A neutrino produced as an electron neutrino can later be detected as a muon neutrino. Super-Kamiokande confirmed this experimentally in 1998 (and it was part of the 2015 Nobel Prize in Physics). The whole effect follows from one fact, namely that the states the weak interaction produces, the flavor eigenstates, are not the states with a definite mass. In this post we derive the survival probability for the two-flavor case from scratch. Nothing beyond basic QM is needed, just two bases, one rotation angle and time evolution.
The free Hamiltonian has mass eigenstates∣ν1⟩,∣ν2⟩ with masses m1,m2. The weak interaction, on the other hand, couples to the flavor eigenstates∣νe⟩,∣νμ⟩, so the neutrino that comes out of a decay is in general not an energy eigenstate. The two bases are related by a rotation through the mixing angleθ,
Let the neutrino start out as a pure electron neutrino, i.e. ce(0)=1, cμ(0)=0. To get the mass-basis amplitudes we invert the rotation. It is orthogonal, so the inverse is the transpose, and
c1(0)=cosθ,c2(0)=−sinθ
(These can also be read off directly from the expression for ∣νe⟩ above.) Putting them in,
Neutrinos are ultra-relativistic, Ei≫mic2. We take both mass states to have the same momentum p (the usual textbook simplification, which is not strictly right but gives the correct answer here) and expand,
Ei=p2c2+mi2c4≈pc(1+2p2mi2c2)
so that the energy difference is
E2−E1≈2E(m22−m12)c4=2EΔm2c4
where E≈pc is the common energy. Finally we trade the time for the distance travelled, L=ct,
2ℏ(E2−E1)t=4EℏcΔm2c4L
and putting everything together,
P(νe→νe)=1−sin2(2θ)sin2(4EℏcΔm2c4L)
The mixing angle sets the depth of the oscillation, and Δm2 sets how fast it oscillates as a function of L/E.
This formula is also why oscillations were such a big deal. The argument of the sin2 contains Δm2=m22−m12. If both neutrinos were massless, Δm2=0 and the probability would be exactly 1 for all L, i.e. no oscillation at all. Spelled out step by step,
oscillations need two different phasese−iE1t/ℏ and e−iE2t/ℏ to interfere,
for massless particles Ei=pc for all i, so the phases are identical and never interfere,
so a non-trivial oscillation pattern needs m1=m2, and hence at least one of the two masses is non-zero.
NOTE
Oscillations only see mass differences, not absolute masses. An oscillation experiment can tell you that Δm2=0, but it cannot tell you the absolute mass scale.
In nature there are three flavors and not two, so the next question is how many of the neutrinos have to be massive. With three flavors there are three mass eigenstates ∣ν1⟩,∣ν2⟩,∣ν3⟩ and three flavor eigenstates ∣νe⟩,∣νμ⟩,∣ντ⟩. Each pair of mass eigenstates gives its own oscillation frequency, set by Δmij2=mi2−mj2, in the same way as the single Δm2 in the two-flavor formula. Only two of these are independent, since Δm312=Δm322+Δm212.
Experimentally, two different frequencies are seen. Solar neutrinos and the KamLAND reactor experiment measure Δm212≈7.5×10−5eV2, while atmospheric and accelerator neutrinos measure ∣Δm312∣≈2.5×10−3eV2, roughly 30 times larger. (The sign of Δm312 is still not known, and this is the mass-ordering question.)
That is enough to answer the question. Suppose only one mass eigenstate were massive, say m1=0 and m2=m3=0. Then Δm232=0 and Δm212=Δm312=−m12, so there would be only one non-zero splitting and a single oscillation frequency. Two different non-zero splittings require at least two massive eigenstates. As far as oscillations are concerned, the lightest one could still be exactly massless.