001dB. It is a challenging measurement, and uncertainty prevents determining loss to that accuracy, but the objective is to show that InsertionLoss is almost entirely due to MismatchLoss in this example.
Above, marker frequency has been set so that capacitor reactance is ~100Ω.
So based on previous discussion, we expect InsertionLoss of 3.01dB, |s21| to be -3.01dB.
Above are the values of |s11| and |s21| at the marker, and calculation of InsertionLoss, Loss, and MismtachLoss.
it can be seen that Loss at 0.0016dB is much less than InsertionLoss at 3.200dB, a result of the large input mismatch, MismatchLoss is 3.198dB, most of the InsertionLoss. The precision give here is beyond uncertainty, the key message is that MismatchLoss almost equals InsertionLoss, and Loss (which gives rise to heating) is very very small, of the order of 0.002dB.
Conclusions
The experiment results confirms that this is a nearly lossless network element (100pF series silver mica) that gives rise to a significant InsertionLoss, and that most of that is MismatchLoss, very very little energy is converted to heat in that element.
It all turns on the concepts and meanings of the terms InsertionLoss, Loss and MismatchLoss, and the careful consistent application of those to the problem.