What Are Virtual Particles? What Experiments Really Measure

Virtual particles help physicists calculate interactions. They are internal contributions to a quantum-field calculation, not particles that briefly escape the rules of physics. The distinction matters: experiments can confirm an interaction without detecting a virtual particle as a separate object. Neither energy conservation nor the uncertainty principle requires particles to borrow energy from empty space.

What are virtual particles?

In quantum field theory, particles are excitations of fields. Calculating how those excitations interact often requires an approximation called perturbation theory. Physicists express a scattering amplitude as a series of terms, then use that amplitude to predict probabilities.

Feynman diagrams organize those terms. External lines represent the incoming and outgoing states of the calculation. Internal lines represent propagators: mathematical factors connecting interactions. Physicists call the particles associated with those internal lines virtual particles.

Consider two electrons scattering. One leading diagram connects their electron lines with an internal photon line. That virtual photon carries the momentum transferred between them. It does not describe an extra photon flying into a detector.

The diagram is only part of the calculation. Contributions interfere, and higher-order terms need not always shrink neatly. Perturbation theory works best when the relevant expansion remains controlled. Our quantum-mechanics guide explains why adding amplitudes differs from adding ordinary probabilities.

A diagram is not a detector image

What does off-shell mean?

A free, stable particle obeys the relativistic relation between energy, momentum and rest mass:

E² = p²c² + m²c⁴

Here E is energy, p is momentum magnitude, m is rest mass, and c is the speed of light. Physicists call this relation the mass shell. A freely propagating photon has m = 0, so E = pc.

An internal propagator is not restricted to that relation. Its four-momentum can be off-shell. For example, momentum transfer between elastically scattering electrons can be spacelike: its energy is too small for a free particle with that momentum.

That does not give a photon a new, negative rest mass. It means the internal contribution is not a free photon. Nor must every internal momentum stay strictly away from the mass shell; propagators can approach or encounter on-shell regions.

This is why “virtual means off-shell” works as an introduction but needs qualification. The distinction concerns the role of a contribution in the calculation, not a second species of particle.

Do virtual particles borrow energy?

No. In ordinary scattering calculations with translation-invariant backgrounds, four-momentum is conserved at each Feynman-diagram vertex. A vertex is a point where lines meet. The outgoing energy and momentum balance the incoming quantities.

The internal line can be off-shell while that balance remains exact. Confusing the mass-shell relation with energy conservation creates the familiar story of an energy loan. They are different constraints.

Energy–time uncertainty does not permit temporary violations of conservation laws. It describes relations involving energy spread and a relevant timescale, such as evolution or decay. Time does not play the same operator role as position in standard quantum mechanics.

Alternative perturbation methods arrange intermediate states differently. Those intermediate quantities are still components of an amplitude, not measurable stages during which nature suspends a law. The uncertainty principle needs no repayment deadline.

Does the vacuum contain particles that flicker into existence?

A quantum vacuum is not a classical state with every field fixed exactly at zero. Fields have fluctuations and correlations even in their lowest-energy state. But fluctuations do not automatically mean little objects appear and disappear along definite trajectories.

For a time-independent Hamiltonian, a vacuum energy eigenstate evolves only by an overall phase. Its ordinary expectation values remain stationary. Unequal-time correlations can still depend on the interval between measurements. Stationarity therefore does not remove quantum structure.

Loop diagrams capture some of that structure in perturbation theory. Reading a closed loop as a literal particle’s itinerary adds a story that the calculation does not establish.

The Casimir force illustrates the difference. In a peer-reviewed analysis, Robert Jaffe showed how to compute it through interactions between material charges and currents without invoking zero-point energies. The force is measurable. It does not uniquely establish a picture of particles popping out of nothing.

Are virtual particles real? Ask what the experiment measures

The word “real” often hides two questions. Do quantum-field interactions produce measurable effects? Yes. Does each internal line correspond to an independently observable particle? No.

Experiments measure quantities such as scattering rates, energy levels and magnetic moments. The theory must reproduce them after combining the relevant contributions. Individual diagrams can depend on gauge choices or on how physicists organize the approximation. They do not all have separate observable meanings.

Other methods can compute some of the same physics without summing ordinary particle-exchange diagrams. For example, a 2021 lattice-QCD calculation evaluated the leading hadronic vacuum-polarization contribution to the muon’s magnetic moment numerically.

That does not mean lattice methods remove quantum fluctuations or forbid diagrammatic descriptions. It shows that a literal population of virtual particles is not required to formulate the calculation. The measured effect has a firmer status than a story about its intermediate steps.

Why unstable particles complicate the picture

A particle need not reach a detector intact to count as an established discovery. Experiments identify short-lived particles through the distributions of their decay products.

The Z boson provides a clear example. Its reconstructed mass distribution has a resonance near 91.2 GeV, with a natural width near 2.5 GeV. The LEP and SLD electroweak analysis reports the precision measurements. The width reflects its finite lifetime.

Physicists often call production near the resonance on-shell production, even though the unstable particle is an intermediate state. Far from the resonance, off-shell contributions can still affect the observed final state.

The Higgs decay H → ZZ* → four leptons uses this distinction. A 125 GeV Higgs cannot produce two on-shell Z bosons. The asterisk marks an off-shell Z contribution. Energy conservation holds throughout.

What off-shell Higgs measurements actually establish

The Higgs also contributes to Z-pair production at invariant masses far above its 125 GeV resonance. Experiments fit the distributions of the final particles, including interference with non-Higgs processes. They do not catch a heavy virtual Higgs in isolation.

A CMS analysis published in 2025 reports 3.8σ evidence for off-shell Higgs production. Its combined width estimate is 3.0^+2.0_−1.5 MeV. An ATLAS analysis published in 2025 reports 3.7σ evidence and 4.3^+2.7_−1.9 MeV. The quoted width intervals correspond to 68% confidence.

Both estimates agree with the Standard Model prediction of about 4.1 MeV. However, extracting a width requires assumptions connecting on-shell and off-shell couplings. New contributions could alter that connection. These are model-dependent inferences, not direct measurements of a resolved 4 MeV-wide peak.

Light-by-light scattering tests the calculation

Classical Maxwell theory in vacuum lets light beams pass through one another without scattering. Quantum electrodynamics predicts a small interaction. Its leading contribution contains loops of charged fields.

ATLAS tested this in collisions where lead nuclei passed without a hadronic collision. Their electromagnetic fields supplied the interacting photons. The 2019 observation contained 59 candidates against an expected background of 12 ± 3 events. The excess reached 8.2σ.

This supports the quantum prediction for photon scattering. It does not reveal the trajectory of a charged particle inside a loop. Nor does a loop require every internal momentum to be off-shell at every point of the integration.

Can changing a vacuum produce real photons?

Yes, if a physical system supplies energy. In a 2011 superconducting-circuit experiment, researchers rapidly changed the electromagnetic boundary condition of a transmission line. They detected microwave photons and two-mode squeezing, a quantum correlation in the radiation.

This dynamical Casimir effect differs from the force between stationary plates. A drive changes the system in time and supplies the emitted energy.

The phrase “virtual photons become real” offers one interpretation. The direct experimental claim is more specific: modulation of the circuit generated correlated radiation. It did not extract unlimited energy from an undisturbed vacuum.

Why the Hawking pair story is only an analogy

The familiar picture places a particle pair near a black-hole horizon. One partner falls in; the other escapes. It can suggest the result, but it is not a literal account of the calculation.

In Hawking’s 1975 work, quantum fields on a collapsing black-hole background produce an outgoing thermal flux. The calculation concerns field modes and the definition of particles in that spacetime. It does not require identifiable virtual pairs to split at a specific point.

The broader lesson applies throughout this article. Identify the measured quantity, the complete theoretical prediction and the assumptions connecting them. Virtual particles are useful components of that prediction. Treating each one as a short-lived object adds claims that the experiment has not tested.