An ode to instantons

arXiv:2603.06575 · hep-th, quant-ph · Submitted 2026-03-06 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: High Energy Physics: Theory. The day in high-energy theory: gravity, strings, fields and the questions left open.

Kai: We're looking at the paper: "An ode to instantons".

Title and authors: Kai: The title itself, "An ode to instantons," suggests a focus on these specific configurations that are central to semiclassical approximations in quantum mechanics >

Mira: It sounds poetic, but it really points to the mathematical structures they are using—the instantons—which are basically solutions to equations of motion that describe tunneling or transition processes >

Lev: For us, instantons are just a way of visualizing the dominant paths in the path integral, and this paper is showing how those paths can exist in different time regimes >

Kai: It's about using these instantons—the complex saddle points they mentioned—to give answers to standard problems like under-barrier transmission or over-barrier reflection >

Mira: They are taking a known technique from one area and applying it to make headway in another, which is what makes this work so interesting for us as theorists >

Lev: I wonder if the authors have found a way to make these instanton contributions more reliable when you move from simple systems like the harmonic oscillator to field theories with more fields >

The paper's summary: Kai: So, what they actually do is present this formalism for semiclassical time evolution using the method of steepest descent by finding these complex saddle points >

Mira: They express the wave function as a collection of contributions from these saddles, and the prefactor involves the determinant of a fluctuation operator which they compute using the Gelfand–Yaglom method >

Lev: That sounds like a lot of machinery. From an engineering perspective, computing that determinant is often where you run into numerical instabilities if you don't handle it right >

Kai: But they say this calculation connects back to known results in one-dimensional quantum mechanics, like the harmonic oscillator and under-barrier transmission >

Mira: And they show how coherent state evolution of the harmonic oscillator can be reproduced by solving differential equations in the complex plane, which is a neat link between these ideas >

Lev: So they're showing that this path integral approach isn't just abstract; it actually maps onto solvable problems we already know how to handle >

The paper's improvements: Kai: One area they focus on is the decay of metastable states, where they decompose the initial state into resonant energy levels and sum up contributions from various complex time contours >

Mira: This summation reveals that the probability inside the false vacuum decays exponentially at late times because of a combinatorial factor counting bounce multiplicity >

Lev: That combinatorial factor is what makes it complicated for us on hardware. If you have many bouncing solutions, you have to account for all of them, which is computationally heavy >

Kai: But they show that for times of order t equals one over gamma n, the collective contribution from all bouncing solutions with a bounce number l equal to one competes with the non-bouncing part >

Mira: And they compare this method against other ways of computing the wave function outside the false vacuum, like WKB or numerical Schrödinger equation solutions >

Conclusion: Kai: So, to wrap up, they've compared four methods for finding the wave function outside the false vacuum—WKB, indirect semiclassical path integral, numerical solutions—and they show that those first two methods only become good approximations after a time of order one over omega >

Mira: The main result is that if you find every relevant saddle point, all the path integral calculations match known WKB results for all examples >

Lev: That's solid. It means we can trust this formalism to reproduce established results in these cases, provided we can actually find those saddle points properly >

Kai: The big implication is that the path integral approach gives us a way to compute decay rates in quantum field theory with time dependence by studying these simpler quantum mechanics analogs >

Mira: And they leave the general problem of which complex saddle points contribute to the time-evolved state open for future work, speculating on whether the normalization constant for subleading saddle points is one or one over two times >

Lev: I just hope that NP equals one for all contributing saddle points, because if it’s half for solutions with bounces, that adds a layer of complexity we'd have to deal with when trying to implement this on real hardware >

Laboratory for Theoretical Fundamental Physics, EPFL · Institute for Theoretical Physics, KU Leuven · Dipartimento di Fisica, Sapienza Universita di Roma · INFN Sezione di Roma

hep-th, quant-ph

Submitted: 2026-03-06

Updated: 2026-11-18

Comments: In memory of Sidney Coleman on the 89th anniversary of his birth and the 19th anniversary of his passing. v2: minor adjustments throughout, and significant improvement of the argument in section 3.4.4. (comparison with numerics in an example)

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 75/100

The gist: We present a formalism for semiclassical time evolution in quantum mechanics, building on a century of work <ref:2603.06575#pg2> We identify complex saddle points in real time…

Key concepts

Semiclassical Formalism
This method approximates the path integral in the limit where Planck's constant ($\hbar$) approaches zero. It achieves this by finding specific points, called complex saddle points, in different time spaces (real and complex) that dominate the path integral's contribution to the quantum wave function.
Complex Saddle Points
These are crucial mathematical structures identified in the semiclassical method. They are not just simple peaks but can exist in real time, complex time, or both. They serve as the primary contributors to approximating the quantum evolution, linking quantum mechanics to classical solutions.
Decay of Metastable States
This involves studying how a system trapped in a false vacuum eventually decays. The formalism reveals that decay is governed by contributions from 'bouncing solutions' occurring across various complex time contours, which collectively determine the exponential decay probability over late times.

Terminology

Summary

We present a formalism for semiclassical time evolution in quantum mechanics, building on a century of work <ref:2603. p. 0

Where the paper says it

An ode to instantons Oliver Janssen1 , Joel Karlsson2,3 , Flavio Riccardi4,5 and Mattia Varrone1 1Laboratory for Theoretical Fundamental Physics, EPFL, 1015 Lausanne, Switzerland 2 Institute for Theoretical Physics, KU Leuven, 3001 Leuven, Belgium 3Leuven Gravity Institute, KU Leuven, 3001 Leuven, Belgium 4Dipartimento di Fisica, Sapienza Universit`a di Roma, Piazzale Aldo Moro 5, 00185, Roma, Italy 5 INFN sezione di Roma, Piazzale Aldo Moro 5, 00185, Roma, Italy Abstract We present a formalism for semiclassical time evolution in quantum mechanics, building on a century of work. We identify complex saddle points in real time, real saddle points in complex time, and complex saddle points in complex time that reproduce the known answers in classic problems.

06575#pg2> We identify complex saddle points in real time, real saddle points in complex time, and complex saddle points in complex time that reproduce the known answers in classic problems <ref:2603. p. 0
Where the paper says it

An ode to instantons Oliver Janssen1 , Joel Karlsson2,3 , Flavio Riccardi4,5 and Mattia Varrone1 1Laboratory for Theoretical Fundamental Physics, EPFL, 1015 Lausanne, Switzerland 2 Institute for Theoretical Physics, KU Leuven, 3001 Leuven, Belgium 3Leuven Gravity Institute, KU Leuven, 3001 Leuven, Belgium 4Dipartimento di Fisica, Sapienza Universit`a di Roma, Piazzale Aldo Moro 5, 00185, Roma, Italy 5 INFN sezione di Roma, Piazzale Aldo Moro 5, 00185, Roma, Italy Abstract We present a formalism for semiclassical time evolution in quantum mechanics, building on a century of work. We identify complex saddle points in real time, real saddle points in complex time, and complex saddle points in complex time that reproduce the known answers in classic problems.

06575#pg2> For the decay of a metastable state, we find finite time and finite energy analogs of the “bounce” which do not have strict zero or negative modes <ref:2603. p. 0
Where the paper says it

We identify complex saddle points in real time, real saddle points in complex time, and complex saddle points in complex time that reproduce the known answers in classic problems. For the decay of a metastable state, we find finite time and finite energy analogs of the “bounce” which do not have strict zero or negative modes.

06575#pg2> The motivation of this work is to learn how to compute decay rates in quantum field theory in situations with non-trivial time dependence, by first taking a humble step backwards to the fascinating world of quantum mechanics <ref:2603. p. 0
Where the paper says it

The motivation of this work is to learn how to compute decay rates in quantum field theory in situations with non-trivial time dependence, by first taking a humble step backwards to the fascinating world of quantum mechanics. In memory of Sidney Coleman on the occasion of his 89th birthday March 7, 2026 arXiv:2603.

06575#pg2>.

Motivation and Open Problems

This work concerns the semiclassical limit in the path integral approach to quantum theory <ref:2603.06575#pg2> Despite its widespread use, however, some aspects of the path integral approach remain insufficiently understood <ref:2603. p. 1

Where the paper says it

28 4 Conclusion 33 References 34 1 Motivation This work concerns the semiclassical limit in the path integral approach to quantum theory [1] – a venerable topic that hardly requires an introduction. Despite its widespread use, however, some aspects of the path integral approach remain insufficiently understood.

06575#pg2>. Especially in the context of tunneling phenomena in quantum field theory, where the path integral is our only analytic tool, and more specifically in situations with non-trivial time dependence, there are concrete, longstanding open problems <ref:2603. p. 1
Where the paper says it

Despite its widespread use, however, some aspects of the path integral approach remain insufficiently understood. 1 Especially in the context of tunneling phenomena in quantum field theory, where the path integral is our only analytic tool, and more specifically in situations with non-trivial time dependence, there are concrete, longstanding open problems, including 1.

06575#pg2>. The authors list two primary problems concerning time-dependent decay probability and a two-field inflation model <ref:2603. p. 1
Where the paper says it

1 Especially in the context of tunneling phenomena in quantum field theory, where the path integral is our only analytic tool, and more specifically in situations with non-trivial time dependence, there are concrete, longstanding open problems, including 1. What is the time-dependent decay probability Γ(t) per unit spacetime volume of a scalar field ϕ in flat space, with an initial spatially homogeneous but oscillating classical field profile ϕ(x, t) = ϕFV + ε sin(ωt)?

06575#pg2>.

Semiclassical Formalism

The method of steepest descent is used to approximate the path integral in the limit ħ → 0 by identifying complex saddle points <ref:2603.06575#pg2>. The resulting wave function is expressed as a collection of contributions from these saddles, given by formula (14) <ref:2603. p. 5

Where the paper says it

The background (z¯0, z¯) solves mz¨¯ = −∇V (z¯) with mz¯˙ (0) = i∇f(z¯0), and in the prefactor γ(t) becomes det γij (t) where Pd j=1 Fijγjk(t ′ ) = 0, γkl(0) = δkl, mγ˙kl(0) = i∂k∂lf(z¯0) and Fij = −mδij∂ 2 t ′ − ∂i∂jV (z¯). A word about determining the contributing saddles in the first place So far, we have discussed the contributions of saddle points to ψ(x, t), assuming they contribute in the first place.

06575#pg2>. The prefactor in this formula involves the determinant of the fluctuation operator, which can be computed using the Gelfand–Yaglom method <ref:2603. p. 5
Where the paper says it

where detD stands for the determinant with Dirichlet boundary conditions on [0, t]. We can compute this with the Gelfand–Yaglom method [26], which gives detD F = 2πℏ m β(t), (15) where β(t ′ ) is defined by Fβ = 0 , β(0) = 0, β˙(0) = 1 .

06575#pg2>. The calculation connects to known results in one-dimensional quantum mechanics, such as the harmonic oscillator and under-barrier transmission <ref:2603. p. 3
Where the paper says it

classic problems in one-dimensional quantum mechanics, including the harmonic oscillator (§3. 1), the transmission of a particle with insufficient energy to surpass a barrier (§3.

06575#pg2>.

Applications in Quantum Mechanics

The formalism is illustrated through several classic problems in one-dimensional quantum mechanics <ref:2603.06575#pg2>. This includes the coherent state evolution of the harmonic oscillator, where the solution is reproduced by solving differential equations in the complex plane <ref:2603. p. 1

Where the paper says it

Contents 1 Motivation 1 2 Semiclassical time evolution from the path integral 3 3 Examples 6 3. 1 Harmonic oscillator: coherent state evolution .

06575#pg2>. For under-barrier transmission, a complex time contour is used to solve boundary value problems on a half-line in the complex u-plane <ref:2603. p. 8
Where the paper says it

(35) Of course, for a starting position ¯z0 < a on the real line and keeping t ′ on the real axis, the particle never reaches x on the other side of the barrier where we want to evaluate the wave function. To circumvent this problem, we complexify t → u and propose instead to solve this boundary value problem on a half-line in the complex u-plane, for a given real x to the right of the feature.

06575#pg2>. Over-barrier reflection is studied by searching for trajectories satisfying specific boundary conditions on a complex time contour <ref:2603. p. 4
Where the paper says it

For real-valued ¯z, F is Hermitian on this space for the inner product ⟨w, v⟩ = Z t 0 dt ′ w(t ′ ) ∗ v(t ′ ), (9) so its eigenfunctions wn with eigenvalues λn, n ∈ N, can be normalized to form an orthonormal basis. We will expand the perturbation δz as δz(t ′ ) = α(t ′ )δz0 + w(t ′ ), (10) where α is the function satisfying Fα = 0 with boundary conditions α(0) = 1, α(t) = 0 and w is an arbitrary function with w(0) = w(t) = 0 that we expand as w(t ′ ) = X∞ n=0 cnwn(t ′ ), ⟨wn, wm⟩ = δnm , (11) for coefficients cn ∈ R.

06575#pg2>.

Decay of Metastable States

The decay of a metastable state is studied by decomposing the initial state into resonant energy levels and summing over contributions from various complex time contours <ref:2603.06575#pg2>. This summation reveals that the probability inside the false vacuum decays exponentially at late times due to a combinatorial factor counting bounce multiplicity <ref:2603. p. 0

Where the paper says it

For the decay of a metastable state, we find finite time and finite energy analogs of the “bounce” which do not have strict zero or negative modes. The one-loop phase of the wave function and the multiplicity of bounce solutions at late times is discussed.

06575#pg2>. The analysis shows that for times of order t = O(1/Γn), the collective contribution of all bouncing solutions with l = O(1) competes with the non-bouncing contribution <ref:2603. p. 4
Where the paper says it

Since ¯z(t ′ ) ∈ C this is one complex boundary condition, and because we have one complex parameter ¯z0 to tune, we generally expect a discrete set of solutions. To calculate the possible contribution to ψ(x, t) from the vicinity of a saddle (¯z0, z¯), we expand the exponent in (1) to second order in the perturbations z = ¯z + δz, z0 = ¯z0 + δz0: iS[z]−f(z0) = iS[¯z]−f(¯z0)− i 2 Z t 0 dt ′ δz(t ′ ) m∂2 t ′ + V ′′(¯z) δz(t ′ )+im 2 [δzδz ˙ ] t 0− f ′′(¯z0) 2 δz2 0+O(δ 3 ).

06575#pg2>.

Comparison and Conclusion

The paper compares four methods for computing the wave function outside the false vacuum, including WKB, indirect (complex time) semiclassical path integral, numerical solution to the Schrödinger equation, and direct (real time) semiclassical path integral <ref:2603.06575#pg2>. The results show that the first two methods are only expected to become valid approximations after a time of order 1/ω <ref:2603.06575#pg2>. The paper concludes by speculating on the normalization constant NP for subleading saddle points, suggesting it might be 1 for the special saddle and 1/2 for solutions with bounces <ref:2603. p. 4

Where the paper says it

Plugging (10) into (6) we find iS[z] − f(z0) = iS[¯z] − f(¯z0) + i 2 X∞ n=0 λnc 2 n − 1 2 [f ′′(¯z0) + imα˙(0)] δz2 0 + · · · . (12) Therefore, from the vicinity of this saddle, defining the integration measure [dy(t ′ )] = NP Y∞ n=0 dcn (13) up to a constant normalization factor NP , and integrating the cn’s over R(1 + iε), we find ψ(x, t) ⊃ g(¯z0) exp i ℏ S[¯z] − f(¯z0) ℏ NP s 2πℏ f ′′(¯z0) + imα˙(0) 1 √ i detD F [1 + O(ℏ)] as ℏ → 0 , (14)

06575#pg2>. The authors also note that in all examples, assuming every relevant saddle point was found, the path integral calculations were shown to match known WKB results <ref:2603. p. 6
Where the paper says it

4 Instead, we will solve it implicitly in the examples we discuss below by identifying a subset of all saddle points and showing that these yield the correct result (known independently, via the WKB method or numerics). 4 we propose a concrete, non-trivial example to which a path integral enthusiast could attempt to apply the method of [23] to determine the set of relevant saddles, and their normalizations NP , from first principles.

06575#pg2>.

Resonant Transmission

Resonant transmission is derived by combining complex time contours with the multi-bounce summation of solutions <ref:2603.06575#pg2>. The resulting transmission probability exhibits a Breit–Wigner profile, demonstrating resonant enhancement when the energy matches a resonant eigenstate <ref:2603. p. 1

Where the paper says it

4 Comparison of various methods in an explicit example . 5 Resonant transmission .

06575#pg2>. This enhancement can reach unit probability for strictly symmetric barriers with unit probability <ref:2603. p. 1
Where the paper says it

1 Especially in the context of tunneling phenomena in quantum field theory, where the path integral is our only analytic tool, and more specifically in situations with non-trivial time dependence, there are concrete, longstanding open problems, including 1. What is the time-dependent decay probability Γ(t) per unit spacetime volume of a scalar field ϕ in flat space, with an initial spatially homogeneous but oscillating classical field profile ϕ(x, t) = ϕFV + ε sin(ωt)?

06575#pg2>. The required initial wave packet must be exponentially broader than the double-barrier system itself to observe this effect <ref:2603. p. 7
Where the paper says it

2 Under-barrier transmission In this section, we study scattering by computing the time evolution of semiclassical energy eigenstates with energy E < maxR V . The dynamics of such states is completely captured by their transmission coefficient, through which one can subsequently reconstruct the dynamics of any localized wave packet.

06575#pg2>.

Zero Modes and Negative Modes

In the standard discussion, the imaginary time version of the fluctuation operator has a zero mode proportional to ∂τ zb, which corresponds to tunneling <ref:2603.06575#pg2>. In this formalism, where F does not have a zero mode because of finite energy above the false vacuum, an enhancement factor proportional to the total time comes from large amounts of inequivalent bounces giving identical contributions <ref:2603.06575#pg2>. The negative mode is absent in this case but its presence was implicitly accounted for by avoiding WKB turning points in the complex u-plane <ref:2603. p. 8

Where the paper says it

It remains real-valued and reaches the turning point b after a negative imaginary time duration τa→b ≡ Z b a m κ . (40) 5 In other words, we want to find the functions of t ∈ R for given x ∈ R appearing in the prefactor and exponent in (18), but, instead, we calculate them on a half-line in the complex u-plane, assume analyticity in an appropriate region, and then uniquely analytically continue u → t.

06575#pg2>.

Dominant vs. p. 6

Where the paper says it

(26) This is a coherent state: its time evolution is ψ(x, t) = N0 e −iωt/2 exp − (x − a cos ωt) 2 2d 2 + i a sin ωt(a cos ωt − 2x) 2d 2 . (27) 4Some obvious constraints on contributing saddles include 1) at early enough times, saddles which are naively more dominant than the special saddle that is continuously connected to the identity cannot be relevant, and 2) saddles can only become dominant as t crosses t∗ if they were contributing subdominantly for t < t∗.

Subdominant Contributions

Bouncing solutions give contributions to psi that are exponentially suppressed as ħ → 0 compared to the two non-bouncing contributions <ref:2603.06575#pg2>. Specifically, individual bouncing contributions are subdominant compared to the uncomputed [1 + O(ħ)] correction to the non-bouncing contribution <ref:2603.06575#pg2>. However, at late times, bouncing contributions receive a combinatorial enhancement because there are many giving the same contribution <ref:2603.06575#pg2>. The relative magnitude of these bouncing solutions compared to the non-bouncing contribution is given by 1/l! p. 4

Where the paper says it

Since ¯z(t ′ ) ∈ C this is one complex boundary condition, and because we have one complex parameter ¯z0 to tune, we generally expect a discrete set of solutions. To calculate the possible contribution to ψ(x, t) from the vicinity of a saddle (¯z0, z¯), we expand the exponent in (1) to second order in the perturbations z = ¯z + δz, z0 = ¯z0 + δz0: iS[z]−f(z0) = iS[¯z]−f(¯z0)− i 2 Z t 0 dt ′ δz(t ′ ) m∂2 t ′ + V ′′(¯z) δz(t ′ )+im 2 [δzδz ˙ ] t 0− f ′′(¯z0) 2 δz2 0+O(δ 3 ).

Γnt 2l <ref:2603. p. 0
Where the paper says it

The motivation of this work is to learn how to compute decay rates in quantum field theory in situations with non-trivial time dependence, by first taking a humble step backwards to the fascinating world of quantum mechanics. In memory of Sidney Coleman on the occasion of his 89th birthday March 7, 2026 arXiv:2603.

06575#pg2>.

Final Summary

The study successfully reproduces known WKB results in all examples, confirming the formalism's validity for semiclassical states <ref:2603.06575#pg2>. The key takeaway is that the path integral approach provides a way to compute decay rates in quantum field theory with non-trivial time dependence by studying analogs in quantum mechanics <ref:2603. p. 0

Where the paper says it

The motivation of this work is to learn how to compute decay rates in quantum field theory in situations with non-trivial time dependence, by first taking a humble step backwards to the fascinating world of quantum mechanics. In memory of Sidney Coleman on the occasion of his 89th birthday March 7, 2026 arXiv:2603.

06575#pg2>. The paper leaves open the general problem of determining which complex saddle points contribute to the time-evolved state <ref:2603. p. 2
Where the paper says it

Our contribution is the explicit calculation of the one-loop prefactor, up to a normalization, which we leave as an open question and speculate upon in the conclusion. We also connect with more recent discussions about which saddle points contribute to the path integral [23], setting up the precise question that would determine this for our problem, but unfortunately we are not able to solve it in general.

06575#pg2>. The final speculation is whether NP = 1 for all contributing saddle points or NP = 1/2l for solutions with l bounces <ref:2603. p. 5
Where the paper says it

A word about determining the contributing saddles in the first place So far, we have discussed the contributions of saddle points to ψ(x, t), assuming they contribute in the first place. Firstly, note that for non-quadratic systems the set of solutions to the classical boundary value problem (4)-(5) is generally infinite.

06575#pg2>. The work provides a framework to study quantum tunneling from the path integral perspective <ref:2603. p. 2
Where the paper says it

One dreams to go beyond plausible estimates, however, and calculate the detailed spectrum of masses of these black holes, in particular the dependence on the functional form of the barrier V (ϕ1, ϕ2). To study quantum tunneling from the path integral perspective, the standard approach of our Teacher [13] is to rotate to imaginary time, and (implicitly) assume the initial state is the would-be ground state of the false vacuum.

06575#pg2>. The paper is supported by research grants from the Italian Ministero dell’Universitâ e Ricerca and the Research Foundation - Flanders <ref:2603.06575#pg2>.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Improved AI can compute semiclassical time evolution by identifying complex saddle points in real time, real saddle points in complex time, and complex saddle points in complex time that reproduce known answers in classic problems.

  2. Improved AI can calculate finite time and finite energy analogs of the “bounce” for the decay of a metastable state which do not have strict zero or negative modes.

  3. Improved AI can compute decay rates in quantum field theory with non-trivial time dependence by first taking a step backwards to quantum mechanics, specifically in one-dimensional quantum mechanics problems like the harmonic oscillator and under-barrier transmission.

  4. Improved AI can solve for the wave function evolution using the path integral approach by expanding the exponent to second order in perturbations, allowing for an explicit calculation of the one-loop prefactor up to a normalization factor.

  5. Improved AI can determine whether saddle points contribute to a timeevolved state by searching for a real-valued function r(t) that flows to the saddle under specific boundary conditions.

  6. Improved AI can analyze resonant transmission in two barriers by combining complex time contours with the multi-bounce summation of solutions characterized by bounce multiplicities (l0, l1, l2).

  7. Improved AI can derive the Breit-Wigner profile for transmission probability by relating the sum over bouncing solutions to a characteristic resonant phase shift theta res.

  8. Improved AI can compute resonant enhancement in transmission by showing that exactly at resonance (delta equals zero), the transmission probability is T 2 = 4Y1Y2/(Y1 + Y2)2 when barriers are symmetric.

Abstract

We review a formalism for semiclassical time evolution in quantum mechanics, identifying complex saddle points in real time, real saddle points in complex time, and complex saddle points in complex time that reproduce known answers to classic problems. For the decay of a metastable state, we find finite-time and finite-energy analogs of the "bounce" which have no zero or negative modes. The one-loop phase of the wave function and the multiplicity of bounce solutions at late times are discussed. The motivation for this work is to learn how to compute decay rates in quantum field theory in situations with interesting time dependence, by first taking a humble step back to the fascinating world of quantum mechanics.

Sources

Related papers