QUANTUM ERROR-DETECTION AT LOW ENERGIES
2.4 On Expectation Values of Local Operators in MPS
2.4.2 Transfer Matrix Techniques
Here we establish some essential statements for the analysis of transfer operators. In Section 2.4.2, we introduce generalized (non-standard) transfer operators: these can be used to express the matrix elements of the form hΞ¨|πΉ|Ξ¨0i of local operators πΉ with respect to pairs of MPS (Ξ¨,Ξ¨0). In Section 2.4.2, we establish bounds on the norm of such operators. Relevant quantities appearing in these bounds are the second largest eigenvalueπ
2of the transfer matrix, as well as the sizes of its Jordan blocks.
More General and Mixed Transfer Operators
Consider a single-site operator π β B (Cp). The generalized transfer matrixπΈπ β B (Cπ· βCπ·)is defined as
πΈπ =βοΈ
π,π
hπ|π|πiπ΄π β π΄π . (2.17) We further generalize this as follows: if π
1, . . . , ππ β B (Cp), then πΈπ
1βΒ·Β·Β·βππ β B (Cπ· βCπ·)is the operator
πΈπ
1βΒ·Β·Β·βππ =πΈπ
1Β· Β· Β·πΈπ
π .
This definition extends by linearity to any operator πΉ β B ( (Cp)βπ), and gives a corresponding operatorπΈπΉ β B (Cπ· βCπ·). The tensor network diagrams for these definitions are given in Figure 2.2, and the composition of the corresponding maps is illustrated in Figure 2.3.
In the following, we are interested in inner productshΞ¨(π΄, π , π) |Ξ¨(π΅, π , π)i of two MPS, defined by local tensors π΄ and π΅, with boundary matrices π andπ, which may have different bond dimensions π·
1and π·
2, respectively. To analyze these, it is convenient to introduce an βoverlapβ transfer operator πΈ = πΈ(π΄, π΅) which now depends on both MPS tensors π΄andπ΅. First, we defineπΈ β B (Cπ·1 βCπ·2) by
πΈ =
p
βοΈ
π=1
π΄π β π΅π .
Figure 2.3: The product πΈπ
1βπ
2 = πΈπ
1πΈπ
2 of two transfer operators. Left- multiplication by an operator corresponds to attaching the corresponding diagram on the left.
The definition of πΈπ for π β B (Cp) is analogous to Equation (2.4.2), but with appropriate substitutions. We set
πΈπ =βοΈ
π,π
hπ|π|πiπ΄π β π΅π .
Starting from this definition, the expressionπΈπΉ β B (Cπ·1βCπ·2)forπΉ β B ( (Cp)βπ) is then defined analogously as before.
Norm Bounds on Generalized Transfer Operators
A first key observation is that the (operator) norm of powers of any transfer operator scales (at most) as a polynomial in the number of physical spins, with the degree of the polynomial determined by the size of the largest Jordan block. We need these bounds explicitly and start with the following simple bounds.
Below, we often consider families of parameters depending on the system size π, i.e., the total number of spins. We writeπ βas a shorthand for a parameterπ
βbeing sufficiently largeβ compared to another parameter β. More precisely, this signifies that we assume that |β/π| β 0 forπ β β, and that by a corresponding choice of a sufficiently largeπ, the term |β/π|can be made sufficiently small for a given bound to hold. Oftentimesβwill in fact be constant, withπ β βasπβ β. Lemma 2.4.1. For π > β, the Frobenius norm of the π-th power (π πΌ+π)π of a Jordan blockπ πΌ +π β B (Cβ) with eigenvalueπ, such that |π| β€ 1, and size β is bounded by
k (π πΌ+π)πkπΉ β€ 3β3/2πββ1|π|πβ(ββ1) . (2.18) Furthermore,
k (π πΌ+π)π k β€4πββ1 forπ β . (2.19)
Proof. Forβ =0 the claim is trivial. Assume thatβ >1. Because πβ =0 andππ has exactlyββπ non-zero entries forπ < β, we have
k (π πΌβ+π)πkπΉ β€
ββ1
βοΈ
π=0
π π
|π|πβπ kππkπΉ
β€ |π|πΒ· |π|β(ββ1)
ββ1
βοΈ
π=0
π π
(ββπ)1/2
β€ β1/2|π|π Β· |π|β(ββ1)
ββ1
βοΈ
π=0
π π
.
Since the right hand side is maximal forπ =ββ1, and the binomial coefficient can be bounded from above by ππ
β€ ππ
π
π
β€ 3Β·ππ, we obtain
ββ1
βοΈ
π=0
π π
β€ 3βΒ·πββ1, hence, the first claim follows.
For the second claim, recall that the entries of theπ-th power of a Jordan block are ( (π πΌβ+π)π)π,π =


ο£²

ο£³
π πβπ
ππ+(πβπ) ifπ β₯ π
0 otherwise
(2.20)
forπ, π β [β]. This means that if|π|=1, the maximum matrix element| ( (π πΌβ+π)π)π,π|=
π ββ1
is attained for (π, π) = (1, β). Using the Cauchy-Schwarz inequality, it is straightforward to check that
k (π πΌ+π)πk β€ βmax
π,π
| (π πΌ+π)ππ,π| =βΒ· π
ββ1
= β (ββ1)!
π! (πβ (ββ1))!
. Since (βββ1)! β€ 2 forβ βNand
π! (πβ (ββ1))!
=πββ1(1+π(β/π)) β€2πββ1 forπ β , the claim follows.
Now, we apply Lemma 2.4.1 to (standard and mixed) transfer operators. It is convenient to state these bounds as follows. The first two statements are about the scaling of the norms of powers of πΈ; the last statement is about the magnitude of matrix elements in powers ofπΈ.
Lemma 2.4.2. Letπ(πΈ)denote the spectral radius of a matrixπΈ β B (Cπ·1βCπ·2). (i) Supposeπ(πΈ) β€ 1. Letββbe the size of the largest Jordan block(s) ofπΈ. Then
kπΈπk β€ 4πββ1 forπ β . (ii) Ifπ(πΈ) <1, then
kπΈπkπΉ β€ π(πΈ)π/2 forπ π·
1π·
2. We will often usekπΈπkπΉ β€1as a coarse bound.
(iii) Suppose that π(πΈ) =1. Let ββ denote the size of the largest Jordan block(s) in πΈ. For π, π β [π·
1π·
2], let (πΈπ)π,π denote the matrix element of πΈπ with respect to the standard computational basis {|πi}π·π=1π·2
1 . Then the following holds: for all π, π β [π·
1π·
2], there is a constant ππ,π with ππ,π = π(1) as πβ βand someβ β {1, . . . , ββ}such that
| (πΈπ)π,π|=ππββ1(1+π(πβ1)) .
Proof. Forπβspec(πΈ), let us denote byπ πΌβ(π)+πβ(π)the associated Jordan block, whereβ(π)is its size. Then
kπΈπk = max
πβspec(πΈ)
k (π πΌβ(π)+πβ(π))πk β€ max
πβspec(πΈ)
k (π πΌβ(π) +πβ(π))πkπΉ
where we assumed thatπ ββ β₯ β(π), |π| β€ 1 and (2.4.1). This shows claim (i).
Claim (ii) immediately follows from (2.4.1) and the observation thatππ·1π·
2β1π(πΈ)π = π(π(πΈ)π/2).
For the proof of statement (iii), observe that matrix elements (2.4.2) of a Jordan block (matrix) with eigenvalueπ(with|π|=1) of sizeβscale as
( (π πΌ+π)π)π,π
= 1 (πβ π)!
π! (πβ (πβπ))!
= 1 (πβ π)!
ππβπ(1+π(1/π)) forπ > πand are constant otherwise. Becauseπβ π β {1, . . . , ββ1}whenπ > π, this is of the form πβ(1+π(1/π)) for some β β [ββ 1]. Since πΈπ is similar (as a matrix) to a direct sum of such powers of Jordan blocks, and the form of this scaling does not change under linear combination of matrix coefficients, the claim follows.
Now let us consider the case where πΈ = |βiihhπ| βπΈΛ is the transfer operator of an injective MPS, normalized with maximum eigenvalue 1. Let π
2 < 1 denote the second largest eigenvalue. Without loss of generality, we can take the MPS to be in canonical form, so thatπΈ has a unique right fixed-point given by the identity matrix πΌ and a unique left fixed-point given by some positive-definite diagonal matrixΞ with unit trace. We can then write the Jordan decomposition of the transfer matrix as
πΈ =|πΌiihhΞ| βπΈ ,Λ (2.21)
where |πΌii and |Ξii denotes the vectorization of πΌ and Ξ, respectively, and where
Λ
πΈ denotes the remaining Jordan blocks of πΈ. Note that powers of πΈ can then be expressed as
πΈπ =|πΌiihhΞ| βπΈΛπ .
We can bound the Frobenius norm of the transfer matrix as
kπΈπk2πΉ =k |πΌiihhΞ| β πΈΛπk2πΉ =tr(πΌ)tr(Ξ2) + kπΈΛπk2πΉ β€ π·+ kπΈΛπk2πΉ, where tr(πΌ) =π·and tr(Ξ2) β€tr(Ξ)2=1. In particular, sinceπ(πΈΛ) =π
2, we obtain from Lemma 2.4.2(ii) that
kπΈΛπkπΉ β€ππ/2
2 forπ π· . (2.22)
This implies the following statement:
Lemma 2.4.3. The transfer operatorπΈ of an injective MPS satisfies kπΈπkπΉ β€
β
π·+1 forπ π· . (2.23)
We also need a bound on the norm kπΈβ
πΉ(π
1β π
2) k, where πΈ is a mixed transfer operator,πΉ β B ( (Cp)βπ) is an operator acting onπ sites, and whereππ βCπ·π for
π =1,2.
Lemma 2.4.4. LetπΈ
1= πΈ(π΄)andπΈ
2=πΈ(π΅) be the transfer operators associated with the tensors π΄ and π΅, respectively, with bond dimensions π·
1 and π·
2. Let πΈ = πΈ(π΄, π΅) β B (Cπ·1 β Cπ·2) be the combined transfer operator. Let π
1 β Cπ·1 andπ
2βCπ·2 be unit vectors. Then k (πΈπΉ)β (π
1βπ
2) k β€ kπΉk
βοΈ
kπΈπ
1k Β· kπΈπ
2k , (2.24)
kπΈπΉ(π
1βπ
2) k β€ kπΉk
βοΈ
kπΈπ
1k Β· kπΈπ
2k , (2.25)
kπΈπΉkπΉ β€ π·
1π·
2kπΉk
βοΈ
kπΈπ
1k Β· kπΈπ
2k , (2.26)
for allπΉ β B ( (Cp)βπ).
Proof. Writing matrix elements in the computational basis as πΉπ
1Β·Β·Β·ππ,π
1Β·Β·Β·ππ :=hπ
1Β· Β· Β·ππ|πΉ|π
1Β· Β· Β·ππi, we have
πΈπΉ = βοΈ
(π
1,...,ππ),(π
1,..., ππ)
πΉπ
1Β·Β·Β·ππ,π
1Β·Β·Β·ππ(π΄π
1 β π΅π
1) (π΄π
2 βπ΅π
2) Β· Β· Β· (π΄π
π β π΅π
π). Therefore,
(πΈπΉ)β = βοΈ
(π
1,...,ππ),(π
1,..., ππ)
πΉπ
1Β·Β·Β·ππ,π
1Β·Β·Β·ππ(π΄β
ππ
βπ΅β
ππ
) Β· Β· Β· (π΄β
π2
β π΅β
π2
) (π΄β π
1 βπ΅β
π1
)
= βοΈ
(π
1,...,ππ),(π
1,..., ππ)
(π πΉ πβ )ππΒ·Β·Β·π
1,ππΒ·Β·Β·π1(π΄β
ππ
β π΅β
ππ
) Β· Β· Β· (π΄β
π2
βπ΅β
π2
) (π΄β
π1
β π΅β
π1
) ,
whereπis the permutation which maps the π-th factor in the tensor product(Cp)βπ to the (πβ π +1)-th factor, and where πΉ is obtained by complex conjugating the matrix elements in the computational basis. This means that
(πΈπΉ)β = πΈβ
π πΉ πβ
, (2.27)
with πΈβ being the mixed transfer operator πΈβ = πΈ(π΄β , π΅β ) obtained by replacing eachπ΄π (respectivelyπ΅π) with its adjoint.
Now consider
k (πΈπΉ)β (π
1βπ
2) k2= (hπ
1| β hπ
2|)πΈπΉ(πΈπΉ)β (|π
1i β |π
2i)
= (hπ
1| β hπ
2|)πΈπΉπΈβ
π πΉ πβ
(|π
1i β |π
2i) . This can be represented diagrammatically as
k (πΈπΉ)β (π
1βπ
2) k2= .
In particular, we have k (πΈπΉ)β (π
1βπ
2) k2 = hπ| (πΉ βπΌβπ) (πΌβπ βπ πΉ πβ ) |πi, (2.28)
whereπ, πβ (Cp)β2π are defined as
|πi= ,
|πi= .
It is straightforward to check that kπk2= (hπ
1| β hπ
1|)πΈπ
1(πΈβ
1)π(|π
1i β |π
1i) , kπk2= (hπ
2| β hπ
2|)πΈπ
2(πΈβ
2)π(|π
2i β |π
2i) . Since k (πΈβ
π)πk = kπΈπ
πk for π = 1,2, it follows with the submultiplicativity of the operator norm that
kπk2 β€ kπΈπ
1k2, kπk2 β€ kπΈπ
2k2. (2.29)
Applying the Cauchy-Schwarz inequality to (2.4.2) yields k (πΈπΉ)β (π
1βπ
2) k2β€ k (πΉβ β πΌβπ)πk Β· k (πΌβπ βπ πΉ πβ )πk
β€ kπΉk2Β· kπk Β· kπk,
where we used the fact that the operator norm satisfies kπΉβ k = kπΉk = kπΉk and kπΌ β π΄k =kπ΄k. The claim (2.4.4) follows from this and (2.4.2).
The claim (2.4.4) follows analogously by using Equation (2.4.2). Finally, the claim (2.4.4) follows from (2.4.4) and
kπΈπΉk2 = βοΈ
πΌ1,πΌ
2β[π·
1]
βοΈ
π½1, π½
2β[π·
2]
| (hπΌ
1| β hπ½
1|)πΈπΉ(|πΌ
2i β |π½
2i) |2
β€ βοΈ
πΌ1,πΌ
2β[π·1]
βοΈ
π½1, π½
2β[π·
2]
kπΈπΉ(|πΌ
2i β |π½
2i) k2
β€ π·2
1π·2
2max
πΌ, π½
kπΈπΉ(|πΌ
2i β |π½
2i) k2, where we employed the orthonormal basis{|πΌi}πΌβ[π·
1] and{|π½i}π½β[π·
2] forCπ·1and Cπ·2, respectively, and applied the Cauchy-Schwarz inequality.
The main result of this section is the following upper bound on the matrix elements of geometricallyπ-local operators with respect to two MPS.
Theorem 2.4.5. Let|Ξ¨1i= |Ξ¨(π΄
1, π
1, π)i,|Ξ¨2i= |Ξ¨(π΄
2, π
2, π)i β (Cp)βπbe two MPS with bond dimensionsπ·
1andπ·
2, where
ππ =|ππihππ|, with kππk = kππk =1 for π =1,2, are rank-one operators. LetπΈ =πΈ(π΄
1, π΄
2) β B (Cπ·1 βCπ·2) denote the combined transfer operator defined by the MPS tensors π΄
1 and π΄
2, ββ
π the size of the largest Jordan block ofπΈπ =πΈ(π΄π) for π =1,2, andββthe size of the largest Jordan block of πΈ = πΈ(π΄
1, π΄
2). Assume that the spectral radii π(πΈ), π(πΈ
1), and π(πΈ
2) are contained in [0,1]. Then, for anyπΉ β B ( (Cp)βπ), we have
|hΞ¨1| (πΉβ πΌ(
Cp)βπβπ) |Ξ¨2i| β€ 16Β· kπΉk Β·π(β
β 1+ββ
2β2)/2(πβπ)βββ1 forπ π·
1, π·
2and (πβπ) π·
1π·
2. Proof. The matrix elementsπΌ= hΞ¨1| (πΉ βπΌ(
Cp)βπβπ) |Ξ¨2iof interest can be written as
πΌ= (hπ
1| β hπ
2|)πΈπΉπΈπβπ(|π
1i β |π
2i) . By the Cauchy-Schwarz inequality, we have
|πΌ| β€ kπΈβ
πΉ(|π
1i β |π
2i) k Β· kπΈπβπ(|π
1i β |π
2i k
β€ kπΉk
βοΈ
kπΈπ
1k Β· kπΈπ
2k Β· kπΈπβπk,
by the definition of the operator norm and Lemma 2.4.4. Then, the claim follows from Lemma 2.4.2 (i), which provides the bounds
kπΈπ
πk β€ 4πβ
β πβ1
for π =1,2, kπΈπβπk β€ 4(πβπ)βββ1
by our assumptions : π(πΈπ) β [0,1],π(πΈ) β [0,1], andπ π·π β₯ ββ
π for π =1,2, as well asπβπ π·
1π·
2 β₯ ββ.
2.5 No-Go Theorem: Degenerate Ground Spaces of Gapped Hamiltonians