Mouheymen Khamadja
@mouhaqh.bsky.social
Theoretical physics graduate from UC1 | Junior Researcher @CQTech
Looking for a PhD
Working on quantum algorithms for quantum simulation
Looking for a PhD
Working on quantum algorithms for quantum simulation
I hope that by the time those QPUs exist, we will have some better methods and ones that truly fuflill Feynman's prophecy. I will be working to make that happen and I hope to see some new approaches appear in the near future.
November 29, 2025 at 1:54 PM
I hope that by the time those QPUs exist, we will have some better methods and ones that truly fuflill Feynman's prophecy. I will be working to make that happen and I hope to see some new approaches appear in the near future.
I agree, but I don't expect a test with a large molecule like FeMoco will be doable, because measuring the Hamiltonian is still a huge problem, we can estimate the number of shots needed and we'll find that we need astronomically fast QPUs to run this computation in a lifetime.
November 29, 2025 at 1:51 PM
I agree, but I don't expect a test with a large molecule like FeMoco will be doable, because measuring the Hamiltonian is still a huge problem, we can estimate the number of shots needed and we'll find that we need astronomically fast QPUs to run this computation in a lifetime.
Unless one of these problems is solved with guarantees, VQAs will perish, or remain as an educational tool and an easy entry for beginners in the field.
i'm shifting away from VQAs too, but i am surprised at how much people are shifting towards classical ML for quantum systems because of this!
i'm shifting away from VQAs too, but i am surprised at how much people are shifting towards classical ML for quantum systems because of this!
November 29, 2025 at 12:37 PM
Unless one of these problems is solved with guarantees, VQAs will perish, or remain as an educational tool and an easy entry for beginners in the field.
i'm shifting away from VQAs too, but i am surprised at how much people are shifting towards classical ML for quantum systems because of this!
i'm shifting away from VQAs too, but i am surprised at how much people are shifting towards classical ML for quantum systems because of this!
There might be a tight room for some advantage from VQAs, but that would mean finding a system of sufficient size (>50 qubits) that has an efficiently measurable observable, with an ansatz that is hard to simulate classically but doesn't suffer BPs.
November 29, 2025 at 12:34 PM
There might be a tight room for some advantage from VQAs, but that would mean finding a system of sufficient size (>50 qubits) that has an efficiently measurable observable, with an ansatz that is hard to simulate classically but doesn't suffer BPs.
Adaptive VQAs are a bit of a pickle, they might avoid BPs because they do some sort of sequential warmstarts that keeps the initial states in [good patches](arxiv.org/pdf/2502.07889) , but they probably are classically simulable, check out the majorana propagation paper : arxiv.org/pdf/2503.18939
arxiv.org
November 29, 2025 at 12:28 PM
Adaptive VQAs are a bit of a pickle, they might avoid BPs because they do some sort of sequential warmstarts that keeps the initial states in [good patches](arxiv.org/pdf/2502.07889) , but they probably are classically simulable, check out the majorana propagation paper : arxiv.org/pdf/2503.18939
Getting the basics down first!
November 28, 2025 at 10:07 PM
Getting the basics down first!
To give credit where it's due, this is what @mvscerezo.bsky.social & @qzoeholmes.bsky.social have shown in their works and have been saying in their talks for a while now : "Working with VQAs you're either gonna fave Barren Plateaus or your circuits are classically simulable"
November 28, 2025 at 2:43 PM
To give credit where it's due, this is what @mvscerezo.bsky.social & @qzoeholmes.bsky.social have shown in their works and have been saying in their talks for a while now : "Working with VQAs you're either gonna fave Barren Plateaus or your circuits are classically simulable"
Those were my 13 cents😅, thank you for reading and would love to hear what everyone makes of this!
(14/14)
(14/14)
November 28, 2025 at 2:10 PM
Those were my 13 cents😅, thank you for reading and would love to hear what everyone makes of this!
(14/14)
(14/14)
A silly example:
Why not instead of using a QC to compute all the energy properties of 2 molecules and then use those values to determine if the molecules will bind or not, we instead simulate the molecules approaching and seeing if they bind or not, without looking under the hood.
(13)
Why not instead of using a QC to compute all the energy properties of 2 molecules and then use those values to determine if the molecules will bind or not, we instead simulate the molecules approaching and seeing if they bind or not, without looking under the hood.
(13)
November 28, 2025 at 2:09 PM
A silly example:
Why not instead of using a QC to compute all the energy properties of 2 molecules and then use those values to determine if the molecules will bind or not, we instead simulate the molecules approaching and seeing if they bind or not, without looking under the hood.
(13)
Why not instead of using a QC to compute all the energy properties of 2 molecules and then use those values to determine if the molecules will bind or not, we instead simulate the molecules approaching and seeing if they bind or not, without looking under the hood.
(13)
The recent work by google (quantum echoes algorithm) is one simple example of what Feynman dreamed of, simulating QS with QS. We can't have a one-size-fits-all algorithm, rather we will have a good method that needs a lot of tailoring to be applied to each single problem to get advantage.
(12)
(12)
November 28, 2025 at 2:07 PM
The recent work by google (quantum echoes algorithm) is one simple example of what Feynman dreamed of, simulating QS with QS. We can't have a one-size-fits-all algorithm, rather we will have a good method that needs a lot of tailoring to be applied to each single problem to get advantage.
(12)
(12)
Most of the approaches to Quantum simulation with QC are somehow quantizations of a classical algorithm. We need to rethink this and start using QC as experiments in a lab, designing new ways to use them rather than linear algebra accelerators in a classical algorithm.
(11)
(11)
November 28, 2025 at 2:03 PM
Most of the approaches to Quantum simulation with QC are somehow quantizations of a classical algorithm. We need to rethink this and start using QC as experiments in a lab, designing new ways to use them rather than linear algebra accelerators in a classical algorithm.
(11)
(11)
To quote Feynman here: "I think Nature's imagination is so much greater than man's, she's never gonna let us relax!"
However, In my opinion, I think this is great news, this means we haven't yet truly fathomed what we can do with quantum computers.
(10)
However, In my opinion, I think this is great news, this means we haven't yet truly fathomed what we can do with quantum computers.
(10)
November 28, 2025 at 2:01 PM
To quote Feynman here: "I think Nature's imagination is so much greater than man's, she's never gonna let us relax!"
However, In my opinion, I think this is great news, this means we haven't yet truly fathomed what we can do with quantum computers.
(10)
However, In my opinion, I think this is great news, this means we haven't yet truly fathomed what we can do with quantum computers.
(10)
So to summarize, VQE was never meant to bring quantum advantage, 2 million equations later, we have definitive proof of that 😅.
It's a bit overoptimistic to think that we could solve all nature problems with one ansatz and a good optimizer.
(9)
It's a bit overoptimistic to think that we could solve all nature problems with one ansatz and a good optimizer.
(9)
November 28, 2025 at 1:59 PM
So to summarize, VQE was never meant to bring quantum advantage, 2 million equations later, we have definitive proof of that 😅.
It's a bit overoptimistic to think that we could solve all nature problems with one ansatz and a good optimizer.
(9)
It's a bit overoptimistic to think that we could solve all nature problems with one ansatz and a good optimizer.
(9)
This is where one expects the QPU to bring advantage, but here we face another problem: The Barren Plateaus, Circuits that are too complex = Space is too complex to be searched classically, so the bottleneck of classically simulating quantum systems was just reformulated, not solved.
(8)
(8)
November 28, 2025 at 1:52 PM
This is where one expects the QPU to bring advantage, but here we face another problem: The Barren Plateaus, Circuits that are too complex = Space is too complex to be searched classically, so the bottleneck of classically simulating quantum systems was just reformulated, not solved.
(8)
(8)
Now of course, PP and MP will fail if the circuit gates induce a lot of "branching" with the observable, and the computational cost to retain accuracy will become too much to handle. But, this is the case for highly complex circuits, or unstructured ones that are very deep (HEA with many reps)
(7)
(7)
November 28, 2025 at 1:48 PM
Now of course, PP and MP will fail if the circuit gates induce a lot of "branching" with the observable, and the computational cost to retain accuracy will become too much to handle. But, this is the case for highly complex circuits, or unstructured ones that are very deep (HEA with many reps)
(7)
(7)
So the role of the QPU in VQE was just to efficiently do some linear algebra for us and the advantage is just a slight computational speedup that is weakened by the cost of measuring the observable in the end (chemistry observables are costly, we'll come back to simple ones a bit later)
(6)
(6)
November 28, 2025 at 1:46 PM
So the role of the QPU in VQE was just to efficiently do some linear algebra for us and the advantage is just a slight computational speedup that is weakened by the cost of measuring the observable in the end (chemistry observables are costly, we'll come back to simple ones a bit later)
(6)
(6)
Once assigning the parameters to their contribution to the cost function, one can still search classically through that space for the global minimum, the classical optimizer doesn't distinguish a cost function that is estimated on the QPU from one that is obtained through MP or PP
(5)
(5)
November 28, 2025 at 1:42 PM
Once assigning the parameters to their contribution to the cost function, one can still search classically through that space for the global minimum, the classical optimizer doesn't distinguish a cost function that is estimated on the QPU from one that is obtained through MP or PP
(5)
(5)
The Measurement was not accelerated as if given a statevector it can be estimated with similar complexity classically.
Evidence to this is the Majorana Propagation (MP) and Pauli Propagation (PP) algorithms, where the costly part was back-propagating the observable.
(4)
Evidence to this is the Majorana Propagation (MP) and Pauli Propagation (PP) algorithms, where the costly part was back-propagating the observable.
(4)
November 28, 2025 at 1:38 PM
The Measurement was not accelerated as if given a statevector it can be estimated with similar complexity classically.
Evidence to this is the Majorana Propagation (MP) and Pauli Propagation (PP) algorithms, where the costly part was back-propagating the observable.
(4)
Evidence to this is the Majorana Propagation (MP) and Pauli Propagation (PP) algorithms, where the costly part was back-propagating the observable.
(4)
The way VQE was designed, the QPU had one single task (simplifying a bit here), which was to accelerate matrix products that correspond to applying the parametrized gates to the initial state. This is not including maesuring the observable at the end of the circuit.
(3)
(3)
November 28, 2025 at 1:33 PM
The way VQE was designed, the QPU had one single task (simplifying a bit here), which was to accelerate matrix products that correspond to applying the parametrized gates to the initial state. This is not including maesuring the observable at the end of the circuit.
(3)
(3)
Just to set a scope, this thread will be about VQAs for ground/excited state preparation, VQE and its "non-fault tolerant" variants to be specific, time-evolution and QAOA will be set aside for later.
VQE was never meant to save QC or bring Quantum advantage, and here is why:
(2)
VQE was never meant to save QC or bring Quantum advantage, and here is why:
(2)
November 28, 2025 at 1:29 PM
Just to set a scope, this thread will be about VQAs for ground/excited state preparation, VQE and its "non-fault tolerant" variants to be specific, time-evolution and QAOA will be set aside for later.
VQE was never meant to save QC or bring Quantum advantage, and here is why:
(2)
VQE was never meant to save QC or bring Quantum advantage, and here is why:
(2)
haha challenge noted, though I don’t think I’d dare take it on. This notation already deserves a medal for consistency 😄
Also, not to derail this thread, but I’d love to ask your opinion about something related to quantum simulation at some point, if that’s okay!
2/2
Also, not to derail this thread, but I’d love to ask your opinion about something related to quantum simulation at some point, if that’s okay!
2/2
November 23, 2025 at 12:19 AM
haha challenge noted, though I don’t think I’d dare take it on. This notation already deserves a medal for consistency 😄
Also, not to derail this thread, but I’d love to ask your opinion about something related to quantum simulation at some point, if that’s okay!
2/2
Also, not to derail this thread, but I’d love to ask your opinion about something related to quantum simulation at some point, if that’s okay!
2/2
By “best randomness” I meant something like : is the depolarizing channel the most “uniform” in dissipating information in all directions? I was struggling to articulate it, then reread the paper and found I’m clearly not alone in that struggle 😅
1/2
1/2
November 23, 2025 at 12:16 AM
By “best randomness” I meant something like : is the depolarizing channel the most “uniform” in dissipating information in all directions? I was struggling to articulate it, then reread the paper and found I’m clearly not alone in that struggle 😅
1/2
1/2