QUANTUM ERROR-DETECTION AT LOW ENERGIES
2.5 No-Go Theorem: Degenerate Ground Spaces of Gapped Hamiltonians are Constant-Distance AQEDCare Constant-Distance AQEDC
The main result of this section is the following upper bound on the matrix elements of geometrically๐-local operators with respect to two MPS.
Theorem 2.4.5. Let|ฮจ1i= |ฮจ(๐ด
1, ๐
1, ๐)i,|ฮจ2i= |ฮจ(๐ด
2, ๐
2, ๐)i โ (Cp)โ๐be two MPS with bond dimensions๐ท
1and๐ท
2, where
๐๐ =|๐๐ih๐๐|, with k๐๐k = k๐๐k =1 for ๐ =1,2, are rank-one operators. Let๐ธ =๐ธ(๐ด
1, ๐ด
2) โ B (C๐ท1 โC๐ท2) denote the combined transfer operator defined by the MPS tensors ๐ด
1 and ๐ด
2, โโ
๐ the size of the largest Jordan block of๐ธ๐ =๐ธ(๐ด๐) for ๐ =1,2, andโโthe size of the largest Jordan block of ๐ธ = ๐ธ(๐ด
1, ๐ด
2). Assume that the spectral radii ๐(๐ธ), ๐(๐ธ
1), and ๐(๐ธ
2) are contained in [0,1]. Then, for any๐น โ B ( (Cp)โ๐), we have
|hฮจ1| (๐นโ ๐ผ(
Cp)โ๐โ๐) |ฮจ2i| โค 16ยท k๐นk ยท๐(โ
โ 1+โโ
2โ2)/2(๐โ๐)โโโ1 for๐ ๐ท
1, ๐ท
2and (๐โ๐) ๐ท
1๐ท
2. Proof. The matrix elements๐ผ= hฮจ1| (๐น โ๐ผ(
Cp)โ๐โ๐) |ฮจ2iof interest can be written as
๐ผ= (h๐
1| โ h๐
2|)๐ธ๐น๐ธ๐โ๐(|๐
1i โ |๐
2i) . By the Cauchy-Schwarz inequality, we have
|๐ผ| โค k๐ธโ
๐น(|๐
1i โ |๐
2i) k ยท k๐ธ๐โ๐(|๐
1i โ |๐
2i k
โค k๐นk
โ๏ธ
k๐ธ๐
1k ยท k๐ธ๐
2k ยท k๐ธ๐โ๐k,
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
k๐ธ๐
๐k โค 4๐โ
โ ๐โ1
for ๐ =1,2, k๐ธ๐โ๐k โค 4(๐โ๐)โโโ1
by our assumptions : ๐(๐ธ๐) โ [0,1],๐(๐ธ) โ [0,1], and๐ ๐ท๐ โฅ โโ
๐ for ๐ =1,2, as well as๐โ๐ ๐ท
1๐ท
2 โฅ โโ.
2.5 No-Go Theorem: Degenerate Ground Spaces of Gapped Hamiltonians
than constant. We prove this result by employing the necessary condition for approximate error-detection from Lemma 2.3.6 for the code subspaces generated by varying the boundary conditions of an (open-boundary) injective MPS. Note that, given a translation invariant MPS with periodic boundary conditions and bond dimension๐ท, there exists a local gapped Hamiltonian, called the parent Hamiltonian, with a unique ground state being the MPS [28].
We need the following bounds which follow from the orthogonality and normaliza- tion of states in such codes.
Lemma 2.5.1. Let๐ดbe the MPS tensor of an injective MPS with bond dimension๐ท, and let ๐ , ๐ โ B (C๐ท) be such that the states |ฮจ๐i = |ฮจ(๐ด, ๐ , ๐)i and |ฮจ๐i =
|ฮจ(๐ด, ๐ , ๐)i are normalized and orthogonal. Let us write the transfer operator as ๐ธ =|๐ผiihhฮ| โ๐ธห (cf. Equation(2.4.2)). Assume๐ ๐ท. Then
(i) The Frobenius norm of๐ (and similarly the norm of๐) is bounded by k๐k๐น =๐(1) .
(ii) We have
|hhฮ| (๐ โ๐) |๐ผii| =๐(๐
๐/2 2 ),
|hhฮ| (๐ โ ๐) |๐ผii| =๐(๐
๐/2 2 ). In the following proofs, we repeatedly use the inequality
|tr(๐
1. . . ๐๐) | โค k๐
1k๐น ยท k๐
2k๐นยท ยท ยท k๐๐k๐น (2.30) for ๐ทร ๐ท-matrices {๐๐}๐
๐=1. Note that the inequality (2.5) is simply the Cauchy- Schwarz inequality for๐ =2. For ๐ >2, the inequality follows from the inequality for๐ =2 and the submultiplicativity of the Frobenius-norm because
|tr(๐
1. . . ๐๐) | โค k๐
1k๐น ยท k๐
2ยท ยท ยท๐๐k๐น โค k๐
1k๐น ยท k๐
2k๐นยท ยท ยท k๐๐k๐น . Proof. The proof of (i) follows from the fact that the stateฮจ๐ is normalized, i.e.,
1=kฮจ๐k2
=tr
๐ธ๐(๐ โ ๐)
=tr
|๐ผiihhฮ| (๐ โ ๐)
+tr(๐ธห๐(๐ โ ๐))
=tr(ฮ๐ ๐โ ) +tr(๐ธห๐(๐ โ ๐))
โฅ ๐
min(ฮ) ยท k๐k2๐น+tr(๐ธห๐(๐ โ ๐)),
where๐
min(ฮ) denotes the smallest eigenvalue of ฮ, and we make use of the fact that๐ ๐โ is positive with trace tr(๐ ๐โ ) = k๐k2
๐น. Since
|tr(๐ธห๐(๐ โ ๐)) | โค k๐ธห๐k๐น ยท k๐ โ ๐k๐น โค ๐๐/2
2 k๐k2๐น for๐ ๐ท by (2.5) and (2.4.2), we conclude
k๐k2๐น โค
๐min(ฮ) โ๐๐/2
2
โ1
=๐
min(ฮ)โ1(1+๐(๐๐/2
2 )). Then the claim (i) follows since๐
min(ฮ)โ1is a constant.
Now, consider the first inequality in (ii) (the bound for |hhฮ| (๐ โ ๐) |๐ผii| is shown analogously). Using the orthogonality of the states|ฮจ๐iand|ฮจ๐i, we obtain
0=hฮจ๐|ฮจ๐i=tr
๐ธ๐(๐ โ๐)
=tr
(|๐ผiihhฮ| +๐ธห๐) (๐ โ๐)
=hhฮ| (๐ โ๐) |๐ผii +tr(๐ธห๐(๐ โ๐)), hence
|hhฮ| (๐ โ๐) |๐ผii| =|tr(๐ธห๐(๐ โ๐)) | โค k๐ธหk๐น ยท k๐ โ๐k๐น
โค ๐
๐/2
2 k๐k๐นยท k๐k๐น , using (2.4.2). The claim (ii) then follows from (i).
With the following lemma, we prove an upper bound on the overlap of the reduced density matrices๐๐and๐๐, supported on 2ฮ-sites surrounding the boundary, of the global states|ฮจ๐iand|ฮจ๐i, respectively.
Lemma 2.5.2. Let๐ดbe an MPS tensor of an injective MPS with bond dimension๐ท, and let ๐ , ๐ โ B ( (Cp)โ๐) be such that the states |ฮจ๐i = |ฮจ(๐ด, ๐ , ๐)i and |ฮจ๐i =
|ฮจ(๐ด, ๐ , ๐)i are normalized and orthogonal. Letฮ ๐ท. Let S ={1,2, . . . ,ฮ} โช {๐โฮ+1, ๐โฮ+2, . . . , ๐} be the subset of2ฮspins consisting of ฮ systems at the left and andฮsystems at the right boundary. Let ๐๐ = tr[๐]\S|ฮจ๐ihฮจ๐| and ๐๐ =tr[๐]\S|ฮจ๐ihฮจ๐|be the reduced density operators on these subsystems. Then
tr(๐๐๐๐) โค ๐๐
ฮ 2
2
where ๐
2 is the second largest eigenvalue of the transfer operator ๐ธ = ๐ธ(๐ด) and where๐is a constant depending only on the minimal eigenvalue of๐ธ and the bond dimension๐ท.
Figure 2.4: The two expressions in Equation (2.5), where ๐ฟ, ๐, and ๐ are used to denote the sites defined in (2.5).
Proof. For convenience, let us relabel the systems as (๐ฟ
1, . . . , ๐ฟฮ) = (1,2, . . . ,ฮ) (๐
1, . . . , ๐๐โ
2ฮ) = (ฮ+1,ฮ+2, . . . , ๐โฮ) (๐
1, . . . , ๐ ฮ) = (๐โฮ+1, ๐โฮ+2, . . . , ๐)
(2.31) indicating their location on the left, in the middle, and on the right, respectively.
For the tensor productH๐ด โ H๐ต of two isomorphic Hilbert spaces, we denote by F๐ด ๐ต โ B (H๐ดโ H๐ต) the flip-operator which swaps the two systems. The following expressions are visualized in Figure 2.4. We have
tr(๐๐๐๐) =tr( (๐๐ฟ1ยทยทยท๐ฟฮ๐ 1ยทยทยท๐ ฮ
๐ โ ๐
๐ฟ0
1ยทยทยท๐ฟ0
ฮ๐ 0
1ยทยทยท๐ 0
ฮ
๐ ) (F๐ฟ ๐ฟ0 โF๐ ๐ 0)), where F๐ฟ ๐ฟ0 =F๐ฟ1๐ฟ0
1
โF๐ฟ2๐ฟ0
2
โ ยท ยท ยท โF๐ฟฮ๐ฟ0
ฮ
, F๐ ๐ 0 =F๐ 1๐ 0
1
โF๐ 2๐ 0
2
โ ยท ยท ยท โF๐ ฮ๐ 0
ฮ
. DefiningF๐ ๐0 analogously, ๐ผ๐ ๐0 = ๐ผ๐
1๐0
1
โ ยท ยท ยท โ ๐ผ๐
๐โ2ฮ๐0
๐โ2ฮ, and similarly ๐ผ๐ฟ ๐ฟ0
and๐ผ๐ ๐ 0, this can be rewritten (by the definition of the partial trace) as tr(๐๐๐๐)
=(hฮจ๐ฟ ๐ ๐ ๐ | โ hฮจ๐๐ฟ0๐0๐ 0|) (F๐ฟ ๐ฟ0 โ ๐ผ๐ ๐0 โF๐ ๐ 0) (|ฮจ๐ฟ ๐ ๐ ๐ i โ |ฮจ๐๐ฟ0๐0๐ 0i)
=(hฮจ๐ฟ ๐ ๐ ๐ | โ hฮจ๐๐ฟ0๐0๐ 0|) (๐ผ๐ฟ ๐ฟ0 โF๐ ๐0 โ ๐ผ๐ ๐ 0) (|ฮจ๐๐ฟ ๐ ๐ i โ |ฮจ๐ฟ๐0๐0๐ 0i).(2.32)
In the last identity, we have used thatF2 =๐ผ is the identity.
Reordering and regrouping the systems as (๐ฟ
1๐ฟ0
1) (๐ฟ
2๐ฟ0
2) ยท ยท ยท (๐ฟฮ๐ฟ0
ฮ) (๐
1๐0
1) (๐
2๐0
2) ยท ยท ยท (๐๐โ
2ฮ๐0
๐โ2ฮ) (๐
1๐ 0
1) (๐
2๐ 0
2) ยท ยท ยท (๐ ฮ๐ 0
ฮ),
we observe that|ฮจ๐ฟ ๐ ๐
๐ i โ |ฮจ๐ฟ0๐0๐ 0
๐ iis an MPS with MPS tensor๐ดโ๐ดand boundary tensor๐โ๐and|ฮจ๐ฟ ๐ ๐
๐ i โ |ฮจ๐ฟ0๐0๐ 0
๐ iis an MPS with MPS tensor๐ดโ๐ดand boundary tensor๐ โ๐. Let us denote the virtual systems of the first MPS by๐
1๐
2, and those of the second MPS by ๐
1๐
2, such that the boundary tensors are ๐๐1 โ๐๐2 and ๐๐1โ๐๐2, respectively. Let ห๐ธ = ๐ธ๐1๐
1โ๐ธ๐2๐
2be the associated transfer operator.
Then we have from (2.5) tr(๐๐๐๐) =tr
ห ๐ธฮ๐ธห
Fโ๐โ2ฮ๐ธหฮ h
(๐
๐1 โ๐
๐2) โ (๐๐1 โ ๐๐2)i
. (2.33) Recall that๐ธฮ = |๐ผiihhฮ| โ๐ธหฮ, where we have
k๐ธหฮk๐น โคโ๏ธ
๐ท2ยท k๐ธหฮk โค ๐ทยท๐ฮ/2
2 forฮ ๐ท ,
k |๐ผiihhฮ| k๐น =k |๐ผii k2ยท k |ฮii k2 โค ๐ท2.
In the second line, we use the fact that k |ฮii k2 = tr(ฮโ ฮ) = ร
๐๐๐2 โค 1 and k |๐ผii k2 =๐ท2. Therefore, we have
๐ธฮ = โ๏ธ
๐โ{0,1}
๐ป๐, where๐ป
0= |๐ผiihhฮ|and๐ป
1=๐ธหฮsatisfy k๐ป
0k๐น โค ๐ท2, and k๐ป
1k๐น โค ๐ทยท๐ฮ/2
2 forฮ ๐ท . (2.34) Note that
ห
๐ธฮ =๐ธฮ โ๐ธฮ = โ๏ธ
๐1,๐
2โ{0,1}
๐ป๐
1 โ๐ป๐
2 . Inserting this into (2.5) gives a sum of 16 terms
tr(๐๐๐๐) โค โ๏ธ
๐1,๐
2,๐
3,๐
4โ{0,1}
|๐ผ๐
1,๐
2,๐
3,๐
4|, where
๐ผ๐
1,๐
2,๐
3,๐
4 =tr
(๐ป๐1๐1
๐1
โ ๐ป๐2๐2
๐2
)๐ธห
Fโ๐โ2ฮ(๐ป๐1๐1
๐3
โ ๐ป๐2๐2
๐4
) h (๐
๐1โ๐
๐2) โ (๐๐1 โ ๐๐2)i .
Consider the term with๐๐ =0 for all ๐ โ {1, . . . ,4}. This is given by ๐ผ0,0,0,0=tr
(|๐ผiihhฮ|๐1๐1 โ |๐ผiihhฮ|๐2๐2)๐ธห
Fโ๐โ2ฮ(|๐ผiihhฮ|๐1๐1 โ |๐ผiihhฮ|๐2๐2)
ยท h (๐
๐1 โ๐
๐2) โ (๐๐1 โ ๐๐2)i
=hhฮ| (๐ โ๐) |๐ผiihhฮ| (๐ยฏ โ ๐) |๐ผii (hhฮ| โ hhฮ|)๐ธห
Fโ๐โ2ฮ(|๐ผii โ |๐ผii).(2.35) By inserting this into (2.5), we get with Lemma 2.5.1 (ii) and the Cauchy-Schwarz inequality
|๐ผ
0,0,0,0|=๐(๐๐
2) ยท
hhฮ| โ hhฮ|)๐ธห
Fโ๐โ2ฮ(|๐ผii โ |๐ผii
=๐(๐๐
2) ยท k |ฮii โ |ฮii k ยท k๐ธห
Fโ๐โ2ฮ(|๐ผii โ |๐ผii k . With Lemma 2.4.4, this can further be bounded as
|๐ผ
0,0,0,0| =๐(๐๐
2) ยท k |ฮii k2ยท k |๐ผii k2ยท kFโ๐โ2ฮk ยท k๐ธ๐โ2ฮk .
SincekFk =1 andk |ฮii k =๐(1), k |๐ผii k =๐(1) andk๐ธ๐โ2ฮk =๐(1)(cf. (2.4.3)), we conclude that
|๐ผ
0,0,0,0|=๐(๐๐
2) . (2.36)
The remaining terms|๐ผ๐
1,๐
2,๐
3,๐
4|with(๐
1, ๐
2, ๐
3, ๐
4) โ (0,0,0,0)can be bounded as follows: using inequality (2.5), we have
|๐ผ๐
1,๐
2,๐
3,๐
4|= tr
(๐ป๐
1 โ ๐ป๐
2)๐ธห
Fโ๐โ2ฮ(๐ป๐
3 โ๐ป๐
4)h
(๐ โ๐) โ (๐ โ ๐)i
โค k๐ป๐
1 โ ๐ป๐
2k๐น ยท k๐ธ
Fโ๐โ2๐k๐น ยท k๐ป๐
3 โ ๐ป๐
4k๐น ยท k๐ โ๐ โ๐ โ ๐k๐น
=k๐k2๐น ยท k๐k2๐น ยทยฉ
ยญ
ยซ ร4
๐=1
k๐ป๐
๐k๐นยช
ยฎ
ยฌ
ยท k๐ธ
Fโ๐โ2ฮk๐น
=๐(๐ฮ/2
2 ) ยท k๐k2๐นยท k๐k2๐น ยท k๐ธ
Fโ๐โ2ฮk๐น , where we use (2.5) and the assumption that (๐
1, ๐
2, ๐
3, ๐
4) โ (0,0,0,0). We use Lemma 2.4.4 and (2.4.3) to get the upper bound k๐ธ
Fโ๐โ2ฮk โค ๐ท2k๐นโ๐โ2ฮk ยท k๐ธ๐โ๐ทk =๐(1). Thus
|๐ผ๐
1,๐
2,๐
3,๐
4| =๐(๐ฮ/2
2 ) for (๐
1, ๐
2, ๐
3, ๐
4) โ (0,0,0,0) . (2.37) Combining (2.5) with (2.5), we conclude that
|tr(๐๐๐๐) | โค โ๏ธ
๐1,๐
2,๐
3,๐
4โ{0,1}
|๐ผ๐
1,๐
2,๐
3,๐
4| โค |๐ผ
0,0,0,0| +15 max
(๐1,๐
2,๐
3,๐
4)โ (0,0,0,0)
|๐ผ๐
1,๐
2,๐
3,๐
4|
=๐(๐ฮ/2
2 ) .
The claim follows from this.
Recall that we call (a family of subspaces) C โ (Cp)โ๐ an approximate error- detection code if it is an (๐ , ๐ฟ) [ [๐, ๐ , ๐]]-code with๐ โ 0 and๐ฟ โ 0 for๐ โ โ. Our main result is the following:
Theorem 2.5.3. LetC โ (Cp)โ๐be an approximate error-detecting code generated from a translation-invariant injective MPS of constant bond dimension๐ทby varying boundary conditions. Then the distance ofCis constant.
Proof. LetC =C๐ โ (Cp)โ๐ be a (family of) subspace(s) of dimensionp๐ defined by an MPS tensor ๐ดby choosing different boundary conditions, i.e.,
C๐={|ฮจ(๐ด, ๐ , ๐)i | ๐ โ X} โ (Cp)โ๐
for some (fixed) subspaceX โ B (C๐ท). For the sake of contradiction, assume that C๐is an (๐๐, ๐ฟ๐) [ [๐, ๐ , ๐๐]]-code with
๐๐, ๐ฟ๐ โ0 and code distance๐๐โ โ for๐โ โ. (2.38) Let |ฮจ๐i = |ฮจ(๐ด, ๐ , ๐)i,|ฮจ๐i = |ฮจ(๐ด, ๐ , ๐)i โ C be two orthonormal states defined by choosing different boundary conditions ๐ , ๐ โ X. From Lemma 2.5.2, we may chooseฮsufficiently large such that the reduced density operators ๐๐, ๐๐ on๐sites surrounding the boundary satisfies
tr(๐๐๐๐) โค ๐๐๐/4
2 for all ๐ โฅ2ฮ. (2.39)
We note thatฮonly depends on the transfer operator and is independent of๐. Fix any constant๐ , ๐ฟ โ (0,1)and choose some๐ โฅ 2ฮsufficiently large such that
๐(๐๐, ๐๐) :=๐ ๐ท2๐๐/4
2
satisfies
๐ < 1โ10๐ and ๐ฟ < (1โ๐)2. (2.40) Since by assumption๐๐ โ โ, there exists some๐
0 โNsuch that ๐๐ > ๐ for all๐ โฅ ๐
0. (2.41)
Combining (2.5), (2.5), and (2.5) with Lemma 2.3.6, we conclude thatC๐is not an (๐ , ๐ฟ) [ [๐, ๐ , ๐๐]]-code for any๐ โฅ ๐
0.
By assumption (2.5), there exists some๐
1 โNsuch that ๐๐ < ๐ and ๐ฟ๐ < ๐ฟ for all๐ โฅ ๐
1. Let us set๐ =max{๐
0, ๐
1}. Then we obtain thatC๐is not an (๐๐, ๐ฟ๐) [ [๐, ๐ , ๐๐]]- code for any๐ โฅ ๐, a contradiction.
In terms of the TQO-1 condition (cf. [57]), Theorem 2.5.3 shows the absence of topological order in 1D gapped systems. The theorem also tells us that we should not restrict our attention to the ground space of a local Hamiltonian when looking for quantum error-detecting codes.5 In the following sections, we bypass this no-go result by extending our search for codes to low-energy states. In particular, we show that single quasi-particle momentum eigenstates of local gapped Hamiltonians and multi-particle excitations of the gapless Heisenberg model constitute error-detecting codes. See Sections 2.6 and 2.7, respectively.