---
title: Multitrace in Mathematics, Physics, & Numerical Analysis
url: https://www.emergentmind.com/topics/multitrace
type: topic
---

# Multitrace in Mathematics, Physics, & Numerical Analysis

Searching arXiv for recent papers on “multitrace” and the specified article to ground the encyclopedia entry in current literature.
arXiv search query: `all:multitrace`
Multitrace is a technical term that appears in several distinct settings in contemporary mathematics, mathematical physics, and numerical analysis. In associative algebra it denotes a refinement of the ordinary trace adapted to semisimple quotients with several simple matrix blocks [2510.05820]. In matrix models and fuzzy field theory it denotes effective interactions built from products of traces, such as \((\mathrm{Tr}M^2)^2\) or \(\mathrm{Tr}M\,\mathrm{Tr}M^3\), that arise after integrating out angular degrees of freedom [2006.13577, 1003.4683]. In AdS/CFT it denotes deformations \(W(\mathcal O)\) built from powers of a single-trace operator and encoded by modified boundary conditions [1005.4921, 1911.12328]. In computational PDE it denotes multitrace formulations, where Dirichlet and Neumann traces on subdomain boundaries are coupled through Calderón projectors [1605.04422, 1701.03084]. A plausible unifying description is that the term labels constructions in which a single trace datum is replaced either by several trace components or by interactions among several trace quantities.

## 1. Terminological range

The term has no single discipline-independent meaning. In ring theory, the central object is a trace-like invariant attached to the simple components of the semisimple quotient of a finite-dimensional algebra [2510.05820]. In random-matrix and fuzzy-field-theory work, the defining feature is an action depending on several trace monomials rather than on a single-trace potential \(\operatorname{Tr}V(M)\) [2006.13577, 1412.6255]. In holography, multitrace refers to deformations of a conformal field theory by a functional \(W(\mathcal O)\), with double-trace deformations as the most prominent special case [1005.4921, 2411.03297]. In boundary-integral methods, multitrace formulations rewrite transmission problems in terms of trace data carried separately by each subdomain [1605.04422].

These uses are not interchangeable. The algebraic multitrace of an element \(a\in A\), the multitrace matrix-model potential of a Hermitian matrix \(M\), and a holographic multitrace deformation \(W(\mathcal O)\) are different constructions with different domains of definition. What they share is structural rather than literal: each replaces a single trace condition by a multi-component or multi-trace object. This suggests that “multitrace” functions less as a universal definition than as a family resemblance across trace-based formalisms.

## 2. Multitrace as an invariant in associative algebra

In "Commutators on Generalized Block-Triangular Algebras" [2510.05820], multitrace is introduced for a finite-dimensional unital associative algebra \(A\) over an algebraically closed field \(K\) of characteristic zero. Choosing a Wedderburn–Malcev decomposition
\[
A = B \oplus \operatorname{rad}(A),
\]
with semisimple part
\[
B \simeq M_{d_1}(K) \oplus \cdots \oplus M_{d_r}(K),
\]
one writes \(a\in A\) as \(a=b+j\), where \(b\in B\) and \(j\in \operatorname{rad}(A)\). Under the above isomorphism, \(b\) corresponds to a tuple \((b_1,\dots,b_r)\) with \(b_i\in M_{d_i}(K)\), and the multitrace is defined by
\[
\operatorname{mtr}(a) = \{ \operatorname{tr}(b_1), \operatorname{tr}(b_2), \dots, \operatorname{tr}(b_r) \}.
\]
The braces indicate a multiset, so the order of components is irrelevant. The condition \(\operatorname{mtr}(a)=0\) means \(\operatorname{tr}(b_i)=0\) for all \(i\).

A central point is well-definedness. Although the semisimple complement \(B\) is not unique, the property of having multitrace zero does not depend on the choice of \(B\). The reason given is that any other semisimple complement is conjugate to \(B\) by an inner automorphism coming from \(1+r\) with \(r\in\operatorname{rad}(A)\), and inner automorphisms preserve the ordinary trace on each matrix block [2510.05820].

The main theorem identifies multitrace as the exact commutator obstruction for generalized block-triangular algebras. Writing \(1_A=e_1+\cdots+e_r\) for the primitive central idempotents corresponding to the simple blocks of \(B\), and defining Peirce components
\[
\operatorname{rad}(A)_{ij}:=e_i\operatorname{rad}(A)e_j,
\]
the algebra is called generalized block-triangular when
\[
\operatorname{rad}(A)_{ij}=0 \quad \text{for all } i\ge j.
\]
For this class, an element \(a\in A\) is a commutator if and only if it has multitrace zero [2510.05820]. The theorem extends the Albert–Muckenhoupt–Shoda characterization for \(M_n(K)\), where ordinary trace zero is equivalent to being a commutator.

The proof strategy is inductive on the number \(r\) of simple components. The easy direction is that every commutator has multitrace zero, because commutators in matrix algebras have trace zero on each block. The converse uses the block-triangular structure to split a multitrace-zero element into a part in a smaller generalized block-triangular subalgebra, a part in the last semisimple block, and an off-diagonal radical term. The last term is absorbed by solving a Sylvester-type equation
\[
j_r=a_0' z-zc'
\]
in a radical bimodule, using a module-theoretic Sylvester–Rosenblum theorem and a scalar shift that produces spectral disjointness [2510.05820].

Two consequences are emphasized. First, in these algebras the set of commutators is closed under addition, because
\[
c(A)=\{[a,b]:a,b\in A\}=\{a\in A:\operatorname{mtr}(a)=0\}
\]
is a vector space. Second, for a finite-dimensional algebra whose semisimple quotient has \(r\) simple components, one always has \(\dim_K A/[A,A]\ge r\), and generalized block-triangular algebras attain this lower bound exactly [2510.05820]. In this setting, multitrace is therefore the correct multi-component analogue of the ordinary trace.

## 3. Multitrace matrix models and fuzzy scalar field theory

In matrix-model and fuzzy-field-theory literature, a multitrace model is an effective matrix theory whose action depends on products of traces and on combinations of moments, rather than only on \(\operatorname{Tr}M^k\). The standard mechanism is angular integration. A scalar field on a fuzzy space is represented by an \(N\times N\) Hermitian matrix \(M\), and the kinetic term is not invariant under arbitrary unitary rotations. After diagonalizing \(M=U^\dagger \Lambda U\), the difficult \(U\)-integral is packaged into an effective eigenvalue action, which naturally acquires multitrace structure [2006.13577].

For the fuzzy sphere, the review "Multitrace matrix models of fuzzy field theories" [2006.13577] writes the scalar action as
\[
S[M]= \operatorname{Tr}\!\left(\frac{1}{2} m^2 M^2 + g M^4 + \frac{1}{2} M \mathcal{K} M\right),
\qquad
\mathcal{K}M = [L_i,[L_i,M]].
\]
Because the kinetic term breaks unitary invariance, the angular integral cannot be performed exactly. The effective action is then expressed in terms of symmetrized moments
\[
t_n=\operatorname{Tr}\!\left(M-\frac{1}{N}(\operatorname{Tr}M)\mathbb{I}\right)^n,
\]
and the phrase multitrace matrix model refers to the resulting dependence on several such invariants. The same review describes a nonperturbative second-moment approximation
\[
S_{\mathrm{eff}}=\frac12 F(t_2)+R,
\qquad
F(t_2)=\log\!\left(\frac{t_2}{1-e^{-t_2}}\right),
\]
as well as a perturbative effective action involving
\[
-\frac{1}{432}t_3^2-\frac{1}{3456}\left(t_4-2t_2^2\right)^2.
\]
Its central conclusion is that higher moments such as \(t_3\) and \(t_4-2t_2^2\) must be included nonperturbatively to reproduce the phase structure reliably [2006.13577].

A more explicit derivation appears for fuzzy \(\mathbf{CP}^n\) in "The Multitrace Matrix Model of Scalar Field Theory on Fuzzy CP\(^n\)" [1003.4683]. There the field \(\Phi\) has action
\[
S[\Phi]=\operatorname{tr}\!\left(\Phi[L_i,[L_i,\Phi]]+r\,\Phi^2+g\,\Phi^4\right),
\]
and a high-temperature expansion of the kinetic term to order \(\beta^3\), together with Haar orthogonality, tensor-product decomposition into irreducible representations, Young projectors, and characters, yields an effective multitrace model. After re-exponentiation, the action takes the form \(S=S_1+S_2+S_3\), with leading term
\[
S_1=\frac{\operatorname{tr}(K)}{N^2-1}\operatorname{tr}(\Phi^2) -\frac{\operatorname{tr}(K)}{N^3-N}\operatorname{tr}(\Phi)^2.
\]
The higher-order terms contain \(\operatorname{tr}(\Phi^4)\), \(\operatorname{tr}(\Phi^2)^2\), and \(\operatorname{tr}(\Phi^3)\operatorname{tr}(\Phi)\), among others. The paper analyzes the resulting single-cut, double-cut, and asymmetric single-cut saddle solutions and finds three phases for \(\mathbf{CP}^1\), \(\mathbf{CP}^2\), and \(\mathbf{CP}^3\) [1003.4683].

The same program was reformulated by a bootstrapping method in "Bootstrapping Fuzzy Scalar Field Theory" [1412.6255]. Instead of computing all angular integrals directly, the paper constrains the multitrace action through shift symmetry, differential operators acting on the exponential of the action, and higher derivatives at \(\Phi=0\). The kinetic part is organized as
\[
S_{\mathrm{MT,kin}}[\Phi] = \sum_T a_T\, s(T),
\]
where \(T\) runs over partitions and \(s(n)\) denotes shift-invariant trace polynomials such as
\[
s(2)=\mathrm{tr}(\Phi^2)-\frac{1}{N}\mathrm{tr}(\Phi)^2.
\]
The coefficients are computed through fourth order in \(\beta\), confirming large-\(N\) constraints previously derived by Polychronakos while showing that an implicit expectation about the shape of the multitrace terms is false [1412.6255].

A focused two-dimensional analysis is given in "A Multitrace Approach to Noncommutative \(\Phi_2^4\)" [1410.4881]. There the kinetic term of noncommutative scalar theory on the fuzzy sphere or the regularized Moyal-Weyl plane is converted into an effective multitrace correction
\[
\Delta V = \frac{r^2}{4}\!\left(v_{2,1}T_2+\frac{2N}{3}w_1 t_2\right) +\frac{r^4}{24}\!\left( v_{4,1}T_4-\frac{4}{N^2}v_{2,2}T_2^2 +4w_2(t_1t_3-t_2^2) +\frac{4}{N}w_3 t_2T_2 \right)+O(r^6),
\]
with
\[
T_2=Nt_2-t_1^2, \qquad T_4=Nt_4-4t_1t_3+3t_2^2.
\]
In the \(M\to -M\) symmetric truncation this becomes a doubletrace quartic matrix model, and the resulting saddle-point problem yields an analytic prediction for the one-cut to two-cut transition line and a lower estimate for the triple point [1410.4881].

## 4. Phase structure, asymmetry, and emergent geometry

A recurring theme in multitrace matrix models is that multitrace couplings alter the phase structure of the quartic one-matrix model. In the basic real quartic model
\[
V_0 = B\,\mathrm{Tr} M^2 + C\,\mathrm{Tr} M^4,
\]
the disordered phase and the non-uniform ordered phase are stable, while the uniform ordered phase is metastable. "The multitrace matrix model: An alternative to Connes NCG and IKKT model" [1608.02758] generalizes this to a quartic multitrace potential
\[
V_1 = D(\mathrm{Tr} M^2)^2 + B'(\mathrm{Tr} M)^2 + C'\,\mathrm{Tr}M\,\mathrm{Tr}M^3 + D'(\mathrm{Tr} M)^4 + A'\,\mathrm{Tr}M^2(\mathrm{Tr}M)^2 + \cdots .
\]
The term \(\mathrm{Tr}M\,\mathrm{Tr}M^3\) is singled out as crucial for stabilizing the Ising-like uniform ordered phase. For the coupling choice
\[
D = 3N,\qquad B' = \sqrt{N},\qquad C = -N,\qquad D' = 0,\qquad A' = 0,
\]
the paper states that the critical behavior across the Ising line matches the Onsager universality class [1608.02758].

The same work then decomposes a \(2N\times 2N\) matrix as
\[
M=M_0\mathbf{1}_{2N}+M_1,\qquad \mathrm{Tr}M_1=0,\qquad M_1=\sigma_a X_a,\qquad M_0=a+m,
\]
and derives an effective \(SO(3)\)-symmetric three-matrix potential
\[
V = -C\,\mathrm{Tr}[X_a,X_b]^2 + 2C\,\mathrm{Tr}(X^2)^2 + 4D(\mathrm{Tr}X^2)^2 + 2(B + B_0 a^2)\mathrm{Tr}X^2 + 2i a \gamma\, \epsilon_{abc}\,\mathrm{Tr}X_aX_bX_c.
\]
Here the first term is identified as Yang–Mills, the cubic term as Myers/Chern–Simons, and the crucial observation is that the Chern–Simons term is proportional to the order parameter \(a\), so it is nonzero only in the uniform ordered phase [1608.02758]. This is the mechanism by which a fuzzy two-sphere and its non-commutative gauge theory emerge dynamically.

Several works develop the same picture through phase diagrams. "Phase diagram of scalar field theory on fuzzy sphere and multitrace matrix models" [1601.05628] distinguishes the symmetric one-cut, asymmetric one-cut, and symmetric two-cut phases and finds a triple point at
\[
g_c \approx 0.02
\]
within a nonperturbative approximation, to be compared with the numerical estimate \(0.125\text{--}0.15\). "Phase diagrams of the multitrace quartic matrix models of noncommutative \(\Phi^4\)" [1509.03726] reconstructs the Ising, matrix, and stripe boundaries using an exact Metropolis algorithm in eigenvalues and reports a triple point highlighted in the abstract and conclusion as
\[
(\tilde b,\tilde c)=(-1.55,0.4),
\]
with measured Ising critical exponents \(\nu=1\), \(\beta=\frac18\), \(\gamma=\frac74\), \(\alpha=0\), and \(\eta=\frac14\), in agreement with the Onsager values [1509.03726].

The interpretation of these phases as indicators of geometry is developed explicitly in "Quantized Noncommutative Geometry from Multitrace Matrix Models" [2110.06677] and "Quantum Gravity as a Multitrace Matrix Model" [1706.07724]. Both works identify the stable uniform-ordered phase as the signal of an emergent geometric regime, while the stripe or non-uniform ordered phase is tied to noncommutativity and UV/IR mixing. The latter paper formulates the guiding principles
\[
\text{I: uniform second-order phase transition} \Rightarrow \text{geometry},
\]
\[
\text{II: critical exponents} \Rightarrow \text{dimension},
\]
\[
\text{III: Wigner semicircle law} \Rightarrow \text{metric},
\]
and argues that a single Hermitian matrix with multitrace interactions can sustain emergent geometry, growing dimensions, and topology change [1706.07724].

More recent work sharpens the asymmetric sector. "Cubic asymmetric multitrace matrix model" [2407.20014] studies the action
\[
S(M)=\frac{1}{N}\left(\frac12r\,\mathrm{tr}M^2+g\,\mathrm{tr}M^4-c_1c_3\right),
\qquad
c_n=\frac{1}{N}\mathrm{tr}(M^n),
\]
and shows that the \(-c_1c_3\) deformation stabilizes an asymmetric one-cut phase. For \(t=-1\) the numerical saddle-point analysis yields a triple point at
\[
g_c=+1.732525\pm0.000005,\qquad r_c=-5.26502\pm0.00001,
\]
and no stable asymmetric two-cut phase is found [2407.20014]. This provides a clean example in which multitrace interactions change the physical phase diagram without changing the symmetric sector qualitatively.

## 5. Multitrace deformations in AdS/CFT and holography

In holography, multitrace deformations are boundary deformations of a conformal field theory by a functional of an operator \(\mathcal O\). "Multitrace deformations, Gamow states, and Stability of AdS/CFT" [1005.4921] studies large-\(N\) CFTs deformed by
\[
I_{\text{CFT}}+\int d^dx\,W(\mathcal O),
\]
and maps such deformations into boundary terms for a bulk scalar in \(AdS_{d+1}\). Near the boundary,
\[
\phi=\alpha z^{\Delta_-}+\beta z^{\Delta_+},
\]
and the generalized boundary condition is
\[
\left[z\phi'-\Delta_-\phi\right]_{\epsilon}=\left[\tilde W'(\phi)\right]_{\epsilon}.
\]
For a doubletrace deformation \(W=\frac{f}{2}\mathcal O^2\), the large-\(N\) beta function becomes
\[
\epsilon^{-1}\frac{d\bar f}{d\epsilon^{-1}}=\bar f^2-2\nu\,\bar f
\]
in the conformal case. In the Breitenlohner–Freedman window \(0<\nu<1\), the theory flows between two fixed points and exhibits a resonance at the scale separating the two CFT regimes; on the gravity side this resonance is an IR non-normalizable Gamow state [1005.4921].

A different application appears in "Light scalar modes from holographic deformations" [1911.12328]. There a bulk scalar in a slice of AdS\(_5\) is used to compare single-trace and multitrace deformations of a confining gauge theory. The multitrace deformation is a double-trace term
\[
W[\mathcal O]=-\frac{\xi}{\Lambda_{\rm UV}^{2\Delta-4}}\mathcal O^2,
\]
with no new elementary scalar. The holographic generating functional becomes
\[
\mathcal{S}_{\rm holo}[\varphi_0] = -\frac12\int d^4p\, \varphi_0(p)\, \frac{\Sigma(p)/g_5^2}{1-\xi\,\Sigma(p)/(g_5^2\Lambda_{\rm UV}^2)}\, \varphi_0(-p),
\]
and the physical masses satisfy
\[
\frac{1}{\xi}=\frac{\Sigma(p)}{g_5^2\Lambda_{\rm UV}^2}.
\]
The key conclusion is that multitrace deformation can produce a light scalar below the confinement scale that is a pure composite state, whereas the corresponding light mode in the single-trace setup is an elementary-composite admixture [1911.12328].

The nonlinear role of such deformations in global AdS is studied in "Multitrace deformations and the nonlinear stability of Anti-de Sitter space" [2411.03297]. For a top-down AdS\(_4\) Einstein-scalar system with \(m^2=-2\), the boundary condition is written as
\[
\beta=\beta_{BC}(\alpha),\qquad \beta_{BC}(\alpha)=\frac{\delta \mathcal W}{\delta \alpha},
\qquad
\mathcal W(\mathcal O)=\sum_{n=1}^{\infty}\frac{c_n}{n}\mathcal O^n.
\]
Because \(\Delta=1\), the operators \(\mathcal O\) and \(\mathcal O^2\) are relevant, \(\mathcal O^3\) is marginal, and \(\mathcal O^n\) for \(n\ge 4\) are irrelevant. The paper finds a one-to-one correspondence between the type of multitrace deformation and the resonant character of the AdS normal-mode spectrum: marginal and irrelevant deformations preserve exact resonance and the Bizoń–Rostworowski instability, while relevant double-trace deformations break exact resonance and destroy that instability at sufficiently small amplitude [2411.03297].

## 6. Multitrace formulations in domain decomposition and boundary integral equations

In numerical analysis, multitrace formulations are not deformations of an action and not products of traces. They are trace-based formulations of elliptic transmission problems in which each subdomain carries its own Dirichlet and Neumann trace unknowns. "An introduction to Multitrace Formulations and Associated Domain Decomposition Solvers" [1605.04422] develops this viewpoint from a one-dimensional model problem and then extends it to Lipschitz domains in arbitrary dimension.

For the model equation
\[
-\frac{d^2u}{dx^2}+a^2u=0
\]
on \(\mathbb R\setminus\{0\}\), the representation formula is
\[
u(x)=\left[\frac{du}{dx}\right]\mathcal G(x)-[u]\frac{d\mathcal G}{dx}(x),
\qquad
\mathcal G(x)=\frac{e^{-a|x|}}{2a}.
\]
The trace-to-potential composition \(\mathbb P_\pm=T_\pm\circ G_\pm\) yields Calderón projectors satisfying \(\mathbb P_\pm^2=\mathbb P_\pm\). For two subdomains, the multitrace system is
\[
(\Id-\mathbb P_1)U_1+\sigma_1(U_1-XU_2)=F_1,\qquad
(\Id-\mathbb P_2)U_2+\sigma_2(U_2-XU_1)=F_2,
\]
with a natural block Jacobi iteration. The eigenvalues of the iteration matrix are
\[
\pm\sqrt{\frac{\sigma_1}{1+\sigma_1}},\qquad
\pm\sqrt{\frac{\sigma_2}{1+\sigma_2}},
\]
so the convergence rate depends only on the relaxation parameters. In the limit \(\sigma_j=0\), the iteration matrix is nilpotent and the method converges in finitely many steps; the paper identifies this optimal multitrace solver with the optimal Schwarz method written in trace form [1605.04422].

"Multitrace formulations and Domain Decomposition Methods for the solution of Helmholtz transmission problems for bounded composite scatterers" [1701.03084] extends this framework to 2D Helmholtz transmission with piecewise constant material parameters. Using Green’s representation formula in each subdomain, the local multitrace formulation is expressed through the Calderón block operator
\[
\mathcal{C}_j=
\begin{bmatrix}
K_{\partial\Omega_j,j} & -\varepsilon_j S_{\partial\Omega_j,j}\\[2mm]
\varepsilon_j^{-1}N_{\partial\Omega_j,j} & -K_{\partial\Omega_j,j}^{\top}
\end{bmatrix},
\qquad
\mathcal{C}_j^2=\frac14 I,
\]
coupled by extension-by-zero operators between interfaces. The paper also studies a global skeleton version, g-MTF, and compares these formulations with non-overlapping DDMs using classical Robin or generalized Robin conditions [1701.03084].

A major practical conclusion is that generalized Robin interface conditions based on square-root Fourier multiplier approximations of Dirichlet-to-Neumann maps substantially reduce Krylov iteration counts. The corresponding approximate hypersingular symbol is
\[
p^N(\xi,k+i\sigma)=-\frac12\sqrt{|\xi|^2-(k+i\sigma)^2},
\]
and the resulting generalized Robin DDM performs markedly better than classical Robin DDM, while remaining cheap to implement in a Nyström setting [1701.03084]. In this literature, “multitrace” therefore denotes a trace-coupling architecture for domain decomposition, rather than a multitrace potential or deformation.

## 7. Multitrace amplitudes and other specialized usages

In scattering amplitudes, multitrace refers to color structures with several trace factors. "Expansion of All Multitrace Tree Level EYM Amplitudes" [1708.04514] studies tree-level Einstein–Yang–Mills amplitudes with an arbitrary number of gluon traces and gravitons, denoted schematically by
\[
A_{m,|\mathsf{H}|}(1,2\ldots r\,|\,\pmb{2}\,|\,\ldots\,|\,\pmb{m}\,\Vert\,\mathsf{H}).
\]
Using CHY formulae and BCFW recursion, the paper gives two recursive expansions that reduce an \(m\)-trace EYM amplitude to amplitudes with fewer traces and/or fewer gravitons. After iteration, any multitrace EYM amplitude can be expanded in the Kleiss–Kuijf basis of color-ordered Yang–Mills amplitudes, and the expansion coefficients provide BCJ numerators [1708.04514].

A closely related string-theoretic use appears in "String Correlators: Recursive Expansion, Integration-by-Parts and Scattering Equations" [1907.06041]. There a multitrace correlator in the compactified bosonic string or the holomorphic sector of the heterotic string has the form
\[
\mathcal{I}^{\rm string}_n(z)=R(i_1,\ldots,i_r)\prod_{t=1}^{m+1}\mathrm{PT}(W_t),
\]
with several Parke–Taylor factors \(\mathrm{PT}(W_t)\), one for each color trace. The paper derives a recursive integration-by-parts expansion that decomposes such correlators into a logarithmic part and lower-trace or lower-gluon correlators, and from this obtains a CHY formula for multitrace \((DF)^2+\mathrm{YM}+\phi^3\) amplitudes [1907.06041].

A distinct signal-processing usage appears in "Adaptive Multi-Trace Carving for Robust Frequency Tracking in Forensic Applications" [2005.06686]. There “multi-trace” refers not to algebraic traces but to several time-frequency paths in a spectrogram. The method tracks one frequency trace by dynamic programming, compensates for it, and repeats the procedure to reveal weaker traces. This usage is terminologically separate from the algebraic, holographic, and matrix-model senses, but it illustrates how the word can also denote multiple tracked trajectories rather than multiple trace functionals [2005.06686].

Across these specialized settings, the meaning of multitrace remains domain-dependent. In amplitudes it is a statement about color decomposition; in string correlators it is a product of several Parke–Taylor cycles; in signal processing it is an iterative extraction of multiple traces in a time-frequency representation. The recurrence of the term reflects a shared emphasis on multiplicity of trace-like structures, but not a single formal definition.

Source: https://www.emergentmind.com/topics/multitrace