We Tortured Salt Water With X-Rays

It Remembered.

I spent 700 minutes watching salt water forget how to be liquid.

Not metaphorically. Not “sort of.” I measured it—every 30 seconds, with X-rays that can count the distance between atoms. And somewhere around minute 350, the rules inverted.

Crystals are supposed to be the organized ones. Rigid. Constrained. When atoms lock into that perfect lattice, their bonds should tighten up—everything pulled into place by the crystal’s geometry.

Liquids are chaos. Molecules tumbling freely, bonds relaxed, no memory that survives from one moment to the next. That’s what every chemistry textbook teaches. That’s what thermodynamics says should happen.

I watched the opposite.

The bonds in my dissolved salt water were shorter than the bonds in the crystals. The liquid was more constrained than the solid. And after I cycled it through crystallization and dissolution twenty times, the liquid had developed structure—persistent organization that shouldn’t exist.

The solution remembered being a crystal.

The Setup: Acoustic Levitation and Atomic-Resolution X-rays

Picture a droplet of Epsom salt solution—magnesium sulfate, the same mineral detected all over Mars—floating in mid-air. Not magic. Physics. Two speakers aimed at each other create a standing wave, and the droplet sits trapped in the pressure nodes, suspended by nothing but sound.

Sound waves create pressure nodes that trap a droplet in mid-air while synchrotron X-rays measure atomic-scale distances in real-time. The acoustic levitator exposes the entire droplet surface to air, maximizing evaporation rate while maintaining precise positioning for X-ray diffraction measurements.

Now aim the Advanced Photon Source at it—one of the most intense X-ray beams on the planet. Every 30 seconds, I’m collecting a pair distribution function (PDF), which is a fancy way of saying: I know exactly how far apart every atom is from every other atom. Sulfur to oxygen. Magnesium to oxygen. Water to water.

The droplet evaporates. Crystals form. I watch the atomic distances shift in real-time as liquid becomes solid.

Then—and this is the key part—I add another drop directly onto those crystals.

They dissolve. Everything goes back to liquid.

Then it crystallizes again.

I did this twenty times.

That’s it. That’s the whole experiment.

No exotic chemicals. No extreme pressures or temperatures. Just: evaporate, add water, repeat. The kind of thing a kid could do with a salt shaker and a puddle on the kitchen counter, except I had X-rays that could count atoms.

Here’s what I can’t get over: nobody had done this before.

Not with atomic-resolution measurements. Not systematically. Not asking “what happens to the solution after the crystals dissolve?”

We’ve been studying crystallization for over a century. Thousands of papers on magnesium sulfate alone. We know the phase diagrams, the humidity conditions, the transition temperatures. We can predict which crystal will form under what conditions.

But we’d always treated each crystallization event as independent. Evaporate once, measure the result, done. Start fresh next time.

Nobody had asked: “Does the solution remember?”

Why not? Maybe because it seems too simple. Too obvious. Surely someone would have checked already.

Or maybe because we’re trained to think about equilibrium—the final state—rather than history. Thermodynamics tells you where a system will end up, not how its past shapes that journey.

Or maybe because the experiments that get published are the ones with clear hypotheses and predicted outcomes. “I’m going to cycle this twenty times and see what happens” doesn’t sound like serious science. It sounds like... playing around.

But playing around is how you find things nobody was looking for.

It’s one of those questions that seems obvious in retrospect—of course you should check if cycling matters, systems develop history all the time—but it required breaking a conceptual assumption we didn’t even know we were making.

The assumption: dissolved = reset. Once it’s liquid again, the system has no memory. You’re back to a blank slate.

That assumption is baked into how we teach chemistry. Into how we design experiments. Into the entire framework of classical nucleation theory.

I broke that assumption by accident, really. The acoustic levitator could only hold so much weight, so I had to keep adding small drops instead of starting with one big concentrated solution. I was trying to work around an equipment limitation.

And that limitation forced me to watch the same material transform over and over in the same location, rather than treating each crystallization as a separate event.

Sometimes the most important experimental designs come from asking: “What if I just... kept going?”

Twenty cycles. Not because I had a hypothesis that predicted something interesting would happen after twenty. Just because I wanted to see. Because the setup let me keep adding drops, so... why not? Plus, after the last cycle the droplet fell out of the levitator and the experiment was over.

The inversion point showed up around cycle 15. If I’d stopped at ten—if I’d followed the standard protocol of “measure it once and move on”—I would have missed it entirely.

The difference between measuring crystallization once versus watching what happens when you keep cycling. Standard protocols treat each crystallization as an independent event. By continuously cycling the same material through dissolution and recrystallization, structural reorganization becomes visible—but only after ~15 cycles. Single-measurement approaches would miss this entirely.

Visual sequence of a single dissolution-recrystallization cycle. A droplet evaporates over ~30 minutes, crystals form, then a fresh drop is added directly onto the crystalline mass. Complete dissolution takes several minutes, after which the cycle repeats. By cycle 20, dissolution takes noticeably longer—the first observable hint that the system is changing. This simple visual observation foreshadowed the molecular-scale reorganization revealed by atomic measurements.

The Backwards Bonds

Classical crystallization theory predicts this: atoms in solution should have loose, flexible bonds. When they lock into a rigid crystal lattice, those bonds should become constrained—tighter, shorter, more uniform.

The opposite happened.

The sulfur-oxygen bonds in the sulfate groups (SO₄²⁻) were shorter in the liquid than in the crystal.

- Liquid phase: 1.43-1.44 Ångströms

- Crystal phase: 1.47 Ångströms

When these molecules crystallized, the bonds stretched.

This was surprising but explainable. Crystal packing forces can distort molecular geometry. The sulfate groups have to fit into the crystal’s architecture—its perfectly arranged hydrogen bonding networks, its magnesium ions each surrounded by exactly six water molecules in octahedral coordination. The crystal says: “I don’t care what bond length you prefer. You fit into my architecture now.”

But then something stranger started happening—and it changed everything.

The Inversion Point

Around 350 minutes into the experiment—somewhere between drops 15 and 18—the behavior changed.

In early cycles:

- Add fresh drop → crystals dissolve quickly

- Bond lengths snap back to their short, liquid-phase values (1.43-1.44 Å)

- Clear distinction between dissolved and crystalline states

In later cycles:

- Add fresh drop → crystals take noticeably longer to dissolve

- Bond lengths stay closer to crystalline values even when dissolved

- The distinction starts to blur

S-O bond length evolution during a multi-drop X-ray PDF experiment over approximately 700 minutes. Blue dots track S-O distance measurements, green diamonds mark crystal dissolution events, red circles indicate crystal appearance, and vertical gray lines show when new drops were added. The plot reveals a striking inversion phenomenon: initially, S-O bonds are shorter in the liquid phase (1.43-1.44 Å) than in crystalline phases (1.47 Å), with this relationship inverting after approximately 350 minutes. This evidence of structural reorganization during crystallization challenges conventional crystallization models.

I could see it in real-time. The X-ray diffraction peaks—those sharp spikes that signal crystallinity—were taking progressively more time to disappear after each drop addition.

The crystals forming after fifteen dissolution/recrystallization cycles weren’t just “more of the same.” They were structurally different. Reorganized. Matured.

And critically: when they dissolved, they weren’t returning to a truly random liquid state.

When Liquid and Crystal Blur

In every chemistry class, you learn about phase transitions. Water at 1°C is liquid. Water at -1°C is ice. There’s a discontinuous jump in properties—a sharp boundary between states of matter.

After twenty cycles, that boundary was disappearing in my data.

I was also running parallel experiments with Raman spectroscopy back in my lab—a technique that measures how molecules vibrate. The symmetric stretch vibration of the sulfate group splits into two distinct frequencies when you look closely. Call them ν₁ₕ and ν₁w, representing sulfate groups in slightly different environments.

Comparative analysis of the SO₄ symmetric stretch (ν₁) vibration for crystalline epsomite (left) and dissolved MgSO₄ (right). The spectra illustrate the decomposition of the ν₁ mode into two Gaussian components: a narrow, intense peak (ν₁h) and a broader, less intense peak (ν₁w). In crystalline epsomite (measured at 28.65 min), these components appear at 983 cm⁻¹ (ν₁h) and 981 cm⁻¹ (ν₁w), while in the dissolved state (measured at 49.65 min), they shift to 979 cm⁻¹ (ν₁h) and 983 cm⁻¹ (ν₁w). This splitting reflects differences in SO₄ coordination environments.

In the first few cycles, when crystals formed, these two peaks were separated by about 5 wavenumbers. A clear spectroscopic fingerprint: “this is crystalline, not liquid.”

By cycle twenty, that separation had shrunk to about 3.5 wavenumbers. The peaks were converging.

The liquid’s signature was becoming more crystal-like. The crystal’s signature was becoming more liquid-like. The distinction was blurring.

The PDF measurements told the same story: bond lengths in solution were approaching bond lengths in crystals. The structural difference between “dissolved” and “solid” was shrinking.

Prenucleation Clusters: The Solution That’s Not Quite Liquid

The textbook story of crystallization: dissolved ions bounce around randomly until, by chance, enough stick together to form a stable nucleus. Then the crystal grows.

This is Classical Nucleation Theory. We’ve been teaching it since the 1920s, and it works beautifully for some systems.

It's also increasingly clear it's incomplete for many real systems—including this one.

What my data shows looks more like this:

Even when “fully dissolved,” the ions aren’t random. Magnesium exists as Mg(H₂O)₆²⁺—an ion surrounded by exactly six water molecules in a perfectly octahedral arrangement. Sulfate groups form hydrogen bonds with those coordinated water molecules. These assemblies—call them prenucleation clusters—exist before any crystal appears.

They’re not crystals. But they’re not random soup either. They're something in between—and they matter.

As water evaporates and concentration increases, these clusters grow and start linking together. Eventually they develop into crystalline structures.

But here’s what conventional theory misses: the structure of those prenucleation clusters depends on the history of the solution.

A fresh solution that’s never crystallized has one cluster population. A solution made by dissolving crystals that have been through multiple cycles has a different cluster population.

By cycling twenty times, I was training the system.

Each cycle:

- Clusters form

- They organize into crystals

- Crystals dissolve

- But the clusters that reform remember the crystalline structure

The bond-length evolution I measured is the signature of this training. Early-cycle solutions have sulfate groups in their relaxed, unconstrained geometry (short S-O bonds). Late-cycle solutions have sulfate groups already pre-organized into the distorted configuration they’ll adopt in the crystal (longer S-O bonds).

The solution is ‘rehearsing’ being a crystal.

What This Actually Means

Here’s what keeps me up at night: this feels simultaneously mundane and impossible.

It’s mundane because of course matter changes under repeated stress. Blacksmiths have known for millennia that hammering metal makes it stronger—work hardening. Molecular biologists know that cycling proteins through heat and cooling helps them fold into stable configurations—annealing. Geologists know that clay develops preferred orientations from repeated wetting and drying.

But dissolved ions in water aren’t supposed to remember anything. Thermodynamics is clear: dissolving a crystal increases entropy. Order is destroyed. Information is erased. The system resets to maximum disorder.

That’s what the Second Law of Thermodynamics says should happen.

Except my solution didn’t reset.

Conceptual contrast between classical dissolution theory and observed behavior. Classical thermodynamics predicts that dissolving a crystal erases all structural information, returning the system to maximum entropy (left). Our measurements show that after repeated cycling, dissolved solutions retain partial organization—prenucleation clusters with preferred coordination geometries that serve as templates for recrystallization (right). The solution doesn’t reset to complete disorder; it maintains structural memory.

The disorder had developed structure. A preferred configuration. An organizational template that survived dissolution.

That’s not how liquids are supposed to work.

But it’s clearly how this liquid worked.

Why This Matters: Life in Impossible Places

Okay, so salt water develops structural memory when you cycle it enough times. Neat crystallography result. Why should anyone outside my niche field care?

Because life exists in these environments. And those environments cycle constantly.

Extremophile bacteria thrive in the most concentrated brines on Earth—places where the salt concentration would instantly kill most organisms. They survive in evaporite deposits: salt lakes that fill and dry seasonally, desert playas that see water only during rare rain events, ancient salt deposits with fluid inclusions where microbes have remained viable for thousands or even millions of years.

Extremophile habitats undergo repeated dissolution-crystallization cycles identical to laboratory experiments. Seasonal or episodic flooding of evaporite deposits creates brines that concentrate through evaporation, precipitating salt crystals that later dissolve when water returns. Halophilic bacteria (inset) must survive repeated exposure to cycling conditions where structural memory in the brine potentially creates predictable microenvironments. Natural systems may experience hundreds to thousands of wet-dry cycles over years to millennia.

These environments undergo exactly what I did in my experiment: repeated cycles of dissolution and crystallization. Wet season, dry season. Lake, playa, lake, playa. Over and over.

If structural memory is real—if solutions retain information from previous crystallization events—then these cycling environments aren’t chaotic. They’re predictable.

And predictable means evolvable.

The Evolutionary Advantage of a Solution That Remembers

Think about what an extremophile faces in an evaporating salt lake:

Without structural memory (classical picture):

• Each wet/dry cycle is random

• Crystallization is chaotic—you might get encapsulated in a crystal, you might not

• Osmotic stress is unpredictable—water activity changes erratically

• No way to anticipate or prepare for the next cycle

• Survival is mostly luck

With structural memory (what I observed):

• Each cycle reinforces structural organization

• Crystallization follows predictable pathways

• Specific microenvironments form reproducibly

• The solution “knows” where it’s been

• Organisms can evolve to exploit that predictability

If you’re a halophilic bacterium and the brine you live in has structural memory, you can:

• Sense the state of organization. If prenucleation clusters have characteristic structures, bacteria could evolve chemotaxis toward or away from them. Position yourself in favorable micro-environments before crystallization.

• Time your metabolic state. If bond-length transitions signal approaching crystallization, that’s information you can use. Enter dormancy at the right moment. Activate stress-response proteins when cluster populations shift.

• Exploit specific chemical niches. If solutions with memory develop heterogeneous structure—Mg-rich domains here, SO₄-rich domains there—that’s spatial organization. Different metabolic strategies could partition into different niches.

• Get encapsulated predictably. If you can sense where crystals will form and what their structure will be, you can position yourself to be captured in fluid inclusions—protective pockets within the crystal where you’re shielded from radiation and desiccation.

Over thousands of cycles—seasonal, annual, millennial—natural selection would strongly favor organisms that could read and respond to structural memory.

The environment isn’t just selecting for “can tolerate salt.” It’s selecting for “can predict and exploit the specific organizational patterns that emerge from repeated cycling.”

Life thrives on predictability. Structural memory provides that, even in extreme environments.

Mars, Obviously

Of course this has implications for Mars. Everything involving sulfate minerals has implications for Mars.

Magnesium sulfate deposits are everywhere on the Martian surface: Meridiani Planum, Gale Crater, Valles Marineris. Orbital spectroscopy and rover missions have been mapping them for two decades.

Distribution of magnesium sulfate deposits on Mars identified through orbital spectroscopy (CRISM, OMEGA) and rover investigations (MER, MSL). These sulfate-bearing layers record ancient aqueous activity in Noachian to Hesperian-age terrains. The formation mechanism—single evaporation events versus repeated wet-dry cycling—has direct implications for habitability: cycled deposits indicate persistent, predictable aqueous environments where structural memory could support microbial adaptation. Current sample return missions target these deposits for cycling history analysis. (Modified from Ehlmann, 2014 https://www.hou.usra.edu/meetings/8thmars2014/pdf/1245.pdf)

The question has always been: how did these deposits form?

Did they form from a single evaporation event—one ancient lake that dried up and never refilled?

Or did they form from repeated cycling—lakes that filled and evaporated many times over thousands or millions of years?

That question matters tremendously for habitability. A single evaporation event is a transient environment. Repeated cycling is a persistent, potentially stable environment where life could gain a foothold.

If structural memory exists and can be measured, we have a new tool.

Bond lengths in sulfate minerals could record cycling history. Crystals that formed in one evaporation event versus crystals that cycled twenty times should have measurably different S-O bond lengths, different Raman signatures, different structural states.

We could, in principle, look at a Martian rock sample and say: “This salt cycled approximately X times before final deposition.”

That tells us:

• How long liquid water persisted

• Whether the environment was stable or chaotic

• What kind of evolutionary pressures would have existed if life was present

• Where to look for biosignatures (preferentially in highly-cycled deposits where structural memory would make habitability more predictable)

And yes, we’re testing this. I have a student analyzing 300 feet of drill core from Searles Lake in California—an ancient evaporite deposit with layer after layer of salts formed under different climate conditions. If bond lengths vary systematically with depositional environment, that’s our proof of concept.

Then we apply it to Mars.

The Thing I Can’t Quite Name

But here's what keeps coming back to me, beyond the Mars implications, beyond the extremophile applications, there’s something here I still can’t articulate properly.

This discovery sits at the intersection of too many domains:

• Thermodynamics says it shouldn’t remember

• Kinetics says it could remember

• Information theory says memory requires encoding mechanisms

• Soft matter physics says structured liquids are common

• Quantum mechanics says coherence can persist in warm, wet environments

My experiment shows that simple inorganic ions in water exhibit behaviors we usually associate with:

• Polymers (shear memory)

• Proteins (folding memory)

• Glasses (structural relaxation)

• Quantum systems (coherence, collective states)

• Living systems (learning, adaptation)

And we don’t have adequate language for that outside biology.

Maybe that’s the most important implication. Not the specific mechanism—whether it’s quantum coherence or topological defects or water network memory—but the recognition that information can be stored in physical systems we thought were information-free.

A solution isn’t just a thermodynamic state. It’s a record of its history.

A crystal isn’t just an arrangement of atoms. It’s the endpoint of a pathway that could have gone differently.

The boundary between them isn’t a sharp line. It’s a gradient, a transition zone, a space where both states coexist and influence each other.

What Else Can Remember?

If salt water can remember, what else can?

What other “simple” physical systems are actually carrying information we haven’t learned to read?

What have we been missing because we assumed disorder meant no structure, and dissolution meant no memory?

I measured bonds. I counted cycles. I watched phase boundaries blur.

But I still don’t know what this is.

And that’s what makes it exciting. Sometimes the most important discoveries don't answer questions—they reveal that we've been asking the wrong ones all along.

---

The paper: “Molecular Mechanisms of Magnesium Sulfate Crystallization: Bond Length Inversion and the Role of Hydration in Mineral Formation” is now published in American Mineralogist. All the raw data, every PDF measurement, every Raman spectrum—it’s all there if you want to see exactly what I saw. https://doi.org/10.2138/am-2025-9913

What’s next: We’re analyzing Searles Lake drill core to see if natural evaporite deposits show the same bond-length signatures. If they do, we have a new tool for reading the environmental history written into every salt crystal—on Earth and potentially on Mars.

Methods note: The acoustic levitator design is open-source (TinyLev), the PDF analysis used the Advanced Photon Source at Argonne National Laboratory, and the Raman spectroscopy was done on a standard Horiba ExploRa+ system. If you want to replicate this or try it with other minerals, all the experimental parameters are in the paper.